CN116486403B - Cell discrimination model construction method, device, electronic equipment and storage medium - Google Patents
Cell discrimination model construction method, device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN116486403B CN116486403B CN202310729210.3A CN202310729210A CN116486403B CN 116486403 B CN116486403 B CN 116486403B CN 202310729210 A CN202310729210 A CN 202310729210A CN 116486403 B CN116486403 B CN 116486403B
- Authority
- CN
- China
- Prior art keywords
- cell
- model
- sample
- signal point
- cells
- 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.)
- Active
Links
- 238000010276 construction Methods 0.000 title claims abstract description 25
- 238000003860 storage Methods 0.000 title claims abstract description 14
- 238000013145 classification model Methods 0.000 claims abstract description 110
- 238000010801 machine learning Methods 0.000 claims abstract description 66
- 238000002073 fluorescence micrograph Methods 0.000 claims abstract description 48
- 238000000034 method Methods 0.000 claims abstract description 47
- 238000012549 training Methods 0.000 claims abstract description 28
- 238000013528 artificial neural network Methods 0.000 claims abstract description 18
- 238000001514 detection method Methods 0.000 claims description 16
- 230000004927 fusion Effects 0.000 claims description 16
- 230000015572 biosynthetic process Effects 0.000 claims description 12
- 238000005070 sampling Methods 0.000 claims description 12
- 238000003786 synthesis reaction Methods 0.000 claims description 12
- 238000004590 computer program Methods 0.000 claims description 11
- 239000013598 vector Substances 0.000 claims description 8
- 230000011218 segmentation Effects 0.000 claims description 6
- 230000002194 synthesizing effect Effects 0.000 claims description 4
- 230000000877 morphologic effect Effects 0.000 abstract description 13
- 239000012535 impurity Substances 0.000 abstract description 12
- 239000013589 supplement Substances 0.000 abstract description 3
- 210000004027 cell Anatomy 0.000 description 603
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 description 12
- 210000003855 cell nucleus Anatomy 0.000 description 11
- 239000010931 gold Substances 0.000 description 11
- 229910052737 gold Inorganic materials 0.000 description 11
- 230000001413 cellular effect Effects 0.000 description 7
- 238000004891 communication Methods 0.000 description 5
- 238000007500 overflow downdraw method Methods 0.000 description 5
- FWBHETKCLVMNFS-UHFFFAOYSA-N 4',6-Diamino-2-phenylindol Chemical compound C1=CC(C(=N)N)=CC=C1C1=CC2=CC=C(C(N)=N)C=C2N1 FWBHETKCLVMNFS-UHFFFAOYSA-N 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000002509 fluorescent in situ hybridization Methods 0.000 description 3
- 238000001000 micrograph Methods 0.000 description 3
- 238000007637 random forest analysis Methods 0.000 description 3
- 230000002159 abnormal effect Effects 0.000 description 2
- 230000032823 cell division Effects 0.000 description 2
- 238000003066 decision tree Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 210000004940 nucleus Anatomy 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 239000002994 raw material Substances 0.000 description 2
- 239000008280 blood Substances 0.000 description 1
- 210000004369 blood Anatomy 0.000 description 1
- 210000000349 chromosome Anatomy 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000000799 fluorescence microscopy Methods 0.000 description 1
- 230000004907 flux Effects 0.000 description 1
- 238000007901 in situ hybridization Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/69—Microscopic objects, e.g. biological cells or cellular parts
- G06V20/698—Matching; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/771—Feature selection, e.g. selecting representative features from a multi-dimensional feature space
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10056—Microscopic image
- G06T2207/10061—Microscopic image from scanning electron microscope
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10064—Fluorescence image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30242—Counting objects in image
-
- 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
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Investigating Or Analysing Biological Materials (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a cell discrimination model construction method, a device, electronic equipment and a storage medium, wherein a cell classification model constructed based on a neural network is obtained by training based on a cell fluorescence image and a cell label of a sample cell, and a machine learning model is constructed outside the cell classification model so as to extract morphological and statistical characteristics of cells and signal points through characteristic engineering, and the cells are classified from other dimensions by utilizing the morphological and statistical characteristics so as to supplement the cell classification model; by fusing the machine learning model and the cell classification model, the cells can be classified and judged by utilizing different characteristics from different angles, so that the interference caused by background impurity points and impurities in the cell classification and judgment process is reduced, the accuracy of cell classification and judgment is improved, the number of manually interpreted cells can be effectively reduced while the recall rate is ensured, and the subsequent interpretation efficiency is improved.
Description
Technical Field
The present invention relates to the field of image recognition technologies, and in particular, to a method and apparatus for constructing a cell discrimination model, an electronic device, and a storage medium.
Background
In the fluorescent in situ hybridization (Fluorescence in situ hybridization, FISH) based cell detection process, one sample requires accurate detection of specific types of cells, such as circulating chromosome abnormal cells (Circulating genetically Abnormal Cell, CAC), from tens of thousands of cells in order to obtain good quality detection results. The criteria for CAC cells are determined based on the fluorescent expression of the cells in each channel. However, when the CACs cells are identified by utilizing the fluorescence signal gain of FISH, the hybridized fluorescence signal is actually background impurity points and impurities due to factors such as experimental environment and experimental raw materials, which can influence the discrimination result of the cells, so that the accuracy of the cell discrimination result is lower, the number of the detected specific type of discrimination cells is higher, and when the subsequent manual determination is carried out, the number of the discrimination cells needing the manual determination is higher, and the subsequent discrimination efficiency and flux are influenced.
Disclosure of Invention
The invention provides a cell discrimination model construction method, a cell discrimination model construction device, electronic equipment and a storage medium, which are used for solving the defect that in the prior art, a cell discrimination result is easily interfered by factors such as an experimental environment, experimental raw materials and the like, so that the accuracy of the cell discrimination result is lower.
The invention provides a cell discrimination model construction method, which comprises the following steps:
cell fluorescence images and cell labels of sample cells are based, and a cell classification model constructed based on a neural network is obtained through training;
determining cell characteristics and signal point characteristics of sample cells, taking classification probability of the sample cells output by the cell classification model as candidate characteristics, carrying out characteristic selection on the candidate characteristics to obtain selected characteristics, and constructing a machine learning model based on the selected characteristics to serve as a cell discrimination model;
or determining cell characteristics and signal point characteristics of the sample cells as candidate characteristics, performing characteristic selection on the candidate characteristics to obtain selected characteristics, constructing a machine learning model based on the selected characteristics, and fusing the cell classification model and the machine learning model to obtain a cell discrimination model.
According to the method for constructing the cell discrimination model provided by the invention, the cell fluorescence image and the cell label thereof based on the sample cells are trained to obtain the cell classification model constructed based on the neural network, and the method specifically comprises the following steps:
obtaining cell fluorescence images and cell labels of different types of real sample cells in a clinical sample;
Clustering the cell fluorescence images of the real sample cells of different types, and obtaining target cell clusters corresponding to the target types;
generating a cell synthesis image of the target type based on the cell fluorescence images of each real sample cell of the target type in the target cell cluster;
and training to obtain the cell classification model based on the cell fluorescence image and the cell label thereof and the cell synthesis image and the cell label thereof.
According to the method for constructing the cell discrimination model provided by the invention, the cell fluorescence image of the true sample cell is obtained based on the following steps:
acquiring contour information of each real sample cell in a cell image based on a cell segmentation model;
and cutting and synthesizing the region corresponding to the real sample cell in each channel fluorescent image based on the outline information of the real sample cell to obtain a cell fluorescent image of the real sample cell.
According to the method for constructing a cell discrimination model provided by the invention, the method for determining the cell characteristics and the signal point characteristics of the sample cells and the classification probability of the sample cells output by the cell classification model are taken as candidate characteristics, and specifically comprises the following steps:
Acquiring characteristic values of cell characteristics and signal point characteristics of real sample cells in a clinical sample;
determining the point of a vector formed by the cell characteristics and signal point characteristics of each real sample cell of the target type and the classification probability of the corresponding real sample cell output by the cell classification model in a characteristic space as a characteristic point corresponding to the corresponding real sample cell;
sampling in the feature space based on the feature points corresponding to each real sample cell of the target type to generate new feature points, and determining the feature values of the cell features and the signal point features of the synthesized sample cells and the classification probability of the corresponding synthesized sample cells output by the cell classification model based on the new feature points;
and determining the characteristic values of the cell characteristics and the signal point characteristics of the real sample cells and the synthesized sample cells, and the classification probability of the corresponding sample cells output by the cell classification model as the characteristic values of candidate characteristics.
According to the method for constructing the cell discrimination model provided by the invention, the cell characteristics and the signal point characteristics of the sample cells are determined as candidate characteristics, and the method specifically comprises the following steps:
obtaining cell images containing real sample cells of a target type from cell images of different clinical samples in different fields of view, and determining contour information of each real sample cell in the cell images containing the real sample cells of the target type;
Acquiring characteristic values of cell characteristics and signal point characteristics of the real sample cells of the target type based on contour information of the real sample cells of the target type;
randomly sampling contour information of real sample cells of non-target types based on contour information of each real sample cell in the cell image containing the real sample cells of target types; the quantity difference between the sampled non-target type real sample cells and the target type real sample cells meets the preset condition;
acquiring characteristic values of cell characteristics and signal point characteristics of the non-target type real sample cells based on contour information of the non-target type real sample cells;
and determining the characteristic values of the cell characteristics and the signal point characteristics of the real sample cells of the target type and the characteristic values of the cell characteristics and the signal point characteristics of the real sample cells of the non-target type as the characteristic values of the candidate characteristics.
According to the cell discrimination model construction method provided by the invention, the characteristic value of the cell characteristic of any real sample cell is determined from the cell image containing any real sample cell based on the contour information of any real sample cell;
The characteristic value of the signal point characteristic of any real sample cell is obtained by determining the signal point region of any real sample cell in each channel fluorescent image by detecting the signal point in the corresponding any real sample cell region in each channel fluorescent image by using a signal point detection model based on the contour information of any real sample cell.
According to the method for constructing the cell discrimination model provided by the invention, the cell classification model and the machine learning model are fused to obtain the cell discrimination model, and the method specifically comprises the following steps:
adding a fusion layer after the cell classification model and the machine learning model to obtain the cell discrimination model; the fusion layer is used for determining an average value or a maximum value between the classification probabilities of the cells to be classified, which are respectively output by the cell classification model and the machine learning model.
The invention also provides a cell discrimination model construction device, which comprises:
the cell classification model construction unit is used for training to obtain a cell classification model constructed based on the neural network based on the cell fluorescence image and the cell label of the sample cells;
the model fusion unit is used for determining cell characteristics and signal point characteristics of sample cells and classifying probability of the sample cells output by the cell classifying model as candidate characteristics, selecting the candidate characteristics to obtain selected characteristics, and constructing a machine learning model based on the selected characteristics to serve as a cell discrimination model; or, the method is used for determining the cell characteristics and the signal point characteristics of the sample cells as candidate characteristics, carrying out characteristic selection on the candidate characteristics to obtain selected characteristics, constructing a machine learning model based on the selected characteristics, and fusing the cell classification model and the machine learning model to obtain a cell discrimination model.
The invention also provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the cell discrimination model construction method according to any one of the above when executing the program.
The present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a cell discrimination model construction method as described in any one of the above.
The invention also provides a computer program product comprising a computer program which when executed by a processor implements a method of cell discrimination model construction as described in any one of the above.
According to the cell discrimination model construction method, the device, the electronic equipment and the storage medium, the cell classification model constructed based on the neural network is obtained through training based on the cell fluorescence image and the cell label of the sample cell, and a machine learning model is constructed outside the cell classification model so as to extract morphological and statistical characteristics of cells and signal points through characteristic engineering, and the cells are classified from other dimensions by utilizing the morphological and statistical characteristics so as to supplement the cell classification model; by fusing the machine learning model and the cell classification model, the cells can be classified and judged by utilizing different characteristics from different angles, so that the interference caused by background impurity points and impurities in the cell classification and judgment process is reduced, the accuracy of cell classification and judgment is improved, the number of manually interpreted cells can be effectively reduced while the recall rate is ensured, and the subsequent interpretation efficiency is improved.
Drawings
In order to more clearly illustrate the invention or the technical solutions of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are some embodiments of the invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a cell discrimination model construction method provided by the invention;
FIG. 2 is a schematic flow chart of the training method of the cell classification model provided by the invention;
FIG. 3 is a schematic flow chart of a sample construction method according to the present invention;
FIG. 4 is a second flow chart of the sample construction method according to the present invention;
FIG. 5 is a schematic diagram of a cell discrimination model construction apparatus according to the present invention;
fig. 6 is a schematic structural diagram of an electronic device provided by the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
FIG. 1 is a schematic flow chart of a method for constructing a cell discrimination model according to the present invention, as shown in FIG. 1, the method includes:
step 110, training to obtain a cell classification model constructed based on a neural network based on a cell fluorescence image of a sample cell and a cell label thereof;
step 120, determining cell characteristics and signal point characteristics of sample cells, and taking classification probability of the sample cells output by the cell classification model as candidate characteristics, performing characteristic selection on the candidate characteristics to obtain selected characteristics, and constructing a machine learning model based on the selected characteristics to serve as a cell discrimination model;
or, in step 130, determining the cell characteristics and the signal point characteristics of the sample cells as candidate characteristics, performing characteristic selection on the candidate characteristics to obtain selected characteristics, constructing a machine learning model based on the selected characteristics, and fusing the cell classification model and the machine learning model to obtain a cell discrimination model.
Specifically, microscopic scan data of a clinical sample (e.g., a blood sample) is collected, and initial cell classification data of the clinical sample is obtained through an AI model, wherein the current AI model comprises a cell segmentation model (e.g., a Mask-rcnn model) and a signal point detection model (e.g., a Yolo model), so that the number of signal points of each cell under each channel is obtained, and the cells are initially classified according to the number of signal points. The method comprises the steps that an interpretation person interprets interpretation cells in the initialization classification, confirms real target type cells (such as CAC cells) in a sample, obtains cell labels of the sample cells, performs data sampling and division according to the cell labels, and generates a sample set for constructing a subsequent cell classification model and a machine learning model. Wherein the sample set used to construct the cell classification model and the machine learning model are generated independently.
And training the cell classification model constructed based on the neural network based on the cell fluorescence image and the cell label (namely the label indicating the cell type of the sample cell) of each sample cell in the sample set corresponding to the cell classification model. The cell fluorescence image of the sample cell is a superposition synthetic image corresponding to the sample cell area in each channel fluorescence imaging. In some embodiments, the cell classification model may be implemented using a convolutional neural network, such as DenseNet-121. Taking DenseNet-121 as an example, in a cell classification model, data enhancement is performed on an input image first, including scaling and random cropping; then, processing the data enhancement result based on a large-scale convolution and a pooling layer; then, after processing based on several consecutive sub-modules (namely, a transform Block and a transform Layer), a pooling Layer and a full-connection Layer are utilized to output corresponding classification results.
In some embodiments, as shown in fig. 2, the training is performed to obtain a cell classification model constructed based on a neural network based on a cell fluorescence image of a sample cell and a cell label thereof, and specifically includes:
step 210, obtaining cell fluorescence images and cell labels of different types of real sample cells in a clinical sample;
Step 220, clustering the cell fluorescence images of the real sample cells of different types, and obtaining target cell clusters corresponding to the target types;
step 230, generating a cell synthesis image of the target type based on the cell fluorescence images of each real sample cell of the target type in the target cell cluster;
step 240, training to obtain the cell classification model based on the cell fluorescence image and the cell label thereof and the cell synthesis image and the cell label thereof.
Specifically, since there are only a few, a dozen or more cells of the target type (for example, CAC cells) in one clinical sample, but there are hundreds of cells of the non-target type, there is a serious problem of sample imbalance, and the problem of sample imbalance affects the training effect of the cell classification model, thereby reducing the classification accuracy in the practical application process. In order to solve the problem of sample imbalance in the cell classification model training process, a sample set of the cell classification model can be generated based on a KMeansSMOTE mode.
Here, a cell fluorescence image of different types of real sample cells in a clinical sample and their cell tags can be acquired. The cell fluorescence image of the real sample cells in the clinical sample is obtained by obtaining contour information of each real sample cell in the cell image based on a cell segmentation model, and then cutting and synthesizing regions corresponding to the real sample cells in each channel fluorescence image based on the contour information of the real sample cells. The cell image and channel fluorescence image are both microscope images under different channels acquired by the microscope, in some embodiments, the cell image may be a microscope image under the Dapi channel, while each channel fluorescence image is a microscope image under Red, green, aqua and Gold channels.
Based on pixel values of pixel points in the cell fluorescence images of the real sample cells of different types, clustering the cell fluorescence images of the real sample cells of different types to obtain a plurality of cell clusters, and obtaining target cell clusters corresponding to the target types from the cell clusters. According to the cell labels of the real sample cells in each cell cluster, the cell cluster with the largest number of the real sample cells of the target type can be used as the target cell cluster corresponding to the target type. Based on the cell fluorescence images of each real sample cell of the target type in the target cell cluster, a cell synthesis image of the target type is generated based on an oversampling technique. For example, a cell synthesis image of a target type may be obtained by fusion based on pixel values at the same position in a cell fluorescence image of a plurality of real sample cells of the target type in the target cell cluster, and a cell label may be set for the cell synthesis image. And constructing a sample set based on the cell fluorescence image and the cell label of each real sample cell and the cell synthesis image and the cell label generated in the steps, so that a sample set with balanced various types of samples can be obtained, a cell classification model is obtained based on training of the sample set, and the classification precision of the cell classification model is improved.
In order to further improve the accuracy of cell classification, a machine learning model is also constructed besides the cell classification model to extract morphological and statistical features of cells and signal points through feature engineering, and the cells are classified from other dimensions by utilizing the morphological and statistical features so as to supplement the cell classification model. By fusing the machine learning model and the cell classification model, the cells can be classified and judged by utilizing different characteristics from different angles, so that the interference caused by background impurity points and impurities in the cell classification and judgment process is reduced, and the accuracy of cell classification and judgment is improved.
Specifically, the fusion method of the machine learning model and the cell classification model comprises the following two modes:
(1) And determining cell characteristics and signal point characteristics of the sample cells, taking the classification probability of the sample cells output by the cell classification model as candidate characteristics, carrying out characteristic selection on the candidate characteristics, and constructing a machine learning model based on the selected characteristics to serve as a cell discrimination model after obtaining the selected characteristics.
Here, the cell characteristics, the signal point characteristics, and the classification probabilities of the corresponding sample cells output by the cell classification model are all candidate characteristics. I.e. the candidate features comprise cell features, signal point features and classification probabilities (i.e. probabilities that the respective cells belong to the respective types) output by the cell classification model. The cell characteristics of the sample cells include morphological characteristics and statistical characteristics of the cells, such as area, signal intensity, roundness, extensibility, and the like. Optional cellular features may include a variety of the following, and embodiments of the present invention are not specifically limited as to which features are specifically employed as cellular features: 'nucleic_pixels_area' (cell nucleus pixel area), 'dapi_mean_intensity' (DAPI average intensity), 'dapi_intensity_std' (DAPI intensity standard deviation), 'dapi_total_intensity' (DAPI total intensity), 'nucleic_round' (cell nucleus roundness), 'nucleic_orientation' (cell nucleus elongation), 'nucleic_average_diameter' (cell nucleus average diameter), "num_signals" (total number of signal points), "gr_signals_total_intensity" (total intensity of Green signal points), "gr_signals_total_area_pixels '(total area of Green signal points)," number_of_gr_signals' (number of Green signal points), "gr_signal_intensity_to_number_pixel_area '(intensity of Green signal points to area of cell nuclei),' gr_signal_pixel_area_to_pixel_area '(Green signal point pixel area to cell core pixel area),' gr_signal_intensity_to_nucleic_intensity '(Green signal point total area),' gr_signal_intensity_to_pixel_area '(Green signal point intensity to signal point pixel area),' back_intensity_of_gr_signal_pixel '(Green signal point background intensity),' r_signal_total_intensity '(Red signal point total area),' r_sign_area_pixel '(Red signal point number),' r_signal point_to_nucleic_pixel area), 'r_signal_pixel_area_to_nucleic_pixel_area' (Red signal dot pixel area to cell nucleus pixel area), 'r_signal_intensity_to_nucleic_intensity' (Red signal dot intensity to cell nucleus intensity), 'r_signal_intensity_to_signal_pixel_area' (Red signal dot intensity to signal dot pixel area), 'background_intensity_of r_signal dot' (Red signal dot background intensity), 'aq_signals_total_intension' (Aqua signal point total intensity), 'aq_signals_total_area_pixel' (Aqua signal point total area), 'number_of_aq_signals' (Aqua signal point number), 'aq_signal_intension_to_nucleic_pixel_area' (Aqua signal point intensity to nucleus pixel area), and 'aq_signal_pixel_area_to_nucleic_pixel_area' (Aqua signal point pixel area to nucleus pixel area), 'aq_signal_intensity_to_nucleic_intensity' (Aqua signal intensity to cell nucleus intensity), 'aq_signal_intensity_to_signal_pixel_area' (Aqua signal point intensity to signal point pixel area), 'background_intensity_of_aq_signals' (Aqua signal point background intensity), 'gd_signals_total_intensity' (Gold signal point total intensity), 'gd_signal_total_area_pixels' (Gold signal dot total area), 'number_of_gd_signal' (Gold signal dot number), 'gd_signal_intensity_to_nucleic_pixel_area' (Gold signal dot intensity to cell nucleus pixel area), 'gd_signal_pixel_area_to_nucleic_pixel_area' (Gold signal dot pixel area to cell nucleus pixel area), ' gd_signal_intensity_to_nucleic_intensity ' (Gold signal intensity to cell nucleus intensity), ' gd_signal_intensity_to_pixel_area ' (Gold signal dot intensity to signal dot pixel area), ' background_intensity_of_gd_signals ' (Gold signal dot background intensity), ' gr_255_area_pixels ' (Green pixel value is area of 255), ' r_255_area_pixels ' (Red pixel value is area of 255), ' aq_255_area_pixels ' (Aqua pixel value is area of 255), ' gd_255_area_pixels ' (Gold pixel value is area of 255), ' gr_255_pixels ' (gold_pixel value is area of 255), the "r_255_area_pixels_to_nucleic_area ' (Red pixel value 255 area to nuclear pixel area), the" aq_255_area_pixel_to_nucleic_area ' (Aqua pixel value 255 area to nuclear pixel area), the "gd_255_area_pixels_to_nucleic_area ' (Gold pixel value 255 area to nuclear pixel area), the" cell_score ' (cell division score output by cell division model), "nucleic_image ' (cell perimeter)," nucleic_aspect ratio), "nucleic_size_dimension ' (cell aspect ratio)," nucleic_solid_dimension), "nucleic_nucleic_pixel_area ' (cell major axis)," nucleic_nucleic_cell major axis ", and" nucleic_cell minor_center "(cell major axis.
The signal point characteristics of the sample cell include morphological characteristics and statistical characteristics of fluorescent signal points under each channel in the cell, for example, target detection scores of the fluorescent signal points under each channel (i.e. probability of each fluorescent signal point output by a signal point detection model), signal point intensity, signal point area, signal point perimeter and the like, including: 'angle' (signal point direction), 'aspect_ratio' (signal point aspect ratio), 'bbox_h' (signal point detection bounding box height), 'bbox_w' (signal point detection bounding box width), 'extent' (signal point range), 'intensity' (signal strength), 'intensity_std' (signal strength standard deviation), 'ma_l' (signal point long axis), 'ma_s' (signal point short axis), 'max_value' (signal point intensity maximum value), 'mean_intensity' (average signal strength), 'min_value' (signal point intensity minimum value), 'pixels_area' (signal point pixel area), 'pixel_diameter),' score '(signal point detection score) output by the' score '(signal point detection model,' solidity; in addition, the signal point features may also include distribution of the above-mentioned morphological features and statistical features (e.g., signal point intensity, signal point area, signal point perimeter, etc.) under each channel, such as minimum, maximum, median, average, variance.
It should be noted that, the characteristic value of the cellular feature of any real sample cell is determined from the cellular image containing the real sample cell based on the contour information of the real sample cell, and the characteristic value of part of the cellular feature needs to be determined by combining with the signal point detection model to perform signal point detection on the real sample cell to obtain a signal point region; the characteristic value of the signal point characteristic of any real sample cell is obtained by detecting the signal point corresponding to the real sample cell area in each channel fluorescent image by using a signal point detection model based on the contour information of the real sample cell, and determining the signal point area of the real sample cell in each channel fluorescent image.
And then, carrying out feature selection on the candidate features by utilizing feature engineering to obtain selected features. The feature engineering can select features based on feature values of various candidate features corresponding to various sample cells, so that the features with distinguishing property are selected as selection features. It can be seen that the selection features are part of the cell features, signal point features and classification probabilities. Wherein feature selection may be performed using an embedded feature selection method, such as a tree model, and in some embodiments, a random forest model may be used. After the selection feature is obtained, a machine learning model can be constructed as a cell discrimination model based on the selection feature. Specifically, a random forest model can be selected as a machine learning model, and model training is performed according to the characteristic values of the corresponding selected characteristics of each sample cell. Wherein, in order to prevent overfitting, each decision tree constructed in the random forest model is pre-pruned, i.e. the growth of the decision tree is stopped after the set maximum depth is reached. The machine learning model constructed in the mode is a cell discrimination model. Therefore, the fusion mode is to integrate the output result of the cell classification model in the construction process of the machine learning model.
Similar to the cell classification model, the problem of sample imbalance also exists in the construction process of the machine learning model, and the problem also causes the insufficient classification precision of the machine learning model. Therefore, to solve the problem of sample imbalance in the machine learning model construction process, a sample set of the machine learning model may be generated based on SMOTE mode.
In some embodiments, as shown in fig. 3, the determining the cell characteristics and the signal point characteristics of the sample cells, and the classification probability of the sample cells output by the cell classification model, as candidate characteristics, specifically includes:
step 310, obtaining characteristic values of cell characteristics and signal point characteristics of real sample cells in a clinical sample;
step 320, determining the feature values of the cell features and the signal point features of each real sample cell of the target type and the points of the vector formed by the classification probability of the corresponding real sample cell output by the cell classification model in the feature space as the feature points corresponding to the corresponding real sample cell;
step 330, sampling and generating new feature points in the feature space based on the feature points corresponding to each real sample cell of the target type, and determining the feature values of the cell features and the signal point features of the synthesized sample cells and the classification probability of the corresponding synthesized sample cells output by the cell classification model based on the new feature points;
Step 340, determining the characteristic values of the cell characteristics and the signal point characteristics of the real sample cells and the synthesized sample cells, and the classification probability of the corresponding sample cells output by the cell classification model as the characteristic values of the candidate characteristics.
Specifically, after the characteristic values of various cell characteristics and signal point characteristics corresponding to various real sample cells in the clinical sample are obtained, the vector formed by combining the characteristic values of various cell characteristics and signal point characteristics corresponding to various real sample cells and the classification probability of corresponding cells output by the cell classification model is constructed. And determining the points of vectors formed by the cell characteristics and the characteristic values of the signal point characteristics of each real sample cell of the target type and the classification probability of the corresponding cell output by the cell classification model in the characteristic space as the corresponding characteristic points of the corresponding real sample cell. Based on the feature points corresponding to each real sample cell of the target type, the over-sampling technology is utilized to sample and generate new feature points in the feature space. The method comprises the steps of selecting a characteristic point corresponding to a real sample cell of one target type at will, determining a characteristic point corresponding to a real sample cell of another target type closest to the characteristic point, and sampling between the two characteristic points to obtain a new characteristic point. And determining the characteristic values of the synthesized sample cells corresponding to various cell characteristics and signal point characteristics and the classification probability of the corresponding synthesized sample cells output by the cell classification model based on the vectors corresponding to the new characteristic points obtained by sampling. The characteristic values of the cell characteristics and the signal point characteristics of the real sample cells and the synthesized sample cells and the classification probability of the corresponding sample cells output by the cell classification model can be used as the characteristic values of each candidate characteristic for characteristic selection and training of a machine learning model.
(2) And determining cell characteristics and signal point characteristics of the sample cells as candidate characteristics, performing characteristic selection on the candidate characteristics to obtain selected characteristics, constructing a machine learning model based on the selected characteristics, and fusing the cell classification model and the machine learning model to obtain a cell discrimination model.
Here, the cell characteristics and signal point characteristics of the sample cells are taken as candidate characteristics. I.e. candidate features include cell features and signal point features. And then, carrying out feature selection on the candidate features by utilizing feature engineering to obtain selected features. The feature selection manner is similar to that in the fusion manner (1), and will not be described here again. After the selection feature is obtained, a machine learning model can be constructed based on the selection feature, and the manner of constructing the machine learning model based on the selection feature is similar to the fusion manner (1), and is not described in detail herein. After the machine learning model is obtained, the cell classification model and the machine learning model can be fused to obtain the cell discrimination model. In some embodiments, a fusion layer may be added after the cell classification model and the machine learning model to obtain a cell discrimination model. That is, the cell discrimination model includes a cell classification model, a machine learning model, and a fusion layer. The fusion layer is used for determining an average value or a maximum value between the classification probabilities of the cells to be classified, which are respectively output by the cell classification model and the machine learning model. Specifically, the cell classification model and the machine learning model respectively classify the cells to be classified to obtain classification probabilities of the cells to be classified, which are respectively output by the cell classification model and the machine learning model, and then the fusion layer obtains an average value or a maximum value between the two classification probabilities to be used as a classification result of the cells to be classified, which is output by the cell discrimination model.
In the fusion method (2), in order to solve the problem of sample imbalance in the machine learning model construction process, the method of generating a sample set of the machine learning model is different from the cell classification model and the fusion method (1).
Specifically, as shown in fig. 4, the determining the cell characteristics and the signal point characteristics of the sample cell as candidate characteristics specifically includes:
step 410, obtaining cell images containing real sample cells of a target type from cell images of different clinical samples in different fields of view, and determining contour information of each real sample cell in the cell images containing the real sample cells of the target type;
step 420, obtaining the characteristic values of the cell characteristics and the signal point characteristics of the real sample cells of the target type based on the contour information of the real sample cells of the target type;
step 430, randomly sampling contour information of the real sample cells of the non-target type based on the contour information of each real sample cell in the cell image containing the real sample cells of the target type; the quantity difference between the sampled non-target type real sample cells and the target type real sample cells meets the preset condition;
Step 440, obtaining the characteristic values of the cell characteristics and the signal point characteristics of the non-target type real sample cells based on the contour information of the non-target type real sample cells;
step 450, determining the characteristic values of the cell characteristics and the signal point characteristics of the real sample cells of the target type and the characteristic values of the cell characteristics and the signal point characteristics of the real sample cells of the non-target type as the characteristic values of the candidate characteristics.
Here, cell images including true sample cells of the target type are acquired from cell images of different clinical samples under different fields of view, and contour information of each true sample cell in the above-described cell images including true sample cells of the target type is determined based on the cell segmentation model. In one aspect, the feature values of the cell features and the signal point features of the true sample cells of the target type may be obtained in the manner provided in the above embodiments based on the profile information of the true sample cells of the target type. On the other hand, considering that the number of the non-target type real sample cells is far greater than that of the target type real sample cells, in order to make up the difference between the two numbers to obtain a more balanced sample set, the contour information of each real sample cell in the cell image containing the target type real sample cells can be randomly sampled based on the contour information of a plurality of non-target type real sample cells, and then the characteristic values of the cell characteristics and the signal point characteristics of the non-target type real sample cells are obtained based on the contour information of the non-target type real sample cells; wherein the difference in number between the sampled non-target type of real sample cells and the target type of real sample cells satisfies a preset condition, in some embodiments the sampled non-target type of real sample cells is the same as the target type of real sample cells. And determining the characteristic values of the cell characteristics and the signal point characteristics of the real sample cells of the target type and the characteristic values of the cell characteristics and the signal point characteristics of the real sample cells of the non-target type as the characteristic values of candidate characteristics for characteristic selection and training of a machine learning model.
It should be noted that, the above two fusion methods can achieve the effect of improving the accuracy of cell classification and discrimination, and the embodiment of the invention does not limit what fusion method is specifically adopted.
Therefore, according to the method provided by the embodiment of the invention, the cell classification model constructed based on the neural network is obtained by training based on the cell fluorescence image and the cell label of the sample cell, and besides the cell classification model, a machine learning model is also constructed, so that morphological and statistical characteristics of cells and signal points are extracted through characteristic engineering, and the cells are classified from other dimensions by utilizing the morphological and statistical characteristics, so that the cell classification model is supplemented; by fusing the machine learning model and the cell classification model, the cells can be classified and judged by utilizing different characteristics from different angles, so that the interference caused by background impurity points and impurities in the cell classification and judgment process is reduced, the accuracy of cell classification and judgment is improved, the number of manually interpreted cells can be effectively reduced while the recall rate is ensured, and the subsequent interpretation efficiency is improved.
The cell discrimination model constructing apparatus provided by the present invention will be described below, and the cell discrimination model constructing apparatus described below and the cell discrimination model constructing method described above can be referred to in correspondence with each other.
Based on any of the above embodiments, fig. 5 is a schematic structural diagram of a cell discrimination model building apparatus according to the present invention, as shown in fig. 5, the apparatus includes: a cell classification model construction unit 510 and a model fusion unit 520.
The cell classification model construction unit 510 is used for training to obtain a cell classification model constructed based on a neural network based on a cell fluorescence image of a sample cell and a cell label thereof;
the model fusion unit 520 is configured to determine a cell feature and a signal point feature of a sample cell, and a classification probability of the sample cell output by the cell classification model is used as a candidate feature, perform feature selection on the candidate feature, and construct a machine learning model based on the selected feature as a cell discrimination model after obtaining the selected feature; or, the method is used for determining the cell characteristics and the signal point characteristics of the sample cells as candidate characteristics, carrying out characteristic selection on the candidate characteristics to obtain selected characteristics, constructing a machine learning model based on the selected characteristics, and fusing the cell classification model and the machine learning model to obtain a cell discrimination model.
According to the device provided by the embodiment of the invention, based on the cell fluorescence image and the cell label of the sample cells, a cell classification model constructed based on a neural network is obtained through training, and besides the cell classification model, a machine learning model is constructed, so that morphological and statistical characteristics of cells and signal points are extracted through characteristic engineering, and the cells are classified from other dimensions by utilizing the morphological and statistical characteristics, so that the cell classification model is supplemented; by fusing the machine learning model and the cell classification model, the cells can be classified and judged by utilizing different characteristics from different angles, so that the interference caused by background impurity points and impurities in the cell classification and judgment process is reduced, the accuracy of cell classification and judgment is improved, the number of manually interpreted cells can be effectively reduced while the recall rate is ensured, and the subsequent interpretation efficiency is improved.
Based on any of the above embodiments, the training of the cell fluorescence image and the cell label thereof based on the sample cell to obtain the cell classification model constructed based on the neural network specifically includes:
obtaining cell fluorescence images and cell labels of different types of real sample cells in a clinical sample;
clustering the cell fluorescence images of the real sample cells of different types, and obtaining target cell clusters corresponding to the target types;
generating a cell synthesis image of the target type based on the cell fluorescence images of each real sample cell of the target type in the target cell cluster;
and training to obtain the cell classification model based on the cell fluorescence image and the cell label thereof and the cell synthesis image and the cell label thereof.
Based on any of the above embodiments, the cell fluorescence image of the real sample cell is obtained based on the following steps:
acquiring contour information of each real sample cell in a cell image based on a cell segmentation model;
and cutting and synthesizing the region corresponding to the real sample cell in each channel fluorescent image based on the outline information of the real sample cell to obtain a cell fluorescent image of the real sample cell.
Based on any one of the above embodiments, the determining the cell characteristics and the signal point characteristics of the sample cells, and the classification probability of the sample cells output by the cell classification model, as candidate characteristics, specifically includes:
acquiring characteristic values of cell characteristics and signal point characteristics of real sample cells in a clinical sample;
determining the point of a vector formed by the cell characteristics and signal point characteristics of each real sample cell of the target type and the classification probability of the corresponding real sample cell output by the cell classification model in a characteristic space as a characteristic point corresponding to the corresponding real sample cell;
sampling in the feature space based on the feature points corresponding to each real sample cell of the target type to generate new feature points, and determining the feature values of the cell features and the signal point features of the synthesized sample cells and the classification probability of the corresponding synthesized sample cells output by the cell classification model based on the new feature points;
and determining the characteristic values of the cell characteristics and the signal point characteristics of the real sample cells and the synthesized sample cells, and the classification probability of the corresponding sample cells output by the cell classification model as the characteristic values of candidate characteristics.
Based on any of the above embodiments, the determining the cell characteristic and the signal point characteristic of the sample cell as candidate characteristics specifically includes:
obtaining cell images containing real sample cells of a target type from cell images of different clinical samples in different fields of view, and determining contour information of each real sample cell in the cell images containing the real sample cells of the target type;
acquiring characteristic values of cell characteristics and signal point characteristics of the real sample cells of the target type based on contour information of the real sample cells of the target type;
randomly sampling contour information of real sample cells of non-target types based on contour information of each real sample cell in the cell image containing the real sample cells of target types; the quantity difference between the sampled non-target type real sample cells and the target type real sample cells meets the preset condition;
acquiring characteristic values of cell characteristics and signal point characteristics of the non-target type real sample cells based on contour information of the non-target type real sample cells;
and determining the characteristic values of the cell characteristics and the signal point characteristics of the real sample cells of the target type and the characteristic values of the cell characteristics and the signal point characteristics of the real sample cells of the non-target type as the characteristic values of the candidate characteristics.
Based on any of the above embodiments, the feature value of the cellular feature of any real sample cell is determined from the cellular image containing the any real sample cell based on the contour information of the any real sample cell;
the characteristic value of the signal point characteristic of any real sample cell is obtained by determining the signal point region of any real sample cell in each channel fluorescent image by detecting the signal point in the corresponding any real sample cell region in each channel fluorescent image by using a signal point detection model based on the contour information of any real sample cell.
Based on any one of the above embodiments, the fusing the cell classification model and the machine learning model to obtain a cell discrimination model specifically includes:
adding a fusion layer after the cell classification model and the machine learning model to obtain the cell discrimination model; the fusion layer is used for determining an average value or a maximum value between the classification probabilities of the cells to be classified, which are respectively output by the cell classification model and the machine learning model.
Fig. 6 is a schematic structural diagram of an electronic device according to the present invention, and as shown in fig. 6, the electronic device may include: processor 610, memory 620, communication interface (Communications Interface) 630, and communication bus 640, wherein processor 610, memory 620, and communication interface 630 communicate with each other via communication bus 640. Processor 610 may invoke logic instructions in memory 620 to perform a cell discrimination model building method comprising: cell fluorescence images and cell labels of sample cells are based, and a cell classification model constructed based on a neural network is obtained through training; determining cell characteristics and signal point characteristics of sample cells, taking classification probability of the sample cells output by the cell classification model as candidate characteristics, carrying out characteristic selection on the candidate characteristics to obtain selected characteristics, and constructing a machine learning model based on the selected characteristics to serve as a cell discrimination model; or determining cell characteristics and signal point characteristics of the sample cells as candidate characteristics, performing characteristic selection on the candidate characteristics to obtain selected characteristics, constructing a machine learning model based on the selected characteristics, and fusing the cell classification model and the machine learning model to obtain a cell discrimination model.
Further, the logic instructions in the memory 620 described above may be implemented in the form of software functional units and may be stored in a computer readable storage medium when sold or used as a stand alone product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
In another aspect, the present invention also provides a computer program product comprising a computer program stored on a non-transitory computer readable storage medium, the computer program comprising program instructions which, when executed by a computer, enable the computer to perform the method of constructing a cell discrimination model provided by the above methods, the method comprising: cell fluorescence images and cell labels of sample cells are based, and a cell classification model constructed based on a neural network is obtained through training; determining cell characteristics and signal point characteristics of sample cells, taking classification probability of the sample cells output by the cell classification model as candidate characteristics, carrying out characteristic selection on the candidate characteristics to obtain selected characteristics, and constructing a machine learning model based on the selected characteristics to serve as a cell discrimination model; or determining cell characteristics and signal point characteristics of the sample cells as candidate characteristics, performing characteristic selection on the candidate characteristics to obtain selected characteristics, constructing a machine learning model based on the selected characteristics, and fusing the cell classification model and the machine learning model to obtain a cell discrimination model.
In yet another aspect, the present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, is implemented to perform the above provided cell discrimination model construction methods, the method comprising: cell fluorescence images and cell labels of sample cells are based, and a cell classification model constructed based on a neural network is obtained through training; determining cell characteristics and signal point characteristics of sample cells, taking classification probability of the sample cells output by the cell classification model as candidate characteristics, carrying out characteristic selection on the candidate characteristics to obtain selected characteristics, and constructing a machine learning model based on the selected characteristics to serve as a cell discrimination model; or determining cell characteristics and signal point characteristics of the sample cells as candidate characteristics, performing characteristic selection on the candidate characteristics to obtain selected characteristics, constructing a machine learning model based on the selected characteristics, and fusing the cell classification model and the machine learning model to obtain a cell discrimination model.
The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (7)
1. A method of constructing a cell discrimination model, comprising:
cell fluorescence images and cell labels of sample cells are based, and a cell classification model constructed based on a neural network is obtained through training;
determining cell characteristics and signal point characteristics of sample cells, taking classification probability of the sample cells output by the cell classification model as candidate characteristics, carrying out characteristic selection on the candidate characteristics to obtain selected characteristics, and constructing a machine learning model based on the selected characteristics to serve as a cell discrimination model;
the determining the cell characteristics and the signal point characteristics of the sample cells and the classification probability of the sample cells output by the cell classification model as candidate characteristics specifically comprises:
acquiring characteristic values of cell characteristics and signal point characteristics of real sample cells in a clinical sample;
determining the point of a vector formed by the cell characteristics and signal point characteristics of each real sample cell of the target type and the classification probability of the corresponding real sample cell output by the cell classification model in a characteristic space as a characteristic point corresponding to the corresponding real sample cell;
sampling in the feature space based on the feature points corresponding to each real sample cell of the target type to generate new feature points, and determining the feature values of the cell features and the signal point features of the synthesized sample cells and the classification probability of the corresponding synthesized sample cells output by the cell classification model based on the new feature points;
And determining the characteristic values of the cell characteristics and the signal point characteristics of the real sample cells and the synthesized sample cells, and the classification probability of the corresponding sample cells output by the cell classification model as the characteristic values of candidate characteristics.
2. The method for constructing a cell discrimination model according to claim 1, wherein the training of the cell classification model constructed based on the neural network based on the cell fluorescence image of the sample cell and the cell label thereof specifically comprises:
obtaining cell fluorescence images and cell labels of different types of real sample cells in a clinical sample;
clustering the cell fluorescence images of the real sample cells of different types, and obtaining target cell clusters corresponding to the target types;
generating a cell synthesis image of the target type based on the cell fluorescence images of each real sample cell of the target type in the target cell cluster;
and training to obtain the cell classification model based on the cell fluorescence image and the cell label thereof and the cell synthesis image and the cell label thereof.
3. The method according to claim 2, wherein the cell fluorescence image of the real sample cell is obtained based on the steps of:
Acquiring contour information of each real sample cell in a cell image based on a cell segmentation model;
and cutting and synthesizing the region corresponding to the real sample cell in each channel fluorescent image based on the outline information of the real sample cell to obtain a cell fluorescent image of the real sample cell.
4. The method according to claim 1, wherein the feature value of the cell feature of any one of the real sample cells is determined from the cell image containing the any one of the real sample cells based on the contour information of the any one of the real sample cells;
the characteristic value of the signal point characteristic of any real sample cell is obtained by determining the signal point region of any real sample cell in each channel fluorescent image by detecting the signal point in the corresponding any real sample cell region in each channel fluorescent image by using a signal point detection model based on the contour information of any real sample cell.
5. A cell discrimination model constructing apparatus comprising:
the cell classification model construction unit is used for training to obtain a cell classification model constructed based on the neural network based on the cell fluorescence image and the cell label of the sample cells;
The model fusion unit is used for determining cell characteristics and signal point characteristics of sample cells and classifying probability of the sample cells output by the cell classifying model as candidate characteristics, selecting the candidate characteristics to obtain selected characteristics, and constructing a machine learning model based on the selected characteristics to serve as a cell discrimination model;
the determining the cell characteristics and the signal point characteristics of the sample cells and the classification probability of the sample cells output by the cell classification model as candidate characteristics specifically comprises:
acquiring characteristic values of cell characteristics and signal point characteristics of real sample cells in a clinical sample;
determining the point of a vector formed by the cell characteristics and signal point characteristics of each real sample cell of the target type and the classification probability of the corresponding real sample cell output by the cell classification model in a characteristic space as a characteristic point corresponding to the corresponding real sample cell;
sampling in the feature space based on the feature points corresponding to each real sample cell of the target type to generate new feature points, and determining the feature values of the cell features and the signal point features of the synthesized sample cells and the classification probability of the corresponding synthesized sample cells output by the cell classification model based on the new feature points;
And determining the characteristic values of the cell characteristics and the signal point characteristics of the real sample cells and the synthesized sample cells, and the classification probability of the corresponding sample cells output by the cell classification model as the characteristic values of candidate characteristics.
6. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the cell discrimination model building method of any one of claims 1 to 4 when the program is executed.
7. A non-transitory computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor implements the cell discrimination model building method according to any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310729210.3A CN116486403B (en) | 2023-06-20 | 2023-06-20 | Cell discrimination model construction method, device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310729210.3A CN116486403B (en) | 2023-06-20 | 2023-06-20 | Cell discrimination model construction method, device, electronic equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116486403A CN116486403A (en) | 2023-07-25 |
| CN116486403B true CN116486403B (en) | 2023-09-08 |
Family
ID=87221689
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310729210.3A Active CN116486403B (en) | 2023-06-20 | 2023-06-20 | Cell discrimination model construction method, device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116486403B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021259393A2 (en) * | 2021-01-08 | 2021-12-30 | 北京安德医智科技有限公司 | Image processing method and apparatus, and electronic device |
| CN114550169A (en) * | 2022-02-23 | 2022-05-27 | 腾讯科技（深圳）有限公司 | Training method, device, equipment and medium for cell classification model |
| CN115063796A (en) * | 2022-08-18 | 2022-09-16 | 珠海横琴圣澳云智科技有限公司 | Cell classification method and device based on signal point content constraint |
| WO2022211402A1 (en) * | 2021-03-30 | 2022-10-06 | 한국과학기술원 | Atherosclerotic plaque tissue analysis method and device using multi-modal fusion image |
| CN115497092A (en) * | 2022-07-29 | 2022-12-20 | 腾讯科技（深圳）有限公司 | Image processing method, device and equipment |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10055551B2 (en) * | 2013-10-10 | 2018-08-21 | Board Of Regents Of The University Of Texas System | Systems and methods for quantitative analysis of histopathology images using multiclassifier ensemble schemes |
| KR20190077061A (en) * | 2016-11-08 | 2019-07-02 | 셀룰러 리서치, 인크. | Cell labeling method |
-
2023
- 2023-06-20 CN CN202310729210.3A patent/CN116486403B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021259393A2 (en) * | 2021-01-08 | 2021-12-30 | 北京安德医智科技有限公司 | Image processing method and apparatus, and electronic device |
| WO2022211402A1 (en) * | 2021-03-30 | 2022-10-06 | 한국과학기술원 | Atherosclerotic plaque tissue analysis method and device using multi-modal fusion image |
| CN114550169A (en) * | 2022-02-23 | 2022-05-27 | 腾讯科技（深圳）有限公司 | Training method, device, equipment and medium for cell classification model |
| CN115497092A (en) * | 2022-07-29 | 2022-12-20 | 腾讯科技（深圳）有限公司 | Image processing method, device and equipment |
| CN115063796A (en) * | 2022-08-18 | 2022-09-16 | 珠海横琴圣澳云智科技有限公司 | Cell classification method and device based on signal point content constraint |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116486403A (en) | 2023-07-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP2302583B1 (en) | Image segmentation | |
| EP3306566B1 (en) | Method and system for processing image | |
| EP3785223B1 (en) | Systems and methods for segmentation and analysis of 3d images | |
| CN109492706B (en) | Chromosome classification prediction device based on recurrent neural network | |
| WO2012169088A1 (en) | Image processing apparatus, image processing method and image processing system | |
| CN110211108A (en) | A kind of novel abnormal cervical cells automatic identifying method based on Feulgen colouring method | |
| US10991098B1 (en) | Methods for automated chromosome analysis | |
| CN114463609B (en) | Method and device for determining fluorescent mixed points in cell microscopic image | |
| CN112036384B (en) | Sperm head shape recognition method, device and equipment | |
| CN114998332B (en) | Method and device for determining karyotype abnormal cells | |
| Buayai et al. | End-to-end automatic berry counting for table grape thinning | |
| Ram et al. | Segmentation and detection of fluorescent 3D spots | |
| CN107492084B (en) | Typical clustering cell nucleus image synthesis method based on randomness | |
| CN113537182B (en) | Automatic identification method and system for metaphase mitosis microscopic image of chromosome | |
| CN109117703A (en) | It is a kind of that cell category identification method is mixed based on fine granularity identification | |
| CN111553888B (en) | Titanium alloy forging microstructure image identification method based on machine learning | |
| CN108921172B (en) | Image processing device and method based on support vector machine | |
| CN104573701B (en) | A kind of automatic testing method of Tassel of Corn | |
| CN116665210B (en) | Cell classification method and device based on multichannel information fusion | |
| CN116486403B (en) | Cell discrimination model construction method, device, electronic equipment and storage medium | |
| Saha et al. | Segmentation of cervical nuclei using SLIC and pairwise regional contrast | |
| García et al. | Pollen grains contour analysis on verification approach | |
| CN113177602A (en) | Image classification method and device, electronic equipment and storage medium | |
| CN118379734B (en) | FISH microscopic image cell segmentation method | |
| EP4379676A1 (en) | Detection system, detection apparatus, learning apparatus, detection method, learning method and program |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |