CN116030958A - Auxiliary diagnosis method, device, equipment and system and storage medium - Google Patents
Auxiliary diagnosis method, device, equipment and system and storage medium Download PDFInfo
- Publication number
- CN116030958A CN116030958A CN202111245816.7A CN202111245816A CN116030958A CN 116030958 A CN116030958 A CN 116030958A CN 202111245816 A CN202111245816 A CN 202111245816A CN 116030958 A CN116030958 A CN 116030958A
- Authority
- CN
- China
- Prior art keywords
- biological sample
- detected
- image
- blood cell
- analysis data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000003745 diagnosis Methods 0.000 title claims abstract description 291
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000004458 analytical method Methods 0.000 claims abstract description 309
- 239000000523 sample Substances 0.000 claims abstract description 297
- 239000012472 biological sample Substances 0.000 claims abstract description 216
- 230000002159 abnormal effect Effects 0.000 claims abstract description 204
- 210000000601 blood cell Anatomy 0.000 claims description 348
- 238000010586 diagram Methods 0.000 claims description 134
- 238000001514 detection method Methods 0.000 claims description 105
- 238000012545 processing Methods 0.000 claims description 71
- 238000000605 extraction Methods 0.000 claims description 60
- 238000004422 calculation algorithm Methods 0.000 claims description 38
- 238000002405 diagnostic procedure Methods 0.000 claims description 34
- 230000005856 abnormality Effects 0.000 claims description 27
- 238000012360 testing method Methods 0.000 claims description 26
- 238000013136 deep learning model Methods 0.000 claims description 24
- 238000007689 inspection Methods 0.000 claims description 21
- 239000013598 vector Substances 0.000 claims description 17
- 230000011218 segmentation Effects 0.000 claims description 15
- 238000004590 computer program Methods 0.000 claims description 12
- 238000013507 mapping Methods 0.000 claims description 10
- 230000028993 immune response Effects 0.000 claims description 9
- 238000001228 spectrum Methods 0.000 claims description 9
- 238000004159 blood analysis Methods 0.000 claims description 8
- 230000035931 haemagglutination Effects 0.000 claims description 7
- 238000012742 biochemical analysis Methods 0.000 claims description 6
- 238000003018 immunoassay Methods 0.000 claims description 6
- 238000007635 classification algorithm Methods 0.000 claims description 4
- 210000004027 cell Anatomy 0.000 description 175
- 210000004369 blood Anatomy 0.000 description 137
- 239000008280 blood Substances 0.000 description 137
- 210000001772 blood platelet Anatomy 0.000 description 75
- 210000003743 erythrocyte Anatomy 0.000 description 69
- 238000003062 neural network model Methods 0.000 description 65
- 238000012549 training Methods 0.000 description 42
- 210000003714 granulocyte Anatomy 0.000 description 32
- 210000000265 leukocyte Anatomy 0.000 description 31
- 238000003709 image segmentation Methods 0.000 description 19
- 230000006870 function Effects 0.000 description 14
- 238000002372 labelling Methods 0.000 description 12
- 102000001554 Hemoglobins Human genes 0.000 description 11
- 108010054147 Hemoglobins Proteins 0.000 description 11
- 238000013528 artificial neural network Methods 0.000 description 11
- 230000008569 process Effects 0.000 description 10
- 210000004556 brain Anatomy 0.000 description 9
- 238000006243 chemical reaction Methods 0.000 description 7
- 238000005534 hematocrit Methods 0.000 description 6
- 230000009467 reduction Effects 0.000 description 6
- 238000004891 communication Methods 0.000 description 5
- 238000010276 construction Methods 0.000 description 5
- 238000013135 deep learning Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 4
- 210000000440 neutrophil Anatomy 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 230000015271 coagulation Effects 0.000 description 3
- 238000005345 coagulation Methods 0.000 description 3
- 238000011478 gradient descent method Methods 0.000 description 3
- 230000000877 morphologic effect Effects 0.000 description 3
- 150000007523 nucleic acids Chemical class 0.000 description 3
- 102000039446 nucleic acids Human genes 0.000 description 3
- 108020004707 nucleic acids Proteins 0.000 description 3
- 238000012216 screening Methods 0.000 description 3
- 239000000243 solution Substances 0.000 description 3
- 238000006467 substitution reaction Methods 0.000 description 3
- 230000003319 supportive effect Effects 0.000 description 3
- 108060003951 Immunoglobulin Proteins 0.000 description 2
- 210000003651 basophil Anatomy 0.000 description 2
- 230000015572 biosynthetic process Effects 0.000 description 2
- 230000001413 cellular effect Effects 0.000 description 2
- 201000010099 disease Diseases 0.000 description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 102000018358 immunoglobulin Human genes 0.000 description 2
- 229940072221 immunoglobulins Drugs 0.000 description 2
- 238000002347 injection Methods 0.000 description 2
- 239000007924 injection Substances 0.000 description 2
- 210000004698 lymphocyte Anatomy 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 210000003924 normoblast Anatomy 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 208000000649 small cell carcinoma Diseases 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 206010003445 Ascites Diseases 0.000 description 1
- 102000009123 Fibrin Human genes 0.000 description 1
- 108010073385 Fibrin Proteins 0.000 description 1
- BWGVNKXGVNDBDI-UHFFFAOYSA-N Fibrin monomer Chemical compound CNC(=O)CNC(=O)CN BWGVNKXGVNDBDI-UHFFFAOYSA-N 0.000 description 1
- 206010018910 Haemolysis Diseases 0.000 description 1
- 206010048612 Hydrothorax Diseases 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000000149 argon plasma sintering Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 230000023555 blood coagulation Effects 0.000 description 1
- 238000004820 blood count Methods 0.000 description 1
- 210000001124 body fluid Anatomy 0.000 description 1
- 239000010839 body fluid Substances 0.000 description 1
- 230000007177 brain activity Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 210000001175 cerebrospinal fluid Anatomy 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000004090 dissolution Methods 0.000 description 1
- 238000002091 elastography Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 210000003979 eosinophil Anatomy 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 229950003499 fibrin Drugs 0.000 description 1
- 238000000684 flow cytometry Methods 0.000 description 1
- 238000012757 fluorescence staining Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000008588 hemolysis Effects 0.000 description 1
- 230000001900 immune effect Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 210000004180 plasmocyte Anatomy 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 210000001995 reticulocyte Anatomy 0.000 description 1
- 238000000790 scattering method Methods 0.000 description 1
- 230000001568 sexual effect Effects 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 210000001179 synovial fluid Anatomy 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 210000004881 tumor cell Anatomy 0.000 description 1
- 210000002700 urine Anatomy 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Investigating Or Analysing Biological Materials (AREA)
Abstract
The application provides an auxiliary diagnosis method, an auxiliary diagnosis device, an auxiliary diagnosis system and a computer readable storage medium, wherein the method comprises the following steps: acquiring biological sample analysis data of an object to be detected, extracting characteristics of the biological sample analysis data, and outputting a recognition result of target characteristics carried in the biological sample analysis data; acquiring clinical information of the object to be detected; and determining abnormal characteristics of a sample to be detected of the object to be detected based on the biological sample analysis data, the identification result and the clinical information, and outputting auxiliary diagnosis decision information corresponding to the abnormal characteristics.
Description
Technical Field
The present disclosure relates to the field of medical devices, and in particular, to a method, an apparatus, a device, a system, and a computer storage medium for assisting diagnosis.
Background
Currently, the application of the blood cell analyzer is more and more widespread, the technical development is more and more mature, the blood sample of a patient is automatically detected by the blood cell analyzer, no matter the blood cell analyzer with different fluorescent channels at three classifications, five classifications or higher end has little change in the interface displayed to a user, and the blood cell analyzer consists of dozens of report parameters, a plurality of graphs and alarm information from different channels with different numbers, namely, after the blood sample of the patient is detected and analyzed by the blood cell analyzer, a numerical report consisting of a large amount of data and graphs is displayed to an examining physician.
The examining physician needs to make comprehensive decisions using these numeric reports to send the examining report to the clinician. However, the large number of blood samples that a testing physician needs to process each day, the extremely high effort does not guarantee that they have sufficient time to understand the test data for each patient, and is also very dependent on the individual testing level of the testing physician, thus resulting in a low testing efficiency.
Disclosure of Invention
In order to solve the existing technical problems, the application provides an auxiliary diagnosis method, device, equipment and system with higher inspection efficiency and more accurate detection result, and a computer storage medium.
In order to achieve the above purpose, the technical solution of the embodiments of the present application is implemented as follows:
in a first aspect, an embodiment of the present application provides an auxiliary diagnostic method, applied to an auxiliary diagnostic device, including:
acquiring biological sample analysis data of an object to be detected, extracting characteristics of the biological sample analysis data, and outputting a recognition result of target characteristics carried in the biological sample analysis data;
acquiring clinical information of the object to be detected;
and determining abnormal characteristics of a sample to be detected of the object to be detected based on the biological sample analysis data, the identification result and the clinical information, and outputting auxiliary diagnosis decision information corresponding to the abnormal characteristics.
In a second aspect, embodiments of the present application provide an auxiliary diagnostic apparatus, including:
the first acquisition module is used for acquiring biological sample analysis data of an object to be detected;
the processing module is used for extracting the characteristics of the biological sample analysis data and outputting the identification result of the target characteristics carried in the biological sample analysis data;
the second acquisition module is used for acquiring clinical information of the object to be detected;
and the auxiliary diagnosis decision module is used for determining abnormal characteristics of a sample to be detected of the object to be detected based on the biological sample analysis data, the identification result and the clinical information, and outputting auxiliary diagnosis decision information corresponding to the abnormal characteristics.
In a third aspect, an embodiment of the present application provides an image recognition device, including a processor and a memory, where the memory stores a computer program executable by the processor, and when the computer program is executed by the processor, the computer program implements an auxiliary diagnostic method according to any embodiment of the present application.
In a fourth aspect, an embodiment of the present application provides an auxiliary diagnostic system, which includes a biological sample analyzer, a test information system, and an auxiliary diagnostic device according to any one of the embodiments of the present application, where the auxiliary diagnostic device is configured to obtain biological sample analysis data of an object to be detected from the biological sample analyzer, and obtain clinical information of the object to be detected from the test information system.
In a fifth aspect, embodiments of the present application provide a computer readable storage medium having a computer program stored thereon, which when executed by a controller, provides a method of assisting in diagnosis according to any of the embodiments of the present application.
According to the auxiliary diagnosis method, the device, the equipment and the system and the computer readable storage medium, according to the biological sample analysis data, the identification result of the target feature in the biological sample analysis data and the clinical information of the object to be detected, the abnormal feature of the sample to be detected of the object to be detected is determined, and the auxiliary diagnosis decision information corresponding to the abnormal feature is output, so that the abnormal feature contained in the biological sample analysis data is used as the identification object, the biological sample analysis data and the clinical information of the object to be detected are combined to determine the abnormal feature of the sample to be detected, the auxiliary diagnosis decision information corresponding to the abnormal feature is output based on the abnormal feature, and an accurate auxiliary diagnosis decision can be obtained without depending on the personal experience level of a checking doctor, so that the checking efficiency can be improved, the detection result is more accurate, and the diagnosis decision is more interpretable.
Drawings
FIG. 1 is a schematic diagram of an alternative application scenario of an auxiliary diagnostic method according to an embodiment;
FIG. 2 is a flow chart of an auxiliary diagnostic method in one embodiment;
FIG. 3 is a schematic diagram of a deep learning model according to an embodiment;
FIG. 4 is a flow chart of an auxiliary diagnostic method in another embodiment;
FIG. 5 is a schematic diagram of an auxiliary diagnostic method according to an embodiment;
FIG. 6 is a schematic diagram of an auxiliary diagnostic method in a specific example;
FIG. 7 is a schematic diagram of an auxiliary diagnostic method in another embodiment;
FIG. 8 is a flow chart of an auxiliary diagnostic method applied to an image recognition model in one embodiment;
FIG. 9 is a flowchart of an auxiliary diagnostic method applied to an image recognition model in another embodiment;
FIG. 10 is a schematic diagram of iterative training of a first sub-image model in an image recognition model in an embodiment;
FIG. 11 is a schematic diagram of iterative training of a second sub-image model in an image recognition model according to another embodiment;
FIG. 12 is a flowchart of a method for constructing an auxiliary diagnostic knowledge graph in an embodiment;
FIG. 13 is a schematic diagram of a method for constructing an auxiliary diagnosis knowledge graph according to an embodiment;
FIG. 14 is a schematic diagram of a method for constructing an auxiliary diagnostic knowledge graph in accordance with another embodiment;
FIG. 15 is a schematic diagram of an auxiliary diagnostic device according to an embodiment;
fig. 16 is a schematic structural view of an auxiliary diagnostic device in an embodiment.
Detailed Description
The technical scheme of the application is further elaborated below by referring to the drawings in the specification and the specific embodiments.
For the purpose of making the objects, technical solutions and advantages of the present application more apparent, the present application will be described in further detail with reference to the accompanying drawings, and the described embodiments should not be construed as limiting the present application, and all other embodiments obtained by those skilled in the art without making any inventive effort are within the scope of the present application.
In the following description, reference is made to the expression "some embodiments" which describe a subset of all possible embodiments, it being noted that "some embodiments" may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.
In the following description, the terms "first, second, third" and the like are used merely to distinguish between similar objects and do not represent a specific ordering of the objects, it being understood that the "first, second, third" may be interchanged with a specific order or sequence, as permitted, to enable embodiments of the application described herein to be practiced otherwise than as illustrated or described herein.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of the present application only and is not intended to be limiting of the present application.
Referring to fig. 1, a frame diagram of an auxiliary diagnostic system of an optional application scenario of an auxiliary diagnostic method provided in an embodiment of the present application is shown. The auxiliary diagnostic system includes a test information system (Laboratory Information System, LIS) 10, a biological sample analyzer 20, and an auxiliary diagnostic device 30. The test information system 10 is typically provided at a clinical laboratory of a hospital and may be used to receive test data, enter patient test information for storage, and assist the hospital in information management. The biological sample analyzer 20 refers to a device for performing intelligent detection analysis on a collected sample to be measured. The sample to be tested can be a sample containing various biological cell information or other biological information, such as a blood sample, a urine sample, other body fluid (hydrothorax and ascites, cerebrospinal fluid, serosal cavity effusion, synovial fluid) sample and the like, and the biological cell type can be at least one of neutrophils, lymphocytes, monoglobes, eosinophils and basophils; but also immature granulocytes, tumor cells, lymphoblasts, plasma cells, atypical lymphocytes, pre-erythroblasts, basal erythrocytes and polychromatic erythrocytes, orthochromatic erythrocytes, pre-megablasts, basal megablasts, polychromatic megacytes and nucleated erythrocytes selected from the group consisting of orthochromatic megaerythrocytes and megakaryospheres.
The auxiliary diagnosis device 30 is in communication connection with the test information system 10 and the biological sample analyzer 20, and is configured to obtain clinical information of a patient from the test information system 10 and biological sample analysis data of a sample to be tested of the patient from the biological sample analyzer 20, determine abnormal characteristics of the sample to be tested of the patient in combination with the clinical information of the patient, the biological sample analysis data and a case database formed according to a history diagnosis record, and output auxiliary diagnosis decision information corresponding to the abnormal characteristics.
Wherein, the auxiliary diagnostic device 30 may be various intelligent devices with storage and calculation capabilities, which can be physically separated from the inspection information system or the biological sample analyzer, such as a smart phone, a personal computer, etc. loaded with a computer program for implementing the auxiliary diagnostic method of the embodiment of the present application; it may also be an integrated test information system or biological sample cell analyzer, such as one loaded with a computer program implementing the auxiliary diagnostic methods of embodiments of the present application.
For ease of understanding, in the embodiments of the present application, the implementation of the auxiliary diagnostic method will be described in detail by taking a sample to be measured as a blood sample, and the biological sample analyzer refers to a blood cell analyzer. The known blood cell analyzer is a typical application of flow cytometry, the impedance channel for cell counting uses the coulter electrical impedance principle, the colorimetric principle is used for hemoglobin concentration measurement, and the optical channel of the high-end blood cell analyzer for white blood cell classification, reticulocyte identification, nucleated red blood cell identification, basophil identification, low-value platelet identification, primitive cell identification and the like uses a laser scattering method and a nucleic acid fluorescence staining technology, and the aim of the above technologies is to convert biological characteristics of blood cell size, cell content complexity, nucleic acid content and the like into electric pulse signals which can be primitive data collected by the blood cell analyzer. However, it should be noted that, although the description of the embodiments of the present application describes the biological sample analysis data as an example of the detection analysis data of the blood sample, the protection scope of the present application should not be limited in this way.
Referring to fig. 2, an auxiliary diagnostic method provided in an embodiment of the present application may be applied to an auxiliary diagnostic apparatus, where the auxiliary diagnostic method includes, but is not limited to, S101, S103, and S105, and is specifically described as follows:
s101, acquiring biological sample analysis data of an object to be detected, extracting features of the biological sample analysis data, and outputting a recognition result of target features carried in the biological sample analysis data.
The object to be detected refers to the subject of the sample to be detected, and is typically a patient who provides a blood sample, taking the sample to be detected as a blood sample of a human body as an example. The biological sample analysis data refers to corresponding analysis data obtained by detecting various biological samples containing biological cell information or other biological information, and the cell characteristics can be health characteristics or unhealthy characteristics represented by different types of cells in corresponding biological samples, such as abnormal cell types, cell number characteristics, cell size characteristics, cell composition proportion characteristics, cell content characteristics, nucleic acid content characteristics and the like in the biological samples. The target features refer to biological information features determined according to detection requirements of different biological samples, such as taking biological sample analysis data as cell analysis data, the target features refer to one or more cell features determined according to detection requirements of different biological samples, such as PLT histograms in the cell analysis data of blood samples, and the target cell features refer to quantity distribution features of platelets with different sizes.
S103, acquiring clinical information of the object to be detected.
The clinical information of the subject to be tested includes the age, sex, type of test sample, past history of the subject to be tested, and the like.
S105, determining abnormal characteristics of a sample to be detected of the object to be detected based on the biological sample analysis data, the identification result and the clinical information, and outputting auxiliary diagnosis decision information corresponding to the abnormal characteristics.
The abnormal characteristics of the sample to be detected refer to characteristics of the sample, which can be used for representing whether the object to be detected has a disease hidden trouble, and taking a blood sample as an example, the abnormal characteristics of the blood sample refer to characteristics of the blood sample, which can be used for representing whether the object to be detected has a disease hidden trouble. The auxiliary diagnosis decision information corresponding to the abnormal characteristics refers to the combination of the extraction result of the target characteristics in the biological sample analysis data, the clinical information of the object to be detected and other report parameters except the extraction result of the cellular characteristics in the cellular analysis data, and the formed auxiliary diagnosis conclusion determined based on the sample to be detected of the object to be detected can comprise the basis description of the determined abnormal characteristics, which can be words or marked images, and the auxiliary diagnosis decision information can reflect the information on which the decision is made, so that the basis of the abnormality determined in the current detection result can be intuitively known, and the readability of the auxiliary diagnosis result is improved. The auxiliary diagnosis equipment combines the cell analysis data, the recognition result of the target cell characteristics carried by the cell analysis image and the clinical data to determine the abnormal characteristics of the sample to be detected of the object to be detected, and outputs auxiliary diagnosis decision information corresponding to the abnormal characteristics, so that the sample with the abnormality in the detection result of the current sample to be detected can be rapidly screened out through the auxiliary diagnosis decision information, and the basis of the abnormality is determined, thereby facilitating a checking doctor to make a diagnosis and treatment decision in the next step more efficiently according to the auxiliary diagnosis decision information.
In the above embodiment, the auxiliary diagnosis method determines the abnormal characteristics of the sample to be detected of the object to be detected according to the biological sample analysis data, the identification result of the target characteristics in the biological sample analysis data and the clinical information of the object to be detected, and outputs the auxiliary diagnosis decision information corresponding to the abnormal characteristics, so that the abnormal characteristics contained in the biological sample analysis data can be used as the identification object, the abnormal characteristics of the sample to be detected can be determined by combining the biological sample analysis data and the clinical information of the object to be detected, the auxiliary diagnosis decision information corresponding to the abnormal characteristics can be output based on the abnormal characteristics, and an accurate auxiliary diagnosis decision can be obtained without depending on the personal experience level of a testing doctor, thereby improving the testing efficiency, ensuring more accurate detection results and more interpretability of the diagnosis decision.
In an alternative embodiment, the sample to be tested is a blood sample of a patient, and the biological sample analysis data is blood cell analysis data. S101, acquiring biological sample analysis data of an object to be detected, extracting features of the biological sample analysis data, and outputting a recognition result of target features carried in the biological sample analysis data, wherein the method comprises the following steps:
And acquiring biological sample analysis data of an object to be detected, extracting characteristics of a biological sample analysis image in the biological sample analysis data through an image processing model, and outputting a recognition result of a target object carried by the biological sample analysis image.
The biological sample analysis data is exemplified by blood cell analysis data, and the biological sample analysis image is referred to as a blood cell analysis image. The blood cell analysis data comprise various detection data obtained by detecting a blood sample, such as blood cell analysis images, blood report parameters, flag alarm parameters and the like. In an alternative specific example, the blood reporting parameters include Red Blood Cells (RBC), hemoglobin (HGB), hematocrit (HCT), mean red blood cell volume (MCV), mean red blood cell hemoglobin content (MCH), white Blood Cells (WBC), mean red blood cell hemoglobin concentration (MCHC), red blood cell distribution width (RDW-CV), red blood cell distribution width standard deviation (RDW-SD), platelets (PLT). The blood cell analysis image comprises histograms corresponding to different cells in blood and scatter diagrams representing the distribution characteristics of various cells, such as platelet histogram, red blood cell histogram, white blood cell DIFF scatter diagram and RET scatter diagram. The characteristics of different cells in the blood are correspondingly carried by different blood cell analysis images, so that whether the content and the distribution of corresponding cells in the blood are abnormal or not can be characterized, and the target object carried in the blood cell analysis images refers to the characteristics of the cells correspondingly carried by the blood cell analysis images. Taking the platelet histogram as an example, the object carried by the platelet histogram is the abnormal type of the curve distribution form in the blood cell histogram, and the abnormal type can be left shift of the histogram, right shift of the histogram, tail drag of the right side of the histogram, and the like. The image processing model performs feature extraction on the blood cell analysis image so as to identify a target object carried by the blood cell analysis image and output the target object. The image processing model can take the abnormal cell characteristics carried in the blood cell analysis image as a target object, and can determine a support basis for the abnormality existing in the blood cell analysis image by extracting the characteristics of the blood cell analysis image, identifying and outputting the abnormal cell characteristics of the corresponding blood cell analysis image as an identification result.
In some embodiments, the obtaining biological sample analysis data of the object to be detected, extracting features of a biological sample analysis image in the biological sample analysis data by using an image processing model, and outputting a recognition result of a target object carried by the biological sample analysis image includes:
acquiring biological sample analysis data of an object to be detected output by a biological sample analyzer;
extracting features of biological sample analysis images in the biological sample analysis data through an image processing model, and outputting recognition results of target features respectively carried by the biological sample analysis images; the biological sample analysis image includes at least one of: a blood cell histogram, a blood cell scatter plot, a pulse signal waveform plot, a pulse signal signature plot, a thromboelastography, a biochemical or immune response curve.
The biological sample analyzer takes a blood cell analyzer as an example, and the auxiliary diagnosis equipment is in communication connection with the blood cell analyzer and directly takes blood cell analysis data output by the blood cell analyzer as input. The blood cell analysis data comprises blood cell analysis images carrying different blood cell characteristics, the blood cell analysis images can comprise histograms corresponding to different blood cells in blood, scatter diagrams representing distribution characteristics of various types of cells, pulse signal waveform diagrams and pulse signal characteristic diagrams, wherein the blood cell histograms and the blood cell scatter diagrams can be as follows: platelet histogram, red cell histogram, white cell DIFF scatter plot, RET scatter plot, histogram formed by pulse characteristic values such as peak height, front peak width, back peak width, half peak width of pulse signal (such as RBC, PLT and WBC histogram formed by peak height of pulse signal), scatter plot formed by two-dimensional, three-dimensional or even multidimensional formed by characteristic values of pulse signal in several dimensions (such as DIFF three-dimensional scatter plot data set formed by peak height of SFL, SSC and FSC three-channel pulse signal), two-dimensional scatter plot projection plot formed by three-dimensional and high-dimensional scatter plot data; the pulse signal waveform diagram may be as follows: the blood cell analyzer detects a waveform chart formed by the collected original pulse signals of the blood sample and a pulse signal waveform chart formed by the pulse signals screened by the pulse recognition algorithm; the pulse signal feature map may be as follows: and a vector characteristic diagram formed by pulse characteristic values such as peak height, front peak width, rear peak width, half peak width and the like of the pulse signals. The characteristics of different cells in the blood are correspondingly carried by different blood cell analysis images, so that whether the content and distribution of the corresponding cells in the blood are abnormal or not can be characterized, and the characteristics of the target cells carried in the blood cell analysis images refer to the characteristics of different types of cells correspondingly carried by the blood cell analysis images. The biological sample analyzer may also include a coagulation analyzer, an immunoassay analyzer, etc., and the biological sample image data may also include a thromboelastography, a biochemical or an immune response curve, respectively. In this embodiment, the image processing model takes a blood cell analysis image as input, performs feature extraction on the blood cell analysis image, and outputs recognition results of target cell features carried by the blood cell analysis image respectively.
In the above embodiment, the auxiliary diagnostic apparatus performs analysis by acquiring biological sample analysis data of the biological sample analyzer by cooperating with the biological sample analyzer, and can accurately determine whether there is an abnormality in the sample to be detected of the object to be detected to provide an inspection physician with auxiliary diagnosis.
In another optional embodiment, the step S101 of obtaining biological sample analysis data of the object to be detected, extracting features from the biological sample analysis data, and outputting a recognition result of the target features carried in the biological sample analysis data includes:
acquiring biological sample analysis data of an object to be detected; the biological sample analysis data includes at least one of: a blood cell histogram, a blood cell scatter diagram, a pulse signal waveform diagram, a pulse signal feature diagram, a one-dimensional pulse signal vector, a pulse signal feature value vector, a multi-dimensional scatter data set, elasticity diagram data and immune response original data;
and extracting the characteristics of the biological sample analysis data through a deep learning model or a traditional algorithm model, and outputting the identification result of the target characteristics carried in the biological sample analysis data.
Taking the sample to be detected as a blood sample of a patient as an example, the biological sample analysis data refers to blood cell analysis data. The blood cell analysis data can comprise a blood cell histogram, a blood cell scatter diagram, a pulse signal waveform diagram, a pulse signal characteristic diagram, a one-dimensional pulse signal vector, a pulse signal characteristic value vector and a multi-dimensional scatter data set, and based on the characteristics of different types of blood cell analysis data, an algorithm model is established to extract and classify the blood cell characteristics in the different types of blood cell analysis data according to the known analysis algorithm of the different types of blood cell analysis data, and the identification result of the target blood cell characteristics is output; or collecting a training sample set to train the deep learning model, extracting and classifying the characteristics of the blood cells in the blood cell analysis data of different types through the trained deep learning model, and outputting the identification result of the characteristics of the target blood cells. The biological sample analysis data may also be sample data containing other biological information, and the target feature is information representing characteristics of corresponding biological information, such as information representing whether a CRP reaction curve output by an immune analyzer has an abnormal peak, whether an impedance detection channel of a blood cell analyzer has a hole blockage, whether a DIFF detection channel of the blood cell analyzer finds immature granulocyte, and the like.
Optionally, the feature extraction is performed on the biological sample analysis data by a deep learning model or a traditional algorithm model, and the output of the identification result of the target feature carried in the biological sample analysis data includes:
extracting features of the biological sample analysis data through a deep learning model based on image classification; or carrying out feature extraction on blood cell features in the blood cell analysis data based on a traditional algorithm model constructed by an image morphology algorithm, an image classification algorithm, a clustering algorithm or a threshold segmentation algorithm;
classifying according to the feature extraction result, and outputting the identification result of the target feature carried in the biological sample analysis data.
Referring to fig. 3, an optional architecture diagram of a deep learning model based on image classification is shown, where the deep learning model includes an image processing model (acceptance v4+average reduction+concat+softmax) for extracting features from image input data such as a biological sample analysis image, and a feature conversion model (inputmatata+ Feature transform +concat+softmax) for extracting features from numerical input data in the biological sample analysis data, and the image processing model and the feature conversion model share a classification output layer (concat+softmax) for splicing and classifying features, and outputting a recognition result of a target blood cell feature carried in the biological sample analysis data. The image morphology algorithm can comprise a combination of algorithms such as expansion corrosion and the like, and the identification of target features of biological sample analysis data of the image is realized through processing steps such as boundary extraction, skeleton extraction, hole filling, corner extraction, image reconstruction and the like. For example, the image morphology algorithm, the image classification algorithm, the clustering algorithm or the threshold segmentation algorithm can be implemented by adopting a traditional algorithm for processing the blood cell analysis data, for example: for high-dimensional non-image data, the traditional algorithm can adopt strategies such as PCA dimension reduction and the like to extract features, and then uses a distance function to perform condition discrimination, or a Support Vector Machine (SVM) to perform automatic discrimination, or KNN (k-nearest neighbor rule) to perform sample classification. It should be noted that, the present embodiment is intended to protect the identification of the target features of the analysis data of the biological sample based on the establishment of the algorithm model, so as to provide the basis for the auxiliary diagnostic decision information, rather than making an innovation on the adopted algorithm itself.
In the above embodiment, the deep learning model or the conventional algorithm model may support the identification of the target features carried by the biological sample analysis data in more types, and the identification result of the target features is provided as the basis for making the auxiliary diagnosis decision information.
In some embodiments, the acquiring clinical information of the object to be detected includes:
acquiring clinical information of the object to be detected, which is output by an inspection information system; the clinical information comprises the age, sex, sample type to be tested and historical diagnosis record of the object to be tested.
The auxiliary diagnostic equipment is in communication connection with the inspection information system, and clinical information of the object to be detected is directly obtained from the inspection information system as input. The age, sex, type of sample to be detected and historical diagnosis record of the object to be detected are different, and the corresponding sample standard values are also different correspondingly, for example, the age, sex, type of sample to be detected and historical diagnosis record of the object to be detected are different, and the corresponding normal morphological standards of different cells in the blood sample are also different, so that the auxiliary diagnosis equipment is combined with the clinical information of the object to be detected to judge whether the blood sample of the object to be detected has abnormal characteristics, and the accuracy of the auxiliary diagnosis result is improved. The type of the sample to be tested refers to the type of executing different detection items on the blood sample, such as blood routine detection, whole blood detection and the like. The historical diagnostic record includes a diagnostic record of the medical history of the subject to be tested, including a past medical history.
In the above embodiment, the auxiliary diagnostic device, through cooperation with the inspection information system, can accurately determine whether there is an abnormality in the sample to be detected of the object to be detected by acquiring clinical information of the object to be detected in the inspection information system and analyzing the clinical information in combination with biological sample analysis data output by the biological sample analyzer, so as to provide auxiliary diagnosis for an inspection doctor.
Optionally, the acquiring clinical information of the object to be detected further includes:
obtaining detection result data of the object to be detected, which is obtained through a single instrument, joint inspection and/or a blood analysis pipeline; the detection result data comprises at least one of the following: blood cell analysis results, biochemical analysis results, immunoassay results, hemagglutination results, and blood cell microscopic examination results.
The blood cell analyzer may be a single instrument, a single joint inspection instrument (for example, a joint inspection instrument of blood routine and CRP, or a joint inspection product of blood routine and CRP and SAA), or a cascade blood cell analyzer comprising a common sample injection mechanism or a cascade of multiple detection devices with a splice sample injection mechanism, where the auxiliary diagnostic device may obtain detection result data of a sample to be detected from the single instrument, or obtain detection result data of the sample to be detected obtained by joint inspection from the cascade blood cell analyzer. The auxiliary diagnostic equipment can be in communication connection with the blood analysis pipeline to acquire detection result data of a sample to be detected, which is obtained by the blood analysis pipeline.
In the above embodiment, the auxiliary diagnostic device may cooperate with a single instrument, a cascade blood cell analyzer, and a blood analysis pipeline to obtain detection result data such as a blood cell analysis result, a biochemical analysis result, an immunoassay result, a hemagglutination result, and a blood cell microscopic examination result of the object to be detected, which are provided as a basis for making auxiliary diagnostic decision information.
In some embodiments, the determining abnormal characteristics of the sample to be detected of the object to be detected based on the biological sample analysis data, the identification result, and the clinical information, and outputting auxiliary diagnostic decision information corresponding to the abnormal characteristics, includes:
inputting an auxiliary diagnosis knowledge graph based on the report parameters, the recognition result and the clinical information in the biological sample analysis data;
and the auxiliary diagnosis knowledge graph determines abnormal characteristics of the biological sample of the object to be detected according to the report parameters, the identification result and the clinical information, and outputs auxiliary diagnosis decision information corresponding to the abnormal characteristics.
The biological sample is a blood sample, the analysis data of the biological sample is taken as an example of analysis data obtained from a blood analyzer, and various detection data obtained by detecting the blood sample by the blood analyzer comprise blood cell analysis images, blood report parameters, flag alarm parameters and the like. In an optional example, the image processing device performs feature extraction on the blood cell analysis image, identifies a target object carried by the blood cell analysis image, where the target object carried by the blood cell analysis image refers to a blood cell feature carried correspondingly in the blood cell analysis image and used for determining whether an abnormal blood cell exists, and the identification result refers to an identification result of the image processing model on whether the abnormal blood cell analysis image exists and a basis for determining whether the corresponding identification result exists in the image. In an alternative specific example, the blood reporting parameters include Red Blood Cells (RBC), hemoglobin (HGB), hematocrit (HCT), mean red blood cell volume (MCV), mean red blood cell hemoglobin content (MCH), mean red blood cell hemoglobin concentration (MCHC), white Blood Cells (WBC), red blood cell distribution width (RDW-CV), red blood cell distribution width standard deviation (RDW-SD), platelets (PLT). The blood cell analysis image comprises histograms corresponding to different cells in blood and scatter diagrams representing distribution characteristics of various blood cells, such as platelet histogram, red blood cell histogram, white blood cell DIFF scatter diagram and RET scatter diagram. The auxiliary diagnosis decision information corresponding to the abnormal features can comprise descriptions corresponding to the abnormal features, and represent the diagnosis basis for decision making. The auxiliary diagnosis knowledge graph is used for determining abnormal characteristics of the blood sample by combining blood report parameters in the blood cell analysis data, recognition results of target objects carried in the blood cell analysis image and the clinical information, forming auxiliary diagnosis decision information corresponding to the abnormal characteristics, and providing a determination basis of the abnormal characteristics of the blood sample through the auxiliary diagnosis decision information for explanation.
In the above embodiment, the auxiliary diagnosis knowledge graph can convert the detection result of the biological sample into the auxiliary diagnosis decision information with stronger readability, so that the abnormal sample in the detection result of the current sample can be conveniently and rapidly screened out and the basis of the abnormality can be determined by the auxiliary diagnosis decision information, and the examining physician can conveniently and efficiently make the next diagnosis and treatment decision according to the auxiliary diagnosis decision information.
In some embodiments, before the inputting of the auxiliary diagnostic knowledge-graph based on the report parameters, the identification result, and the clinical information in the biological sample analysis data, further comprises:
forming an entity set in a knowledge base according to the identification result of the target feature carried by the biological sample analysis data, the clinical information of the detection object and the sample reference parameters;
forming an attribute set and a relationship set in a knowledge base according to the identification result, the clinical information, the comparison result of the report parameter relative to the sample reference parameter and the mapping relation of the historical diagnosis example in the diagnosis example database;
forming a knowledge spectrum triplet set according to the entity set, the attribute set and the relation set, and constructing an auxiliary diagnosis knowledge spectrum according to the knowledge spectrum triplet set.
The recognition result of the target feature carried by the biological sample analysis data is taken as an example of the recognition result of the target object carried by the blood cell analysis image through an image processing model. The sample reference parameters refer to reference values corresponding to detection index parameters respectively contained in different sample types to be detected. The sample reference parameters of different sample types to be detected comprise different numbers and types of detection index parameters, and the value ranges of the reference values of the detection index parameters in the sample reference parameters of the same sample types to be detected are different. The historical diagnosis example refers to a detection result of determining whether the sample to be detected is abnormal based on the biological sample analysis image, clinical information and sample reference parameters of the object to be detected, for example, any blood detection record of the user can form a historical diagnosis example, for example, according to whole blood detection performed by the user a at time 1, the whole blood detection record a1 correspondingly comprises blood cell analysis data which is output by a blood detector for the blood sample of the user a at time 1, clinical information which is correspondingly input by the user a during the blood sample detection at time 1 in a detection information system, and comprehensive judgment made by a detecting doctor by using the data, and the whole blood detection record a1 can be used as a diagnosis example of a diagnosis example database.
The blood cell analysis image output by the blood analyzer in the whole blood detection record a1 can be used as a training sample for training the initial neural network model after being marked, and a trained image processing model is obtained. And taking a blood cell analysis image output by the blood analyzer in the whole blood detection record a1 as input of an image processing model, extracting characteristics of the blood cell analysis image through the image processing model, outputting a recognition result of a target object carried by the blood cell analysis image, and determining abnormal characteristics of a blood sample of the object to be detected by using the recognition result of the image processing model on the target object carried by the blood cell analysis image, clinical information of the detection object and blood reference parameters as input of an auxiliary diagnosis knowledge map, and outputting an auxiliary diagnosis report containing descriptions corresponding to the abnormal characteristics. The auxiliary diagnosis knowledge graph takes triples as nodes for constructing the knowledge graph, each triplet can be characterized in the forms of (entity, relation, entity), (entity, attribute value) and the like, an entity set is formed by extracting the entity from input data of the auxiliary diagnosis knowledge graph, an attribute set and a relation set are respectively formed according to the auxiliary diagnosis knowledge graph based on the extracted attribute and relation in the mapping relation between the input and diagnosis examples in a diagnosis example database which is relied in the determining and outputting process, and a triplet set is formed by establishing triples according to the relation between the entity and the attribute value of the entity, namely, the node set for constructing the auxiliary diagnosis knowledge graph is formed.
In an optional specific example, the recognition result of the image processing model on the target object carried by the blood cell analysis image includes that the PLT histogram label 1 is that there is little red blood cell interference, and the RBC histogram label 0 is that the morphology is normal; the blood report parameters comprise corresponding parameter values of RBC, HGB, HCT, WBC, MCV, MCH, RDW-CV and RDW-SD; clinical information includes patient age, sex, and sample type. The diagnosis example in the diagnosis example database comprises the steps of determining that the platelet false is higher based on the presence of the small red cell interference and the small red cell in the PLT histogram, and determining that the platelet false is higher based on the higher PLT parameter value in the comparison result of the patient blood report parameter relative to the blood reference parameter; wherein, the presence of small-cell erythrocytes is determined based on the RBC histogram morphology normal and the parameter values of MCV, MCHC in the patient blood report parameters. Thus, the first node of the auxiliary diagnostic knowledge graph may include (PLT histogram, tag 1, small red blood cell interference), (RBC histogram, tag 0, morphologically normal), (blood reporting parameters, values corresponding to each parameter), (clinical information, basic information, age, sex, and sample type); the intermediate nodes may include (RBC histogram morphology is normal, MCV, MCH parameter values are low, small cell red blood cells in the comparison result of blood reporting parameters relative to blood reference parameters), (patient clinical information, parameter normal range, corresponding blood reference parameters), and the end nodes may include (PLT histogram has small cell interference, small cell red blood cells, platelet false high), (patient blood reporting parameters, PLT parameter values are high, platelet false high in the comparison result of blood reporting parameters relative to the blood reference parameters), and the auxiliary diagnostic knowledge map is constructed according to the mapping relation relied on from the first node to the determined end node.
In the above embodiment, by constructing the auxiliary diagnosis knowledge graph, combining the image processing model with the identification result of the biological sample analysis image of the object to be detected, the clinical information and the corresponding sample report parameters to determine the abnormal characteristics of the biological sample, the examining physician can be assisted to rapidly screen out the sample with the abnormality in the current sample detection result and determine the basis of the abnormality, so that the examining physician can make the next diagnosis and treatment decision more efficiently according to the auxiliary diagnosis decision information.
In some embodiments, the determining the abnormal feature of the sample to be detected of the object to be detected, and outputting the auxiliary diagnostic decision information corresponding to the abnormal feature include one of the following:
determining abnormal characteristics of a sample to be detected of the object to be detected, correcting a parameter alarm state formed based on the identification result of the target cell characteristics according to the abnormal characteristics, and outputting corrected auxiliary diagnosis decision information corresponding to the abnormal characteristics;
determining abnormal characteristics of a sample to be detected of the object to be detected, and outputting auxiliary diagnosis decision information containing failure cause analysis corresponding to the abnormal characteristics;
Determining abnormal characteristics of a sample to be detected of the object to be detected, determining supporting evidence and/or diagnosis and treatment advice corresponding to the abnormal characteristics, and outputting an auxiliary diagnosis report containing the supporting evidence and/or diagnosis and treatment advice.
The biological sample analyzer performs detection analysis on the sample to be detected, outputs biological sample image data, parameter reports, alarm parameters and the like, for example, the cell analyzer performs detection analysis on the sample to be detected, and outputs detection analysis results of cell analysis data, blood parameter reports, alarm parameters and the like. The auxiliary diagnosis equipment can comprehensively take a detection analysis result output by the biological sample analyzer, clinical information of an object to be detected in the detection information system and the like as input for auxiliary diagnosis solution based on an auxiliary diagnosis knowledge graph, and simulate a detection doctor to make an auxiliary diagnosis decision based on detection result data and output key information based on which the auxiliary diagnosis decision is made, so that dependence on individual experience level of the detection doctor can be reduced, workload of the detection doctor is reduced, and the auxiliary diagnosis decision is more readable, accurate and efficient. An alternative scheme of the auxiliary diagnosis decision information is to correct a parameter alarm state formed based on the identification result of the target feature according to the abnormal feature, output corrected auxiliary diagnosis decision information corresponding to the abnormal feature, and highlight the basis for correcting the parameter alarm state from key information contained in the auxiliary diagnosis decision information and used for making the auxiliary diagnosis decision; in this way, auxiliary diagnostic decisions are made by integrating multiple data sources to avoid one-sidedness of a single data source. Another alternative scheme of the auxiliary diagnosis decision information is to include unqualified reason analysis corresponding to the abnormal characteristics, and unqualified specimen decision logic can be highlighted from key information included in the auxiliary diagnosis decision information and used for making an auxiliary diagnosis decision; in this way, it is possible to distinguish between failure causes such as abnormal samples (chylomicronemia, hemolysis, red blood cell cold coagulation, etc.), abnormal biological sample analyzers (hole blocking, unstable voltage, etc.). Yet another alternative of the auxiliary diagnostic decision information is to determine supporting evidence and/or diagnosis and treatment advice corresponding to the abnormal feature, output an auxiliary diagnostic report comprising the supporting evidence and/or diagnosis and treatment advice, and form an auxiliary diagnostic report according to the key information on which the auxiliary diagnostic decision is made.
In the above embodiment, the auxiliary diagnostic device cooperates with the biological sample analyzer and the inspection information system to synthesize biological sample analysis data, the identification result and the clinical information to make auxiliary diagnostic decisions, and forms auxiliary diagnostic decision information in different forms based on key information on which the auxiliary diagnostic decisions are made, so that the auxiliary diagnostic result is more readable, and the reliability of the quick judgment result is facilitated.
In some embodiments, the determining the abnormal feature of the sample to be detected of the object to be detected, outputting auxiliary diagnostic decision information corresponding to the abnormal feature, includes:
determining abnormal characteristics of a sample to be detected of the object to be detected, determining diagnosis and treatment suggestions corresponding to the abnormal characteristics, and outputting an auxiliary diagnosis report containing text descriptions of the diagnosis and treatment suggestions; or alternatively, the first and second heat exchangers may be,
and determining abnormal characteristics of a sample to be detected of the object to be detected, determining supporting evidence corresponding to the abnormal characteristics, and outputting an auxiliary diagnosis report comprising text description, image data carrying labels and numerical description of the supporting evidence.
The sample to be detected is exemplified by a blood sample of the object to be detected, and the abnormal characteristics of the biological sample may correspond to the diagnosis results of each historical diagnosis example in the diagnosis example database, such as a small number of platelets, an increased number of white blood cells with immature granulocytes, and the like. The diagnosis and treatment advice for the small number of platelets may be advice for checking for autoantibodies, platelet-related immunoglobulins and the like for definitive diagnosis, and the diagnosis and treatment advice for the increased number of leukocytes with immature granulocytes may be advice for performing relevant examinations such as myelocytology examination if necessary. The auxiliary diagnosis report provides abnormal characteristics of the biological sample and text description of the diagnosis and treatment advice corresponding to the abnormal characteristics, so that the auxiliary diagnosis result is more readable. The supportive evidence corresponding to the abnormal feature means that the cause of the abnormal feature is determined, for example, the supportive evidence for the small number of platelets may include comparison of the platelet count value with a corresponding reference value, and the supportive evidence for the increased number of white blood cells with immature granulocytes may include comparison of the white blood cell count value with a corresponding reference value, comparison of the immature granulocyte ratio with a corresponding reference value, comparison of a patient scatter diagram marked with an abnormal site with a reference scatter diagram confirmed by microscopic examination. The auxiliary diagnosis report provides the abnormal characteristics of the blood sample and the corresponding text description of the supporting evidence, the image data carrying the labels and the numerical description thereof, thereby facilitating the checking doctor to know the specific reasons of the abnormality in the current detection result efficiently and accurately, grasping the basis of the conclusion forming process in the auxiliary diagnosis report and facilitating the checking doctor to make decisions according to the basis.
In the above embodiment, the determination basis of the abnormal characteristics of the sample in the detection result of the sample to be detected can be provided through the auxiliary diagnosis report, and the description of the auxiliary diagnosis and the conclusion formation can be described through the text description or the text and image combination description, and the explanation of the supporting evidence of the auxiliary diagnosis conclusion of the auxiliary diagnosis report can also be further provided, so that the auxiliary diagnosis result with higher readability can be output.
In some embodiments, the determining the abnormal characteristic of the sample to be detected of the object to be detected, outputting the auxiliary diagnostic report including the corresponding description of the abnormal characteristic, includes:
acquiring a selection instruction of a simplified diagnosis report or a detailed diagnosis report;
outputting an auxiliary diagnosis report containing the text description of the abnormal cause according to the selection instruction of the simplified diagnosis report, or outputting an auxiliary diagnosis report containing the text description of the supporting evidence, the image data carrying the label and the numerical description according to the selection instruction of the detailed diagnosis report.
The auxiliary diagnostic device can provide a selection button of an auxiliary diagnostic report type through an application program interface, and a user can select the type for acquiring the auxiliary diagnostic report to be a simplified diagnostic report or a detailed diagnostic report by clicking the selection button. The auxiliary diagnosis equipment determines abnormal characteristics of a blood sample of the object to be detected according to a selection instruction of a user for the type of the auxiliary diagnosis report, if the user selects to output a simplified diagnosis report, determines diagnosis and treatment suggestions corresponding to the abnormal characteristics, and outputs the auxiliary diagnosis report containing word descriptions of the diagnosis and treatment suggestions; if the user selects to output the detailed diagnosis report, determining the abnormal characteristics of the blood sample of the object to be detected, determining the supporting evidence corresponding to the abnormal characteristics, and outputting an auxiliary diagnosis report containing the text description of the supporting evidence, the image data carrying the labels and the numerical description.
In the above embodiment, the user may autonomously select the type of the currently received auxiliary diagnostic report, obtain the simplified diagnostic report, quickly browse the current auxiliary diagnostic conclusion, and select to obtain the detailed diagnostic report for the specific basis of knowing the corresponding auxiliary diagnostic conclusion so as to meet the more sexual use demands.
In some embodiments, the determining abnormal characteristics of the sample to be detected of the object to be detected based on the biological sample analysis data, the identification result and the clinical information, and outputting auxiliary diagnostic decision information corresponding to the abnormal characteristics, includes:
inputting an auxiliary diagnostic model based on the report parameters in the biological sample analysis data, the identification result, and the clinical information;
and the auxiliary diagnosis model determines the abnormal characteristic type of the sample of the object to be detected according to the report parameters, the identification result and the clinical information, and outputs an auxiliary diagnosis report corresponding to the abnormal characteristic type.
The sample to be detected takes a blood sample and a recognition result of a target feature carried by analysis data of the biological sample as an example, and takes a recognition result of a target object carried by analysis images of blood cells by an image processing model as an example, blood report parameters in the analysis data of the blood cells, the recognition result of the analysis images of the blood cells by the image processing model and clinical information of the object to be detected as inputs of the auxiliary diagnosis model, and features are learned from a large amount of multi-mode data by using the auxiliary diagnosis model, which is similar to the brain of a professional examining doctor, has no forgetting curve, and can be optimized step by step along with the increase of the data. Alternatively, the auxiliary diagnosis model can be obtained after training by adopting a neural network model, and can also be completed by constructing an auxiliary diagnosis knowledge graph.
In the above embodiment, the auxiliary diagnosis model integrates multi-modal data, including but not limited to numerical data output by the biological sample analyzer, graphic data of biological sample analysis images represented by various histograms and scatter diagrams, text data represented by clinical information, and auxiliary diagnosis reports with higher output readability such as report parameters, and the auxiliary diagnosis results are obtained by simulating the brain of a professional examining doctor through the auxiliary diagnosis model, so that the auxiliary diagnosis results do not need to excessively depend on the personal experience level of the examining doctor, and are more objective, accurate and reliable.
In order to provide a more complete understanding of the auxiliary diagnostic method provided in the embodiments of the present application, a specific example will be described below, with reference to fig. 4 to 7, where the auxiliary diagnostic method includes:
s11, obtaining cell analysis data output by a cell analyzer; as shown in fig. 6, the cell analysis data may be blood cell analysis data obtained by detecting a blood sample of a patient by a blood cell analyzer, including data such as a histogram, a scatter diagram, a blood report parameter, a study parameter, a flag alarm, and the like;
s12, acquiring clinical information related to a patient from a test information system, a single instrument, a cascade analyzer and/or a blood analysis pipeline; the clinical information includes patient age, sex, test sample type, past medical history, etc. obtained from the test information system; blood cell analysis results, hemagglutination analysis results, biochemical analysis results, immunoassay results, blood cell microscopic examination results and the like obtained from a single instrument and/or a cascade analyzer; biochemical analysis results, immunological analysis results, hemagglutination analysis results, blood cytoscopy results, and the like obtained from the blood analysis line.
S13, extracting characteristics of the cell analysis data to obtain a recognition result of target cell characteristics in the cell analysis data; for example, for the cell image data in the cell analysis data, an image processing model may be used to perform feature extraction and identification on the cell image data, taking a histogram and a scatter plot in the blood cell analysis image as an example, the image processing model takes the histogram and the scatter plot in the blood cell analysis data as input, performs feature extraction on the histogram and the scatter plot, and outputs an identification result of the abnormal cell feature carried by the histogram and the scatter plot;
s14, inputting the identification result of the target cell characteristics carried by the cell analysis data, other report parameters in the cell analysis data and clinical information into an auxiliary diagnosis knowledge graph, determining the abnormal characteristics of the blood sample by the auxiliary diagnosis knowledge graph, and outputting auxiliary diagnosis decision information corresponding to the abnormal characteristics. Wherein, the auxiliary diagnosis decision information may include key information according to which an auxiliary diagnosis decision is made, referring to fig. 5, the auxiliary diagnosis decision information may be corrected based on the parameter alarm state output by the cell analyzer, and the basis for correcting the parameter alarm state is highlighted; or the analysis of unqualified reasons corresponding to the abnormal characteristics is included, and unqualified specimen decision logic is highlighted; or setting a standard format of an auxiliary diagnosis report, determining supporting evidence and/or diagnosis and treatment advice corresponding to the abnormal characteristics, and outputting the auxiliary diagnosis report containing the supporting evidence and/or diagnosis and treatment advice.
Referring to fig. 7, the auxiliary diagnostic report may be divided into a simplified diagnostic report and a detailed diagnostic report, wherein the simplified diagnostic report only includes auxiliary diagnostic results described in text, the simplified diagnostic report describes abnormal characteristics of a sample to be detected of a patient, includes abnormal types such as a change in the number of cells, occurrence or non-occurrence of abnormal cells, and the like, and simultaneously provides a diagnosis and treatment suggestion for the discovered abnormal characteristics; the detailed diagnosis report contains the text description and the auxiliary diagnosis results of result interpretation, the detailed diagnosis report describes the abnormal characteristics of the sample to be detected of the patient in detail, comprises the abnormal types such as the number change of cells, the occurrence of abnormal cells and the like, simultaneously gives out the next diagnosis and treatment suggestion aiming at the discovered abnormal characteristics, simultaneously gives out supporting evidence for the discovered abnormal characteristics, for example, gives out the reference range of normal values of corresponding parameters aiming at the abnormality of the numerical value, combines the clinical information such as the age, the sex and the like of the patient, and gives out the scatter diagram image of similar clinically diagnosed cases aiming at the abnormal cells and marks. For some special anomalies, the method can be further explained by combining a knowledge graph, and a reference image of the characteristics of the target cells which are matched with the anomalies and are found through microscopic examination is displayed according to the searching result of the historical diagnosis data.
In the above embodiment, on the one hand, the input data corresponding to the obtained auxiliary diagnostic decision information may be the original pulse signal itself or the pulse signal waveform diagram collected by the blood cell analyzer, the pulse signal itself or the pulse signal waveform diagram screened by the pulse recognition algorithm, the vector composed of pulse characteristic values such as the peak height, the front peak width, the rear peak width, the half peak width, etc. of the pulse signal, the histogram data or the histogram graph formed by the characteristic values of the pulse signal (such as RBC, PLT and WBC histograms formed by the peak heights of the pulse signal), the two-dimensional, three-dimensional or even multi-dimensional scatter data set composed of the characteristic values of the pulse signal in several dimensions (such as DIFF three-dimensional scatter data set composed of the peak heights of SFL, SSC and FSC three-channel pulse signals), the method can also be a scatter diagram graph formed by two-dimensional scatter points or a two-dimensional scatter diagram projection graph formed by three-dimensional and high-dimensional scatter data, wherein the input data are combined with clinical information (age, sex, sample type, microscopic examination result, past medical history and the like) of a patient, parameter reports and/or reference ranges of research parameters, and input data such as biochemical, immune, hemagglutination, blood smear microscopic examination result and the like obtained by combining a single instrument, a joint inspection instrument or a pipeline are combined, and abnormal sample feature extraction (such as reduction of platelet number, occurrence of immature granulocytes in a DIFF scatter diagram and the like) is performed through an AI model and/or an algorithm model established by a traditional algorithm. On the other hand, the auxiliary diagnosis decision information is based on parameter alarm state correction, unqualified specimen decision logic and auxiliary diagnosis reports in a set format, and the knowledge graph of the simulated examining physician decision route is formed together, so that further diagnosis and treatment scheme suggestions aiming at specific abnormal sample characteristics can be generated, key information for making the basis of the auxiliary diagnosis decision is provided, the readability is high, and the credibility of the auxiliary diagnosis decision can be further verified.
In step S13, feature extraction is performed on the cell analysis image in the cell analysis data by using an image processing model to identify the abnormal feature of the target cell carried by the corresponding cell analysis image, the image processing model can simulate the brain of a professional inspection doctor, intelligent identification of the cell analysis image output by the cell analyzer is completed, and the significant feature of the image in the output result is visualized.
Referring to fig. 8, in another aspect of the present application, an auxiliary diagnosis method is provided, which is applied to a deep learning model, and includes:
s201, acquiring a biological sample analysis image of an object to be detected; wherein the biological sample analysis image comprises a blood cell histogram, a blood cell scatter diagram, a pulse signal waveform diagram, a pulse signal characteristic diagram, a thromboelastography, a biochemical or immune response curve.
The object to be detected refers to the subject of the sample to be detected, and is typically a patient who provides a blood sample, taking the sample to be detected as a blood sample of a human body as an example. The biological sample analysis image comprises blood cell histograms corresponding to different cells in blood, blood cell scatter diagrams representing the distribution characteristics of various cells, pulse signal waveform diagrams, pulse signal characteristic diagrams and the like, wherein the blood cell histograms and the blood cell scatter diagrams can be as follows: platelet histogram, red cell histogram, white cell DIFF scatter plot, RET scatter plot, histogram formed by pulse characteristic values such as peak height, front peak width, back peak width, half peak width of pulse signal (such as RBC, PLT and WBC histogram formed by peak height of pulse signal), scatter plot formed by two-dimensional, three-dimensional or even multidimensional formed by characteristic values of pulse signal in several dimensions (such as DIFF three-dimensional scatter plot data set formed by peak height of SFL, SSC and FSC three-channel pulse signal), two-dimensional scatter plot projection plot formed by three-dimensional and high-dimensional scatter plot data; the pulse signal waveform diagram may be as follows: the blood cell analyzer detects a waveform chart formed by the collected original pulse signals of the blood sample and a pulse signal waveform chart formed by the pulse signals screened by the pulse recognition algorithm; the pulse signal feature map may be as follows: and a vector characteristic diagram formed by pulse characteristic values such as peak height, front peak width, rear peak width, half peak width and the like of the pulse signals. The different blood cell analysis images correspond to the characteristics of different cells in the bearing blood, and can be used for representing whether the content and the distribution of the corresponding cells in the blood are abnormal or not. Thrombi elastography is an index for reflecting blood coagulation dynamic changes (including the formation rate of fibrin, the firmness of dissolution and coagulation, elasticity); biochemical or immunological response curves are indicators used to characterize the content or concentration of a test substance.
S203, extracting the characteristics of the biological sample analysis image, and outputting the identification result of the target characteristics carried by the biological sample analysis image.
Optionally, referring to fig. 9, the deep learning model uses an image processing model as a main framework, and includes a plurality of sub-image models for respectively identifying target features carried by different types of biological sample analysis images. Taking a sample to be detected as a blood sample as an example, a biological sample analysis image at least includes a blood cell histogram and a blood cell scatter diagram, and the step S203 of extracting features from the biological sample analysis image and outputting a recognition result of target features carried by the biological sample analysis image includes:
s2031, extracting features of the blood cell histogram through a first sub-image model, and outputting a first recognition result of target cell features carried by the blood cell histogram.
The image processing model comprises a first sub-image model with a blood cell histogram as input. The first sub-image model may refer to a model obtained through deep learning and capable of being used for extracting key features representing whether the target object is carried in the image. Among them, deep Learning (DL) is a new research direction in the field of Machine Learning (ML), which is introduced into Machine Learning to make it closer to the original target, i.e., AI. Deep learning is the inherent regularity and presentation hierarchy of learning sample data, and the information obtained during such learning is helpful in interpreting data such as text, images and sounds. Its final goal is to have the machine have analytical learning capabilities like a person, and to recognize text, image, and sound data. Deep learning enables machines to simulate human brain activities such as audio-visual, thinking and the like in search technology, data mining, machine learning, machine translation, natural language processing, multimedia learning, voice recognition, recommendation and personalization technologies and other related fields to achieve a plurality of achievements, solves a plurality of complex pattern recognition problems, and enables related technologies of artificial intelligence to be greatly improved. The method comprises the steps of carrying out feature extraction on a blood cell histogram by a first sub-image model to obtain a first identification result of target cell features carried by the blood cell histogram, carrying out feature extraction on a feature vector formed by the feature extraction of the first sub-image model to represent image features of the corresponding blood cell histogram, and determining whether the specified target cell features exist in the blood cell histogram or not based on the feature vector. Taking the platelet histogram as an example, the target cell characteristics carried by the platelet histogram may refer to abnormal types of curve distribution forms in the platelet histogram, and may be left-shifted, right-shifted, tail-shaped, and the like on the right side of the histogram.
S2032, extracting the characteristics of the blood cell scatter diagram through a second sub-image model, and outputting a second identification result of the target cell characteristics carried by the blood cell scatter diagram.
The image processing model includes a second sub-image model that is input with the blood cell scatter plot. The second sub-image model is independent of the first sub-image model, and the blood cell histogram and the blood cell scatter diagram are processed and identified in parallel. The second sub-image model obtains a second recognition result of the target cell characteristics carried by the blood cell scatter diagram through feature extraction of the blood cell scatter diagram, the feature vector formed through feature extraction of the second sub-image model characterizes the image characteristics of the corresponding blood cell scatter diagram, and whether the specified target cell characteristics exist in the blood cell scatter diagram is determined based on the feature vector. Taking a lateral fluorescence scatter plot as an example, the target cell characteristics carried by the lateral fluorescence scatter plot may refer to a decrease in the number of leukocytes in the blood cell scatter plot, a left shift in neutrophil nuclei, and the like.
In the above embodiment, the image processing model performs feature extraction on the biological sample analysis image to identify the target feature carried by the biological sample analysis image. For a blood cell histogram and a blood cell scatter diagram in a blood cell analysis image, the image processing model adopts a framework that a first sub-image model and a second sub-image model are used for processing the blood cell histogram and the blood cell scatter diagram in parallel, the first sub-image model and the second sub-image model can respectively take abnormal cell characteristics carried in the blood cell histogram and the blood cell scatter diagram as identification objects, the first sub-image model can identify and output abnormal cell characteristics of the corresponding blood cell histogram as identification results through extracting the characteristics of the blood cell histogram, the second sub-image model can identify and output abnormal cell characteristics of the corresponding blood cell scatter diagram as identification results, and the blood cell histogram and the blood cell scatter diagram are independently and parallelly processed, so that the accuracy and the efficiency of the image processing model are improved, and support basis can be determined for abnormality existing in the blood cell analysis image.
Optionally, the deep learning model further comprises a feature conversion model for extracting features from non-image-class numerical data in the biological sample analysis data. The feature conversion model and the image processing model share a classification output layer, and the classification output layer is used for classifying the features of the blood cell analysis image output by the feature extraction layer of the first sub-image model and the second sub-image model and the features of the logarithmic value data output by the linear classification layer of the feature conversion model after the feature extraction result is spliced, and outputting the recognition result of the target blood cell features carried in the blood cell analysis data.
The deep learning model can process numerical data in the blood cell analysis data and image data in parallel by adopting a feature conversion model, for example, clinical information of a patient, blood report parameters, a blood cell histogram and a blood cell scatter diagram are respectively used as input of the deep learning model, and a recognition result of target blood cell features which are integrally represented by the blood cell analysis data corresponding to one sample to be detected is obtained by utilizing the self-learning of the deep learning model.
Referring to fig. 10 and 11 in combination, the image processing model may employ different neural network models, and in some embodiments, the first sub-image model and the second sub-image model are each an image classification neural network model;
The feature extraction is performed on the blood cell histogram through a first sub-image model, and a first recognition result of the target cell feature carried by the blood cell histogram is output, including: extracting features of the blood cell histogram through a first sub-image model, and outputting a first identification result carrying a classification label of target cell features carried by the blood cell histogram;
the feature extraction is performed on the blood cell scatter diagram through a second sub-image model, and a second recognition result of the target cell feature carried by the blood cell scatter diagram is output, including: and extracting the characteristics of the blood cell scatter diagram through a second sub-image model, and outputting a second identification result carrying a classification label of the target cell characteristics carried by the blood cell scatter diagram.
The classification label of the target cell characteristic carried by the blood cell histogram may refer to a preset type of the target cell characteristic carried by the blood cell histogram, the preset type may be determined according to different dominant appearances of different blood cells in the corresponding blood cell histogram, taking the platelet histogram as an example, the carried target cell characteristic may refer to an abnormal type of a curve distribution form in the blood cell histogram, which may be a left shift of the histogram, a right shift of the histogram, a trailing shape on the right side of the histogram, and the like, and the corresponding classification label includes a label 1 corresponding to the left shift of the histogram, a label 2 corresponding to the right shift of the histogram, and a label 3 corresponding to the right side of the histogram in a trailing shape. The feature extraction is performed on the blood cell histogram through the first sub-image model, and the output of the first identification result carrying the classification label of the target cell feature carried by the blood cell histogram may refer to a classification neural network model for classifying the blood cell histogram obtained through pre-training, and a classification result of the classification neural network model for judging whether the target cell feature carried by the blood cell histogram is of a specified target type or not is obtained. The classifying neural network model comprises a feature extraction layer and a classifying and predicting layer, the blood cell histogram is input into the classifying neural network model, the feature extraction layer is used for extracting the features of the blood cell histogram, the feature extraction layer can be provided with a plurality of layers, the last layer is connected with the classifying and predicting layer, the classifying and predicting layer is used for classifying and predicting according to the feature vectors output by the feature extraction layer, and the corresponding classifying label is determined.
In the same way, the classification label of the target cell characteristics carried by the blood cell scatter diagram can refer to a preset type of the target cell characteristics carried by the blood cell scatter diagram, the preset type can be determined according to different dominant manifestations of different blood cells in the corresponding blood cell scatter diagram, taking the lateral fluorescence scatter diagram as an example, the target cell characteristics carried by the blood cell scatter diagram can refer to leucocyte quantity reduction, neutrophil occurrence nucleus left shift and the like in the blood cell scatter diagram, and the corresponding classification label comprises a label 1 corresponding to leucocyte quantity reduction and a label 2 corresponding to neutrophil occurrence nucleus left shift. The feature extraction is performed on the blood cell scatter diagram through the second sub-image model, and the output of the second identification result carrying the classification label of the target cell feature carried by the blood cell scatter diagram may refer to a classification neural network model for classifying the blood cell scatter diagram obtained through pre-training, and a classification result of the classification neural network model for judging whether the target cell feature carried by the blood cell scatter diagram is of a specified target type is obtained. The blood cell scatter diagram is input into the classification neural network model, the characteristics of the blood cell scatter diagram are extracted through the characteristics extraction layer, the characteristics extraction layer can comprise a plurality of layers, the last layer is connected with the classification prediction layer, the classification prediction layer carries out classification prediction according to the characteristic vectors output by the characteristics extraction layer, and corresponding classification labels are determined.
In the above embodiment, the image processing model adopts the image classification neural network model, and the trained image classification neural network model compares the prediction category of the corresponding blood cell histogram and the blood cell scatter diagram with the standard target category, so as to determine the classification label of the target cell characteristics respectively carried by the corresponding blood cell histogram and the blood cell scatter diagram based on the similarity between the prediction category and the target category, thereby completing the intelligent recognition of the blood cell analysis image output by the blood cell analyzer by the brain of the substitution professional examining doctor, and visualizing the image salient features in the output result.
In some embodiments, before the obtaining the blood cell analysis image of the object to be detected, the method further includes:
acquiring a sample image set; the sample image set comprises a blood cell histogram carrying a classification label of the target cell characteristic as a first sample image and a blood cell scatter diagram carrying a classification label of the target cell characteristic as a second sample image; the classification labels of the first sample image comprise a positive sample label of small red blood cell interference and a negative sample label of normal morphology, and the classification labels of the second sample image comprise a positive sample label of abnormal immature granulocytes and a negative sample label of normal morphology;
Performing iterative training on an initial first image classification neural network model by adopting the first sample image until a first loss function of the first image classification neural network model converges, so as to obtain a trained first sub-image model;
and carrying out iterative training on the initial second image classification neural network model by adopting a second sample image until a second loss function of the second image classification neural network model converges, so as to obtain a trained second sub-image model.
The first sub-image model and the second sub-image model are obtained through pre-training, and when the method is implemented, an initial classification neural network model, such as a classical image classification neural network of ResNet50, inceptionV4 and the like, can be constructed, and the classification neural network model is trained by the following modes: firstly, acquiring a sample image, and carrying out category labeling on the sample image, wherein the category labeling can be carried out on the sample image according to the label information capable of uniquely characterizing the category identity of the target cell characteristics, if the category labeling of the image containing the appointed target object 1 in the sample image corresponds to 1, the category labeling of the image containing the appointed target object 2 in the sample image corresponds to 2, and the category labeling of the image not containing any appointed target object in the sample image corresponds to 0, so as to obtain the sample image labeled with the target category; and then inputting the sample image marked with the target category into a classification neural network model, carrying out category prediction on an object borne by the sample image through the classification neural network model, comparing the prediction category with the standard target category, determining the value of a loss function of the classification neural network model based on the difference between the prediction category and the standard target category, reversely transmitting the value of the loss function into each layer of the classification neural network model, and updating model parameters of each layer through a random gradient descent method (SGD, stochastic Gradient Descent) to realize training of the neural network model.
The sample image for training the first sub-image model adopts a labeled blood cell histogram, the sample image comprises a positive sample image and a negative sample image, and in the embodiment, the classification label of the first sample image comprises a positive sample label of small red blood cell interference and a negative sample label of normal morphology. And carrying out feature extraction on the input blood cell histogram of the object to be detected through the first sub-image model after training, identifying whether the blood cell histogram has small red blood cell interference or is in a normal form, and correspondingly outputting a classification label of the target cell feature carried by the blood cell histogram as a classification label of the small red blood cell interference or a classification label of the normal form. Similarly, the sample image for training the second sub-image model adopts a labeled blood cell scatter diagram, the sample image comprises a positive sample image and a negative sample image, and the classification label of the second sample image comprises a positive sample label of immature granulocyte abnormality and a negative sample label of normal morphology. And performing feature extraction on the input blood cell scatter diagram of the object to be detected through the second sub-image model after training, identifying whether the blood cell scatter diagram has abnormal immature granulocytes or normal morphology, and correspondingly outputting a classification label of the target cell features carried by the blood cell scatter diagram as a classification label of abnormal immature granulocytes or a classification label of normal morphology. The classified neural network model can comprise a pooling layer, a visual characteristic visualization algorithm such as guide-Backpropagation, grad-CAM, score-CAM, group-CAM and the like is adopted, and the image area which determines the corresponding classified label is used as a more focused area of the neural network model to be marked on the original image by utilizing the spatial corresponding relation between the characteristic mapping output by the last convolution layer and the original image and is output synchronously with the classified label.
Optionally, the image processing model includes a regression layer, and a counter-propagating neural network may be used, and through training of sample data, the network weights and thresholds are continuously corrected to make the error function drop along the negative gradient direction, so as to approach the expected output.
In the above embodiment, the first sub-image model and the second sub-image model are respectively obtained after training by using an image classification neural network, and different types of image data respectively establish an independent classification neural network to perform feature extraction, so as to complete intelligent recognition of the blood cell analysis image output by the blood cell analyzer by the brain of a professional examining doctor.
In some embodiments, the first sub-image model and the second sub-image model are each a target detection neural network model;
the feature extraction is performed on the blood cell histogram through a first sub-image model, and a first recognition result of the target cell feature carried by the blood cell histogram is output, including: extracting features of the blood cell histogram through a first sub-image model, and outputting a first recognition result containing an image region mark where the target cell features carried by the blood cell histogram are located;
the feature extraction is performed on the blood cell scatter diagram through a second sub-image model, and a second recognition result of the target cell feature carried by the blood cell scatter diagram is output, including: and extracting the characteristics of the blood cell scatter diagram through a second sub-image model, and outputting a second identification result containing the image region labels of the target cell characteristics carried by the blood cell scatter diagram.
The first sub-image model adopts a target detection neural network model, detects the position of a target object in a corresponding blood cell histogram in an image according to different dominant manifestations of different blood cells in the corresponding blood cell histogram, and outputs a first recognition result containing an image region mark of a target cell characteristic carried by the blood cell histogram.
In this embodiment, taking the PLT histogram as an example, the target cell features carried by the PLT histogram may refer to that the PLT histogram has small red blood cell interference and normal morphology, and the image region label corresponding to the target cell features refers to an image dividing line of a local image region in the PLT histogram, where the small red blood cell interference exists, can be determined according to the image dividing line. The feature extraction is performed on the blood cell histogram through the first sub-image model, and the output of the first identification result carrying the image region label where the target cell feature carried by the blood cell histogram is located may refer to a target detection neural network model for detecting whether the small red blood cell interference exists in the blood cell histogram obtained through pre-training, and a target detection result for judging whether the small red blood cell interference exists on the target cell feature carried by the blood cell histogram through the target detection neural network model is obtained. Optionally, the target detection neural network model includes a convolution layer and a RPN (region proposal networks) network layer, the convolution layer performs feature extraction on an input image to obtain a feature map, and the RPN (region proposal networks) network layer uses a priori anchor to output a rectangular candidate region set with objectless socre, so as to determine an image region where a corresponding target cell feature is located.
In the same way, the labeling of the image area where the target cell features carried by the blood cell scattergram are located may mean that the image dividing line of the local image area where the immature granulocyte abnormality exists can be determined in the blood cell scattergram according to the labeling. The feature extraction is performed on the blood cell scatter diagram through the second sub-image model, and the output of the second identification result carrying the labeling of the image region where the target cell feature carried by the blood cell scatter diagram is located may refer to the target detection neural network model which is obtained through pre-training and used for detecting whether the immature granulocyte abnormal image region exists in the blood cell scatter diagram, and the target detection result of the target detection neural network model for judging whether the immature granulocyte abnormal region exists on the target cell feature carried by the blood cell scatter diagram is obtained.
In the above embodiment, the image processing model adopts the target detection neural network model, and detects whether the corresponding blood cell histogram and blood cell scatter diagram have the preset cell abnormal feature type through the trained target detection neural network model, so as to compare the detected cell abnormal feature type with the standard cell abnormal feature type, determine the type and the position of the target cell feature carried by the corresponding blood cell histogram and blood cell scatter diagram respectively, complete the intelligent recognition of the blood cell analysis image output by the blood cell analyzer instead of the brain of the professional examining doctor, and enable the image salient feature in the output result to be visualized.
In some embodiments, before the obtaining the blood cell analysis image of the object to be detected, the method further includes:
acquiring a sample image set; the sample image set comprises a first sample image which is provided with a blood cell histogram marked in an image area where the target cell characteristics are located and a second sample image which is provided with a blood cell scatter diagram marked in an image area where the target cell characteristics are located; the image region labels of the first sample image comprise positive sample labels of small red blood cell interference image region labels and negative sample labels of normal morphology, and the image region labels of the second sample image comprise positive sample labels of immature granulocyte abnormal image region labels and negative sample labels of normal morphology;
performing iterative training on an initial first target detection neural network model by adopting the first sample image until a first loss function of the first target detection neural network model converges, so as to obtain a trained first sub-image model;
and carrying out iterative training on the initial second target detection neural network model by adopting a second sample image until a second loss function of the second target detection neural network model converges, so as to obtain a trained second sub-image model.
The first sub-image model and the second sub-image model are obtained through pre-training, and when the method is implemented, an initial target detection neural network model, such as a YoloV5 classical target detection neural network, a RetinaNet classical target detection neural network, can be built, and the target detection neural network model is trained in the following manner: firstly, obtaining a sample image, marking the type of a target object and the area of the sample image, wherein the sample image can be marked according to the type identity capable of uniquely representing the characteristics of target cells and the image dividing line of the corresponding position, if the type marking of the image containing the designated target object 1 in the sample image is 1, the image area of the target object 1 is marked by the image dividing line, the type marking of the image containing the designated target object 2 in the sample image is 2, the image area of the target object 2 is marked by the image dividing line, and the type marking of the image not containing any designated target object in the sample image is 0, so as to obtain a marked sample image; and inputting the marked sample image into a target detection neural network model to detect a target object borne by the sample image through the target detection neural network model, comparing the detected abnormal characteristics of the target cells with the abnormal characteristics of cells in the standard sample image to determine the value of a loss function of the target detection neural network model based on the difference between the detected abnormal characteristics of the target cells and the abnormal characteristics of cells in the standard sample image, reversely transferring the value of the loss function to each layer of the target detection neural network model, and updating the model parameters of each layer through a random gradient descent method (SGD, stochastic Gradient Descent) to realize training of the neural network model.
The sample image for training the first sub-image model adopts a labeled blood cell histogram, the sample image comprises a positive sample image and a negative sample image, and in the embodiment, the image region label of the first sample image comprises a positive sample label labeled by a small red blood cell interference image region and a negative sample label labeled by a morphological normal. And carrying out feature extraction on the input blood cell histogram of the object to be detected through the first sub-image model after training, identifying whether the blood cell histogram has small red blood cell interference or is normal in morphology, and correspondingly outputting the classification label of the target cell feature carried by the blood cell histogram as the classification label of the small red blood cell interference and the image area label where the classification label is positioned or the classification label of the normal morphology.
Similarly, a labeled blood cell scatter diagram is adopted for a sample image for training the second sub-image model, the sample image comprises a positive sample image and a negative sample image, and the image region label of the second sample image comprises a positive sample label and a negative sample label with normal morphology, which are labeled in the immature granulocyte abnormal image region. And extracting the characteristics of the input blood cell scatter diagram of the object to be detected through the second sub-image model after training, identifying whether the blood cell scatter diagram has abnormal immature granulocytes or normal morphology, and correspondingly outputting the classification label of the target cell characteristics carried by the blood cell scatter diagram as the classification label of the abnormal immature granulocytes and the image area label thereof or the classification label of the normal morphology.
In the above embodiment, the first sub-image model and the second sub-image model are respectively obtained after training by using the target detection neural network, and different types of image data respectively establish an independent target detection neural network to perform feature extraction, so as to complete intelligent recognition of the blood cell analysis image output by the blood cell analyzer by the brain of the professional examining physician.
In some embodiments, the first sub-image model and the second sub-image model are each an image-segmented neural network model;
the feature extraction is performed on the blood cell histogram through a first sub-image model, and a first recognition result of the target cell feature carried by the blood cell histogram is output, including: extracting features of the blood cell histogram through a first sub-image model, and outputting a first identification result of image region segmentation labels containing target cell features carried by the blood cell histogram;
the feature extraction is performed on the blood cell scatter diagram through a second sub-image model, and a second recognition result of the target cell feature carried by the blood cell scatter diagram is output, including: and extracting the characteristics of the blood cell scatter diagram through a second sub-image model, and outputting a second identification result containing the image region segmentation labels of the target cell characteristics carried by the blood cell scatter diagram.
In this embodiment, taking the PLT histogram as an example, the target cell features carried by the PLT histogram may refer to that the PLT histogram has small red blood cell interference and normal morphology, and the segmentation labeling of the image region where the corresponding target cell features are located refers to that the PLT histogram can determine that the image of the local image with the small red blood cell interference is enhanced and dominant. The feature extraction is performed on the blood cell histogram through the first sub-image model, and the output of the first identification result carrying the segmentation marking of the image region where the target cell feature carried by the blood cell histogram is located may refer to an image segmentation neural network model for detecting whether the small red blood cell interference exists in the blood cell histogram obtained through pre-training, and a target detection result of the image segmentation neural network model for judging whether the small red blood cell interference exists in the local image of the target cell feature carried by the blood cell histogram is obtained.
In the same way, the segmentation and labeling of the image region where the target cell features carried by the blood cell scatter diagram are located can refer to the image enhancement display representation of the local image in which the immature granulocyte abnormality can be determined in the blood cell scatter diagram. The feature extraction is performed on the blood cell scatter diagram through the second sub-image model, and the output of the second identification result carrying the segmentation marking of the image region where the target cell feature carried by the blood cell scatter diagram is located can be that an image segmentation neural network model for detecting whether an immature granulocyte abnormal image exists in the blood cell scatter diagram or not is obtained through pre-training, and a target detection result of the image segmentation neural network model for judging whether the partial image of the immature granulocyte abnormal exists or not on the target cell feature carried by the blood cell scatter diagram is obtained.
In the above embodiment, the image processing model adopts an image segmentation neural network model, and detects whether the corresponding blood cell histogram and blood cell scatter diagram have preset abnormal cell feature types through the trained image segmentation neural network model, so as to compare the detected abnormal cell feature types with standard abnormal cell feature types to determine the types and the positions of target cell features respectively carried by the corresponding blood cell histogram and blood cell scatter diagram, thereby completing the intelligent recognition of the blood cell analysis image output by the blood cell analyzer instead of the brain of a professional inspector, and visualizing the significant image features in the output result.
In some embodiments, before the obtaining the blood cell analysis image of the object to be detected, the method further includes:
acquiring a sample image set; the sample image set comprises a blood cell histogram carrying image region segmentation labels of target cell features as a first sample image and a blood cell scatter diagram carrying image region segmentation labels of the target cell features as a second sample image; the image region segmentation labels of the first sample image comprise labeled positive sample labels for segmenting small red blood cells to interfere with the image region and labeled negative sample labels with normal morphology, and the image region segmentation labels of the second sample image comprise labeled positive sample labels for segmenting five-class abnormal image regions and labeled negative sample labels with normal morphology;
Performing iterative training on an initial first image segmentation neural network model by adopting the first sample image until a first loss function of the first image segmentation neural network model converges, so as to obtain a trained first sub-image model;
and carrying out iterative training on the initial second image segmentation neural network model by adopting a second sample image until a second loss function of the second image segmentation neural network model converges, so as to obtain a trained second sub-image model.
The first sub-image model and the second sub-image model are obtained through pre-training, and when the method is implemented, an initial image segmentation neural network model, such as a classical image segmentation neural network of U-Net, deep LabV3 and the like, can be constructed, and the image segmentation neural network model is trained in the following manner: firstly, acquiring a sample image set, carrying out target object type and image position on the sample image, wherein the sample image can be marked according to the type identity capable of uniquely representing the target cell characteristics and the image enhancement display representation of the corresponding position, wherein the image enhancement display can be enhanced display by adopting different colors, for example, the type label of the image containing the appointed target object 1 in the sample image is corresponding to 1, the image area where the target object 1 is positioned is marked by local image enhancement display, the type label of the image containing the appointed target object 2 in the sample image is corresponding to 2, the image area where the target object 2 is positioned is marked by local image enhancement display, and the type label of the image not containing any appointed target object in the sample image is corresponding to 0, so as to obtain the marked sample image; and inputting the marked sample image into an image segmentation neural network model to detect a target object carried in the sample image through the image segmentation neural network model, comparing the detected target cell abnormal characteristic with the cell abnormal characteristic in the standard sample image to determine the value of a loss function of the image segmentation neural network model based on the difference between the detected target cell abnormal characteristic and the cell abnormal characteristic in the standard sample image, reversely transmitting the value of the loss function to each layer of the image segmentation neural network model, and updating the model parameters of each layer through a random gradient descent method (SGD, stochastic Gradient Descent) to realize training of the neural network model.
The sample image for training the first sub-image model adopts a blood cell histogram with image area segmentation labeling, the sample image comprises a positive sample image and a negative sample image, and in the embodiment, the image area labeling of the first sample image comprises a positive sample label and a negative sample label with normal morphology, wherein the positive sample label is displayed by enhancing a local area of the small red blood cell interference image. And carrying out feature extraction on the input blood cell histogram of the object to be detected through the first sub-image model after training, identifying whether the blood cell histogram has small red blood cell interference or normal morphology, and correspondingly outputting the classification label of the target cell feature carried by the blood cell histogram as the classification label of the small red blood cell interference and the local area enhancement display of the image where the classification label is positioned or the classification label of the normal morphology.
Similarly, the blood cell scatter diagram of the local area enhanced display is adopted for the sample image for training the second sub-image model, the sample image comprises a positive sample image and a negative sample image, and the image area label of the second sample image comprises a positive sample label and a negative sample label with normal morphology of the local image area label of the immature granulocyte abnormality. And performing feature extraction on the input blood cell scatter diagram of the object to be detected through the second sub-image model after training, identifying whether the blood cell scatter diagram has abnormal immature granulocytes or normal morphology, and correspondingly outputting the classification label of the target cell features carried by the blood cell scatter diagram as the classification label of the abnormal immature granulocytes and the local image area enhancement display label of the classification label or the classification label of the normal morphology.
In the above embodiment, the first sub-image model and the second sub-image model are respectively obtained after training by using an image segmentation neural network, and different types of image data respectively establish an independent image segmentation neural network to perform feature extraction, so as to complete intelligent recognition of the blood cell analysis image output by the blood cell analyzer by replacing the brain of a professional examining doctor.
In some embodiments, the acquiring a cell analysis image of the object to be detected comprises:
and obtaining a blood cell histogram and a blood cell scatter diagram of the object to be detected output by the blood cell analyzer.
The image processing model is in communication connection with the blood cell analyzer, and directly takes a blood cell histogram and a blood cell scatter diagram output by the blood cell analyzer as input. The blood cell histogram may include histograms corresponding to different blood cells in blood, such as a platelet histogram, a red blood cell histogram, a white blood cell histogram, etc.; the blood cell scatter plot may include different scatter plots obtained by generating light scattering in different directions for each blood cell by detecting the scattered light, converting the light signal into electrical pulses, such as a white blood cell DIFF scatter plot, a RET scatter plot, and the like.
The characteristics of different cells in the blood are correspondingly carried by different blood cell analysis images, so that whether the content and distribution of the corresponding cells in the blood are abnormal or not can be characterized, and the characteristics of the target cells carried in the blood cell analysis images refer to the characteristics of different types of cells correspondingly carried by the blood cell analysis images. In this embodiment, the image processing model takes a blood cell histogram and a blood cell scattergram as input, performs feature extraction on the blood cell histogram and the blood cell scattergram, and outputs recognition results of target cell features carried by the blood cell histogram and the blood cell scattergram, respectively.
In the above embodiment, the image processing model cooperates with the blood cell analyzer, and directly obtains the blood cell analysis image output by the blood cell analyzer for analysis, so as to accurately determine whether the blood sample of the object to be detected has an abnormality, so as to provide auxiliary diagnosis for the examining physician.
In step S14, the identification result of the target cell feature carried by the cell analysis data, the blood report parameter in the cell analysis data, and the clinical information are comprehensively determined by using the auxiliary diagnosis knowledge graph to determine the abnormal feature of the sample to be detected, and the auxiliary diagnosis decision information corresponding to the abnormal feature is output. The auxiliary diagnosis knowledge graph can be constructed based on a deep learning model or a traditional algorithm model by taking a recognition result of target characteristics, report parameters and clinical information in biological sample analysis data as input.
Referring to fig. 12, in another aspect of the present application, a knowledge graph construction method is further provided, which is applied to an auxiliary diagnostic device, and includes:
s301, forming an entity set in a knowledge base according to a recognition result of target features carried by biological sample analysis data, clinical information of a detection object and sample reference parameters;
S303, forming an attribute set and a relationship set in a knowledge base according to the identification result, the clinical information, the comparison result of the report parameter relative to the sample reference parameter and the mapping relation of the historical diagnosis example in the diagnosis example database;
s305, forming a triplet set according to the entity set, the attribute set and the relation set, and constructing an auxiliary diagnosis knowledge graph according to the triplet set.
The sample reference parameters refer to reference values corresponding to detection index parameters respectively contained in different sample types to be detected, the sample reference parameters can be obtained based on industry standards, for example, in blood reference parameters, the analysis parameter interval of the blood cells of children can be obtained based on the sanitary industry standard WS/T779-2021 of the people's republic of China. The sample reference parameters of different sample types to be detected comprise different numbers and types of detection index parameters, and the value ranges of the reference values of the detection index parameters in the sample reference parameters of the same sample types to be detected are different. The historical diagnosis example refers to a detection result of determining whether a sample is abnormal based on a biological sample analysis image, clinical information and sample reference parameters of an object to be detected, for example, taking the sample to be detected as a blood sample, any blood detection record of a user can form a historical diagnosis example, for example, according to whole blood detection performed by a user a at time 1, the whole blood detection record a1 correspondingly comprises blood cell analysis data output by a blood detector for detecting the blood sample of the user a at time 1, clinical information correspondingly input by the blood detector a at time 1 in a test information system, and comprehensive judgment made by a test doctor by using the data, and the whole blood detection record a1 can be used as a diagnosis example of a diagnosis example database.
The blood cell analysis image output by the blood analyzer in the whole blood detection record a1 can be used as a training sample for training the initial neural network model after being marked, and a trained image processing model is obtained. And taking a blood cell analysis image output by the blood analyzer in the whole blood detection record a1 as input of an image processing model, extracting characteristics of the blood cell analysis image through the image processing model, outputting a recognition result of a target object carried by the blood cell analysis image, and determining abnormal characteristics of a blood sample of the object to be detected by using the recognition result of the image processing model on the target object carried by the blood cell analysis image, clinical information of the detection object and blood reference parameters as input of an auxiliary diagnosis knowledge map, and outputting an auxiliary diagnosis report containing descriptions corresponding to the abnormal characteristics. The auxiliary diagnosis knowledge graph takes triples as nodes for constructing the knowledge graph, each triplet can be characterized in the forms of (entity, relation, entity), (entity, attribute value) and the like, an entity set is formed by extracting the entity from input data of the auxiliary diagnosis knowledge graph, an attribute set and a relation set are respectively formed according to the auxiliary diagnosis knowledge graph based on the extracted attribute and relation in the mapping relation between the input and diagnosis cases in a diagnosis case database which is depended on in the determination output process, and a triplet set is formed by establishing triples according to the relation between the entity and the attribute value of the entity, namely, the node set for constructing the auxiliary diagnosis knowledge graph is formed; and constructing an auxiliary diagnosis knowledge graph based on the node set.
In the above embodiment, by constructing the auxiliary diagnosis knowledge graph and combining the auxiliary diagnosis knowledge graph with the image processing model to determine the abnormal characteristics of the sample according to the identification result, the clinical information and the corresponding report parameters of the biological sample analysis image of the object to be detected, the examining physician can be assisted to rapidly screen the sample with the abnormality in the current biological sample detection result and determine the basis of the abnormality, so that the examining physician can make the next diagnosis and treatment decision more efficiently according to the auxiliary diagnosis report.
In some embodiments, the sample to be detected is a blood sample, the recognition result of the target feature carried by the analysis data of the biological sample mainly includes the recognition result of the image processing model on the target object carried by the analysis image of the blood cell, the image processing model includes a parallel architecture formed by the first sub-image model and the second sub-image model, the recognition result of the target cell feature carried by the histogram of the blood cell and the scatter plot of the blood cell is respectively performed, and the knowledge graph construction method further includes:
and acquiring a first identification result of the target cell characteristics carried by the blood cell histogram and the second identification result of the target cell characteristics carried by the blood cell scatter diagram.
The image processing model comprises a first sub-image model and a second sub-image model which are respectively processed independently and in parallel on a blood cell histogram and a blood cell scatter diagram output by the blood cell analyzer. The input of the construction of the auxiliary diagnosis knowledge graph comprises a first identification result of respectively carrying out identification output on the target cell characteristics carried by the blood cell histogram by the image processing model and a second identification result of respectively carrying out identification output on the target cell characteristics carried by the blood cell scatter diagram, so that the constructed auxiliary diagnosis knowledge graph can take the blood cell histogram and the blood cell scatter diagram obtained by detecting the blood sample of the object to be detected by the blood cell analyzer as input, and form a final auxiliary diagnosis report by combining other report parameters and related clinical information of the object to be detected, the auxiliary diagnosis knowledge graph converts the detection result of the blood sample into the auxiliary diagnosis report with higher readability, the abnormal sample in the current blood sample detection result can be conveniently and rapidly screened out by the auxiliary diagnosis report, and the basis of the abnormality is determined, so that a convenient examination doctor can make a next diagnosis and treatment decision more efficiently according to the auxiliary diagnosis report.
In some embodiments, the constructing an auxiliary diagnostic knowledge-graph from the triplet set includes:
determining a final node according to the triples with attribute values described by the diagnostic results in the diagnosis example database, determining a head node for the triples of the target object carried by the image processing model on the blood cell analysis image, the clinical information of the detection object and the blood reference parameters, and determining an intermediate node according to the triples related to the connection event from the head node to the final node;
and constructing an auxiliary diagnosis knowledge graph based on the first node, the intermediate node and the last node.
The diagnosis result description in the diagnosis example database may include: a textual description of a diagnosis and treatment recommendation corresponding to the abnormal characteristic of the blood sample; or text description of supporting evidence corresponding to abnormal characteristics of the blood sample, image data carrying labels and numerical description.
Referring to fig. 13, in an alternative specific example, the recognition result of the image processing model on the target object carried by the blood cell analysis image includes that the PLT histogram tag 1 is small red blood cell interference, and the RBC histogram tag 0 is morphological normal; the blood report parameters comprise corresponding parameter values of RBC, HGB, HCT, WBC, MCV, MCH, RDW-CV and RDW-SD; clinical information includes patient age, sex, and sample type. The diagnosis example in the diagnosis example database comprises the steps of determining that the platelet false is higher based on the presence of the small red cell interference and the small red cell in the PLT histogram, and determining that the platelet false is higher based on the higher PLT parameter value in the comparison result of the patient blood report parameter relative to the blood reference parameter; wherein, the presence of small-cell erythrocytes is determined based on the RBC histogram morphology normal and the parameter values of MCV, MCHC in the patient blood report parameters. Thus, the first node of the auxiliary diagnostic knowledge graph may include (PLT histogram, tag 1, small red blood cell interference), (RBC histogram, tag 0, morphologically normal), (blood reporting parameters, values corresponding to each parameter), (clinical information, basic information, age, sex, and sample type); the intermediate nodes may include (RBC histogram morphology is normal, MCV, MCH parameter values are low, small cell red blood cells in the comparison result of blood reporting parameters relative to blood reference parameters), (patient clinical information, parameter normal range, corresponding blood reference parameters), and the end nodes may include (PLT histogram has small cell interference, small cell red blood cells, platelet false high), (patient blood reporting parameters, PLT parameter values are high, platelet false high in the comparison result of blood reporting parameters relative to the blood reference parameters), and the auxiliary diagnostic knowledge map is constructed according to the mapping relation relied on from the first node to the determined end node.
Referring to fig. 14, in another alternative specific example, the recognition result of the image processing model on the target object carried by the blood cell analysis image includes that the PLT histogram tag 1 is shorter in PLT histogram, and the DIFF scatter plot tag 1 is that the DIFF two-dimensional main view has immature granulocytes; the blood reporting parameters comprise parameter values corresponding to PLTs; clinical information includes patient age, sex, and sample type. The diagnosis examples in the diagnosis example database comprise the steps of determining that the PLT value is lower based on the PLT histogram is lower and combining the comparison result of the PLT parameter value and the parameter reference range of the patient to determine that the PLT parameter value is lower than the reference value, and correspondingly, assisting in diagnosis decision information such as reduction of the number of platelets, and suggesting to check related tests such as autoantibodies, platelet related immunoglobulins and the like for clear diagnosis. The diagnosis example in the diagnosis example database comprises the steps that the DIFF two-dimensional main view has the immature granulocytes, the comparison result of the parameter values combined with the WBC, IG and IG proportion and the parameter reference range of the patient is that the parameter values of the WBC, IG and IG proportion are larger than the reference value, the white blood cell values are higher, the immature granulocytes are determined to exist, the corresponding auxiliary diagnosis decision information is that the white blood cell number is increased and the immature granulocytes are accompanied, and the relevant examination such as the myelocytometry examination is recommended if necessary.
Therefore, the first node of the auxiliary diagnosis knowledge graph comprises (PLT histogram, label 1, PLT histogram is short), (DIFF scatter diagram, label 1, DIFF two-dimensional main view has immature granulocyte), (blood report parameter, value corresponding to each parameter), (clinical information, basic information, age, sex and sample type); the intermediate node comprises (DIFF two-dimensional main view has immature granulocytes, WBC, IG and IG ratio parameter values larger than a reference value, white blood cell values higher, immature granulocytes, patient clinical information, parameter normal range and corresponding blood reference parameters), (PLT histogram shorter, PLT parameter values and patient parameter reference range comparison results are that the PLT parameter values are smaller than the reference value, white blood cell values higher, immature granulocytes; the end node comprises: (there is little red blood cell interference, little cell red blood cell, platelet false is high) the comparison result of the patient blood report parameter and the blood report parameter relative to the blood reference parameter shows that the PLT parameter value is high and the platelet false is high). And constructing an auxiliary diagnosis knowledge graph according to the mapping relation relied on by the connection event from the first node to the determined end node.
In the above embodiment, the input of the image processing model may include a platelet histogram, a red blood cell histogram, a white blood cell histogram, a peak height, a front peak width, a rear peak width, a half peak width, etc. of a pulse characteristic value formed by a platelet analyzer to perform detection analysis on a blood sample of a subject to be detected, a histogram (such as RBC, PLT, and WBC histograms formed by peak heights of the pulse signal), a white blood cell DIFF scattergram, a RET scattergram, a two-dimensional, three-dimensional, or even multi-dimensional scattergram (such as a DIFF three-dimensional scattergram dataset formed by a SFL, SSC, and FSC three-channel pulse signal), or an original pulse signal for constructing the image data, the input of the auxiliary diagnosis knowledge graph includes a recognition result of the image processing model on the image data, and may further include a blood report parameter outputted by the platelet analyzer, a clinical information related to a patient from a hospital LIS system, an auxiliary diagnosis report with better readability formed by the auxiliary diagnosis map, a two-dimensional or three-dimensional or even multi-dimensional scattergram containing description, a description and a scatter plot formed by characteristic value description to provide a support for the auxiliary diagnosis, and a white blood cell report from a differential image-feature map, or a white blood cell image-feature-derived from a similar image feature-to a similar image feature map, such as a value-derived from a differential image-feature database, or a similar image-derived image feature.
In some embodiments, the method for constructing the auxiliary diagnosis knowledge graph further includes:
acquiring biological sample analysis data of an object to be detected, extracting characteristics of a biological sample analysis image in the biological sample analysis data through an image processing model, and outputting a recognition result of a target object carried by the biological sample analysis image;
acquiring clinical information of the object to be detected;
inputting the report parameters, the identification results and the clinical information in the biological sample analysis data into an auxiliary diagnosis knowledge graph, wherein the auxiliary diagnosis knowledge graph determines abnormal characteristics of a sample to be detected of the object to be detected according to the report parameters, the identification results and the clinical information, and outputs an auxiliary diagnosis report containing descriptions corresponding to the abnormal characteristics;
and updating the diagnosis instance database according to the verified auxiliary diagnosis report.
After the auxiliary diagnosis knowledge graph is constructed, the diagnosis example database is updated in the application process, the updated diagnosis example database can upgrade and perfect the auxiliary diagnosis knowledge graph, no forgetting curve exists at any time, and the auxiliary diagnosis knowledge graph can be better and better represented as the data of the diagnosis example are more and more.
The auxiliary diagnostic method provided by the above embodiment has at least the following advantages:
firstly, constructing a deep learning model based on image processing, directly identifying biological sample image data of a biological sample analyzer, such as a blood cell analysis image output by the blood cell analyzer through the image processing model, efficiently and accurately screening an output result of the blood cell analyzer by using the image processing model instead of a checking doctor, screening out a blood sample with abnormality, providing a source for abnormality determination in the blood cell analysis image, and improving the checking efficiency, wherein a diagnosis example of a professional checking doctor can be used as mode data learned by the image processing model, so that the checking accuracy is improved;
secondly, constructing an auxiliary diagnosis knowledge graph in combination with a recognition result of target features carried in biological sample analysis data, wherein the recognition result can be a recognition result of a blood cell analysis image of an object to be detected by an image processing model, and the clinical information and corresponding blood report parameters determine abnormal features of blood samples, so that an auxiliary diagnosis report containing explanatory description of the abnormality can be formed by assisting a checking doctor in rapidly screening out the blood samples with the abnormality in the current blood sample detection result and determining the basis of the abnormality;
Thirdly, the auxiliary diagnosis equipment extracts target characteristics of image type, numerical type and/or text type data obtained from the biological sample analyzer through a deep learning model or a traditional algorithm model, for example, an image processing model identifies a blood cell analysis image output by the blood cell analyzer, and intelligent identification of blood sample detection is completed through other blood cell analysis data output by the auxiliary diagnosis knowledge image comprehensive blood cell analyzer, identification results of the image processing model and relevant clinical data, so that a numerical report displayed by the blood cell analyzer is converted into a descriptive report, and an accurate auxiliary diagnosis result can be obtained without depending on the personal experience level of a checking doctor.
In another aspect of the embodiment, referring to fig. 15, an auxiliary diagnostic apparatus is further provided, which includes a first obtaining module 311 configured to obtain biological sample analysis data of an object to be detected, and a processing module 312 configured to perform feature extraction on the biological sample analysis data, and output a recognition result of a target feature carried in the biological sample analysis data; a second acquisition module 313, configured to acquire clinical information of the object to be detected; the decision module 314 is configured to determine an abnormal feature of a sample to be detected of the object to be detected based on the biological sample analysis data, the identification result, and the clinical information, and output auxiliary diagnostic decision information corresponding to the abnormal feature.
Optionally, the first obtaining module 311 is specifically configured to obtain biological sample analysis data of an object to be detected, and the processing module 312 is configured to perform feature extraction on a biological sample analysis image in the biological sample analysis data, and output an identification result of a target object carried by the biological sample analysis image.
Optionally, the first obtaining module 311 is specifically configured to obtain biological sample analysis data of the object to be detected output by the biological sample analyzer; the processing module 312 performs feature extraction on the biological sample analysis images in the biological sample analysis data, and outputs recognition results of target features respectively carried by the biological sample analysis images; the biological sample analysis image includes at least one of: a blood cell histogram, a blood cell scatter plot, a pulse signal waveform plot, a pulse signal signature plot, a thromboelastography, a biochemical or immune response curve.
Optionally, the first obtaining module 311 obtains biological sample analysis data of the object to be detected; the biological sample analysis data includes at least one of: a blood cell histogram, a blood cell scatter diagram, a pulse signal waveform diagram, a pulse signal feature diagram, a one-dimensional pulse signal vector, a pulse signal feature value vector, a multi-dimensional scatter data set, elasticity diagram data and immune response original data; the auxiliary diagnosis device further comprises a deep learning model or a traditional algorithm model, performs feature extraction on the biological sample analysis data, and outputs a recognition result of target features carried in the biological sample analysis data.
Optionally, the auxiliary diagnostic device further comprises a deep learning model for extracting features of the biological sample analysis data through the deep learning model based on image classification; or, the traditional algorithm model is used for extracting the characteristics of the biological sample analysis data through the traditional algorithm model constructed based on an image morphology algorithm, an image classification algorithm, a clustering algorithm or a threshold segmentation algorithm; classifying according to the feature extraction result, and outputting the identification result of the target feature carried in the biological sample analysis data.
Optionally, the second obtaining module 312 is specifically configured to obtain clinical information of the object to be detected output by the inspection information system; the clinical information comprises the age, sex, sample type to be tested and historical diagnosis record of the object to be tested.
Optionally, the second obtaining module 312 is configured to obtain detection result data of the object to be detected obtained through a single instrument, joint inspection, and/or a blood analysis pipeline; the detection result data comprises at least one of the following: blood cell analysis results, biochemical analysis results, immunoassay results, hemagglutination results, and blood cell microscopic examination results.
Optionally, the decision module 313 is further configured to input an auxiliary diagnostic knowledge-graph based on the report parameters in the biological sample analysis data, the identification result and the clinical information; and the auxiliary diagnosis knowledge graph determines abnormal characteristics of the biological sample of the object to be detected according to the report parameters, the identification result and the clinical information, and outputs auxiliary diagnosis decision information corresponding to the abnormal characteristics.
Optionally, the system further comprises a construction module, a detection module and a detection module, wherein the construction module is used for forming an entity set in a knowledge base according to the identification result of the target feature carried by the biological sample analysis data, the clinical information of the detection object and the sample reference parameter; forming an attribute set and a relationship set in a knowledge base according to the identification result, the clinical information, the comparison result of the report parameter relative to the sample reference parameter and the mapping relation of the historical diagnosis example in the diagnosis example database; forming a knowledge spectrum triplet set according to the entity set, the attribute set and the relation set, and constructing an auxiliary diagnosis knowledge spectrum according to the knowledge spectrum triplet set.
Optionally, the decision module 313 is further configured to input an auxiliary diagnostic model based on the report parameters in the biological sample analysis data, the identification result and the clinical information; and the auxiliary diagnosis model determines the abnormal type of the sample of the object to be detected according to the report parameters, the identification result and the clinical information, and outputs auxiliary diagnosis decision information corresponding to the abnormal type.
Optionally, the decision module 313 is further configured to determine an abnormal characteristic of the sample to be detected of the object to be detected, correct a parameter alarm state formed based on the recognition result of the target cell characteristic according to the abnormal characteristic, and output corrected auxiliary diagnosis decision information corresponding to the abnormal characteristic; determining abnormal characteristics of a sample to be detected of the object to be detected, and outputting auxiliary diagnosis decision information containing failure cause analysis corresponding to the abnormal characteristics; determining abnormal characteristics of a sample to be detected of the object to be detected, determining supporting evidence and/or diagnosis and treatment advice corresponding to the abnormal characteristics, and outputting an auxiliary diagnosis report containing the supporting evidence and/or diagnosis and treatment advice.
Optionally, the decision module 313 is further configured to determine an abnormal feature of a sample to be detected of the object to be detected, determine a diagnosis and treatment suggestion corresponding to the abnormal feature, and output an auxiliary diagnosis report including a text description of the diagnosis and treatment suggestion; or determining the abnormal characteristics of the sample to be detected of the object to be detected, determining the supporting evidence corresponding to the abnormal characteristics, and outputting an auxiliary diagnosis report comprising the text description, the image data carrying the labels and the numerical description of the supporting evidence.
Optionally, the decision module 313 is further configured to obtain a selection instruction of the simplified diagnostic report or the detailed diagnostic report; outputting an auxiliary diagnosis report containing the text description of the abnormal cause according to the selection instruction of the simplified diagnosis report, or outputting an auxiliary diagnosis report containing the text description of the supporting evidence, the image data carrying the label and the numerical description according to the selection instruction of the detailed diagnosis report.
It should be noted that: in the process of implementing the auxiliary diagnostic method, the auxiliary diagnostic device provided in the above embodiment is only exemplified by the above-mentioned division of each program module, and in practical application, the above-mentioned process allocation may be performed by different program modules according to needs, i.e. the internal structure of the device may be divided into different program modules, so as to complete all or part of the above-mentioned method steps. In addition, the auxiliary diagnostic device and the auxiliary diagnostic method provided in the above embodiments belong to the same concept, and specific implementation processes thereof are detailed in the method embodiments, which are not described herein again.
In another aspect of the embodiments of the present application, referring to fig. 16, the auxiliary diagnostic apparatus further provides an auxiliary diagnostic apparatus, which includes a processor 211 and a memory 212, where a computer program executable by the processor is stored in the memory 212, and the auxiliary diagnostic method described in any embodiment of the present application is implemented when the computer program is executed by the processor, and the auxiliary diagnostic apparatus and the auxiliary diagnostic method provided in the foregoing embodiment can achieve the same technical effects, so that repetition is avoided and no further description is given here.
The embodiment of the application further provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor, implements the respective processes of the above-mentioned auxiliary diagnostic method embodiment, and can achieve the same technical effects, so that repetition is avoided, and no further description is given here. The computer readable storage medium is, for example, read-only memory (ROM), random Access Memory (RAM), magnetic disk or optical disk.
The foregoing is merely specific embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily think about changes or substitutions within the technical scope of the present application, and the changes and substitutions are intended to be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (17)
1. An auxiliary diagnostic method applied to an auxiliary diagnostic device, comprising:
acquiring biological sample analysis data of an object to be detected, extracting characteristics of the biological sample analysis data, and outputting a recognition result of target characteristics carried in the biological sample analysis data;
Acquiring clinical information of the object to be detected;
and determining abnormal characteristics of a sample to be detected of the object to be detected based on the biological sample analysis data, the identification result and the clinical information, and outputting auxiliary diagnosis decision information corresponding to the abnormal characteristics.
2. The auxiliary diagnostic method as set forth in claim 1, wherein the acquiring biological sample analysis data of an object to be detected, extracting features from the biological sample analysis data, and outputting a recognition result of target features carried in the biological sample analysis data, includes:
and acquiring biological sample analysis data of an object to be detected, extracting characteristics of a biological sample analysis image in the biological sample analysis data through an image processing model, and outputting a recognition result of a target object carried by the biological sample analysis image.
3. The auxiliary diagnostic method as set forth in claim 2, wherein the acquiring biological sample analysis data of the object to be detected, performing feature extraction on a biological sample analysis image in the biological sample analysis data by an image processing model, and outputting a recognition result of the target object carried by the biological sample analysis image, includes:
Acquiring biological sample analysis data of an object to be detected output by a biological sample analyzer;
extracting features of biological sample analysis images in the biological sample analysis data through an image processing model, and outputting recognition results of target features respectively carried by the biological sample analysis images; the biological sample analysis image includes at least one of: a blood cell histogram, a blood cell scatter plot, a pulse signal waveform plot, a pulse signal signature plot, a thromboelastography, a biochemical or immune response curve.
4. The auxiliary diagnostic method as set forth in claim 1, wherein the acquiring biological sample analysis data of an object to be detected, extracting features from the biological sample data, and outputting a recognition result of target features carried in the biological sample analysis data, includes:
acquiring biological sample analysis data of an object to be detected; the biological sample analysis data includes at least one of: a blood cell histogram, a blood cell scatter diagram, a pulse signal waveform diagram, a pulse signal feature diagram, a one-dimensional pulse signal vector, a pulse signal feature value vector, a multi-dimensional scatter data set, elasticity diagram data and immune response original data;
And extracting the characteristics of the biological sample analysis data through a deep learning model or a traditional algorithm model, and outputting the identification result of the target characteristics carried in the biological sample analysis data.
5. The aided diagnosis method of claim 4, wherein the feature extraction of the biological sample analysis data by a deep learning model or a conventional algorithm model, and outputting the recognition result of the target feature carried in the biological sample analysis data, comprises:
extracting features of the biological sample analysis data through a deep learning model based on image classification; or, extracting the characteristics of the biological sample analysis data through a traditional algorithm model constructed based on an image morphology algorithm, an image classification algorithm, a clustering algorithm or a threshold segmentation algorithm;
classifying according to the feature extraction result, and outputting the identification result of the target feature carried in the biological sample analysis data.
6. The aided diagnosis method of claim 1, wherein the acquiring clinical information of the object to be detected includes:
acquiring clinical information of the object to be detected, which is output by an inspection information system; the clinical information comprises the age, sex, sample type to be tested and historical diagnosis record of the object to be tested.
7. The aided diagnosis method of claim 6, wherein the acquiring clinical information of the object to be detected further comprises:
obtaining detection result data of the object to be detected, which is obtained through a single instrument, joint inspection and/or a blood analysis pipeline; the detection result data comprises at least one of the following: blood cell analysis results, biochemical analysis results, immunoassay results, hemagglutination results, and blood cell microscopic examination results.
8. The auxiliary diagnostic method according to claim 1, wherein the determining an abnormal feature of the sample to be tested of the object to be tested based on the biological sample analysis data, the identification result, and the clinical information, and outputting auxiliary diagnostic decision information corresponding to the abnormal feature, comprises:
inputting an auxiliary diagnosis knowledge graph based on the report parameters, the recognition result and the clinical information in the biological sample analysis data;
and the auxiliary diagnosis knowledge graph determines abnormal characteristics of the biological sample of the object to be detected according to the report parameters, the identification result and the clinical information, and outputs auxiliary diagnosis decision information corresponding to the abnormal characteristics.
9. The aided diagnosis method of claim 8, wherein before the report parameters, the identification result, and the clinical information in the analysis data based on the biological sample are input into an aided diagnosis knowledge graph, further comprising:
forming an entity set in a knowledge base according to the identification result of the target feature carried by the biological sample analysis data, the clinical information of the detection object and the sample reference parameters;
forming an attribute set and a relationship set in a knowledge base according to the identification result, the clinical information, the comparison result of the report parameter relative to the sample reference parameter and the mapping relation of the historical diagnosis example in the diagnosis example database;
forming a knowledge spectrum triplet set according to the entity set, the attribute set and the relation set, and constructing an auxiliary diagnosis knowledge spectrum according to the knowledge spectrum triplet set.
10. The auxiliary diagnostic method according to claim 1, wherein the determining an abnormal feature of the sample to be tested of the object to be tested based on the biological sample analysis data, the identification result, and the clinical information, and outputting auxiliary diagnostic decision information corresponding to the abnormal feature, comprises:
Inputting an auxiliary diagnostic model based on the report parameters in the biological sample analysis data, the identification result, and the clinical information;
and the auxiliary diagnosis model determines the abnormal type of the sample of the object to be detected according to the report parameters, the identification result and the clinical information, and outputs auxiliary diagnosis decision information corresponding to the abnormal type.
11. The aided diagnosis method of claim 1, wherein the determining an abnormal feature of a sample to be detected of the object to be detected, outputting aided diagnosis decision information corresponding to the abnormal feature, comprises one of:
determining abnormal characteristics of a sample to be detected of the object to be detected, correcting a parameter alarm state formed based on the identification result of the target characteristics according to the abnormal characteristics, and outputting corrected auxiliary diagnosis decision information corresponding to the abnormal characteristics;
determining abnormal characteristics of a sample to be detected of the object to be detected, and outputting auxiliary diagnosis decision information containing failure cause analysis corresponding to the abnormal characteristics;
determining abnormal characteristics of a sample to be detected of the object to be detected, determining supporting evidence and/or diagnosis and treatment advice corresponding to the abnormal characteristics, and outputting an auxiliary diagnosis report containing the supporting evidence and/or diagnosis and treatment advice.
12. The auxiliary diagnostic method according to claim 1, wherein the determining of the abnormal feature of the sample to be detected of the object to be detected, outputting auxiliary diagnostic decision information corresponding to the abnormal feature, comprises:
determining abnormal characteristics of a sample to be detected of the object to be detected, determining diagnosis and treatment suggestions corresponding to the abnormal characteristics, and outputting an auxiliary diagnosis report containing text descriptions of the diagnosis and treatment suggestions; or alternatively, the first and second heat exchangers may be,
and determining abnormal characteristics of a sample to be detected of the object to be detected, determining supporting evidence corresponding to the abnormal characteristics, and outputting an auxiliary diagnosis report comprising text description, image data carrying labels and numerical description of the supporting evidence.
13. The auxiliary diagnostic method as set forth in claim 12, wherein the determining an abnormality characteristic of the sample to be detected of the object to be detected, outputting an auxiliary diagnostic report containing a description corresponding to the abnormality characteristic, includes:
acquiring a selection instruction of a simplified diagnosis report or a detailed diagnosis report;
outputting an auxiliary diagnosis report containing the text description of the abnormal cause according to the selection instruction of the simplified diagnosis report, or outputting an auxiliary diagnosis report containing the text description of the supporting evidence, the image data carrying the label and the numerical description according to the selection instruction of the detailed diagnosis report.
14. An auxiliary diagnostic device, comprising:
the first acquisition module is used for acquiring biological sample analysis data of an object to be detected;
the processing module is used for extracting the characteristics of the biological sample analysis data and outputting the identification result of the target characteristics carried in the biological sample analysis data;
the second acquisition module is used for acquiring clinical information of the object to be detected;
and the auxiliary diagnosis decision module is used for determining abnormal characteristics of a sample to be detected of the object to be detected based on the biological sample analysis data, the identification result and the clinical information, and outputting auxiliary diagnosis decision information corresponding to the abnormal characteristics.
15. An auxiliary diagnostic device comprising a processor and a memory, wherein the memory stores a computer program executable by the processor, the computer program implementing the auxiliary diagnostic method of any one of claims 1 to 13 when executed by the processor.
16. An auxiliary diagnostic system comprising a biological sample analyzer, a test information system, and the auxiliary diagnostic apparatus according to claim 15, wherein the auxiliary diagnostic apparatus is configured to acquire biological sample analysis data of an object to be detected from the biological sample analyzer, and acquire clinical information of the object to be detected from the test information system.
17. A computer readable storage medium, characterized in that the computer readable storage medium has stored thereon a computer program which, when executed by a controller, implements the auxiliary diagnostic method according to any one of claims 1 to 13.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111245816.7A CN116030958A (en) | 2021-10-25 | 2021-10-25 | Auxiliary diagnosis method, device, equipment and system and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111245816.7A CN116030958A (en) | 2021-10-25 | 2021-10-25 | Auxiliary diagnosis method, device, equipment and system and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116030958A true CN116030958A (en) | 2023-04-28 |
Family
ID=86076454
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111245816.7A Pending CN116030958A (en) | 2021-10-25 | 2021-10-25 | Auxiliary diagnosis method, device, equipment and system and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116030958A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117787864A (en) * | 2023-12-27 | 2024-03-29 | 石家庄博瑞迪生物技术有限公司 | Animal tissue sampling tube traceability and warehousing management method |
| CN118039176A (en) * | 2024-03-05 | 2024-05-14 | 河北省儿童医院 | Cytokine detection data processing method, device, terminal and storage medium |
| CN118314379A (en) * | 2024-03-29 | 2024-07-09 | 深圳市心研医疗科技有限公司 | Scatter diagram classification device |
| CN118486450A (en) * | 2024-06-03 | 2024-08-13 | 杭州量磁科技有限公司 | Auxiliary lung cancer diagnosis method and system based on quantum measurement |
-
2021
- 2021-10-25 CN CN202111245816.7A patent/CN116030958A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117787864A (en) * | 2023-12-27 | 2024-03-29 | 石家庄博瑞迪生物技术有限公司 | Animal tissue sampling tube traceability and warehousing management method |
| CN118039176A (en) * | 2024-03-05 | 2024-05-14 | 河北省儿童医院 | Cytokine detection data processing method, device, terminal and storage medium |
| CN118314379A (en) * | 2024-03-29 | 2024-07-09 | 深圳市心研医疗科技有限公司 | Scatter diagram classification device |
| CN118486450A (en) * | 2024-06-03 | 2024-08-13 | 杭州量磁科技有限公司 | Auxiliary lung cancer diagnosis method and system based on quantum measurement |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101981446B (en) | For the method and system using support vector machine to analyze flow cytometry data | |
| CN116030958A (en) | Auxiliary diagnosis method, device, equipment and system and storage medium | |
| KR101106201B1 (en) | Clinical examination analyzing device, clinical examination analyzing method, and storing computer readable recording medium program for allowing computer to execute the method | |
| Bjornsson et al. | Associative image analysis: a method for automated quantification of 3D multi-parameter images of brain tissue | |
| US20240044904A1 (en) | System, method, and article for detecting abnormal cells using multi-dimensional analysis | |
| WO2023186051A1 (en) | Auxiliary diagnosis method and apparatus, and construction apparatus, analysis apparatus and related product | |
| CN116029960A (en) | Auxiliary diagnosis method, image processing model and storage medium | |
| CN116030957A (en) | Knowledge graph construction method, knowledge graph construction device, auxiliary diagnosis equipment and storage medium | |
| CN116046643B (en) | Auxiliary diagnostic information providing device and blood analysis system | |
| US20230066976A1 (en) | Systematic characterization of objects in a biological sample | |
| Souza et al. | Simple hemogram to support the decision-making of COVID-19 diagnosis using clusters analysis with self-organizing maps neural network | |
| Grochowski et al. | Machine learning system for automated blood smear analysis | |
| JP2023101619A (en) | Detecting and reporting subpopulation of neutrophil | |
| Prasad et al. | Deep U_ClusterNet: automatic deep clustering based segmentation and robust cell size determination in white blood cell | |
| CN117524456A (en) | Auxiliary diagnosis method and system | |
| CN116762011A (en) | Detection of probability of developing sepsis | |
| CN116936113A (en) | Auxiliary diagnosis method, device, equipment, server, product, system and medium | |
| D’Souza et al. | Prediction of Anemia Using Naïve-Bayes Classification Algorithm in Machine Learning | |
| CN116978539A (en) | Auxiliary diagnosis method and knowledge graph construction method, device, equipment and medium | |
| CN117405896A (en) | Auxiliary diagnosis method, device, equipment, system and medium | |
| Ayumba | Role of Metrics in Medical Image Analysis based on Unsupervised Machine Learning | |
| McRoy | CLASSIFICATION AND EXPLANATION OF IRON DEFICIENCY ANEMIA FROM COMPLETE BLOOD COUNT DATA USING MACHINE LEARNING | |
| Khurana et al. | Data Science in Healthcare-Current Challenges and Opportunities | |
| Kandasamy et al. | Optimized Deep Learning Networks for Accurate Identification of Cancer Cells in Bone Marrow | |
| Bhimavarapu et al. | Blood Cell Detection and Counting Using Deep Learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |