CN116385756B - Medical image recognition method and related device based on enhancement annotation and deep learning - Google Patents
Medical image recognition method and related device based on enhancement annotation and deep learning Download PDFInfo
- Publication number
- CN116385756B CN116385756B CN202211634149.6A CN202211634149A CN116385756B CN 116385756 B CN116385756 B CN 116385756B CN 202211634149 A CN202211634149 A CN 202211634149A CN 116385756 B CN116385756 B CN 116385756B
- Authority
- CN
- China
- Prior art keywords
- medical image
- target
- deep learning
- detected
- identification
- 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
- 238000000034 method Methods 0.000 title claims abstract description 128
- 238000013135 deep learning Methods 0.000 title claims abstract description 81
- 238000002372 labelling Methods 0.000 claims abstract description 68
- 238000012549 training Methods 0.000 claims abstract description 32
- 230000008569 process Effects 0.000 claims description 92
- 238000001514 detection method Methods 0.000 claims description 30
- 210000001951 dura mater Anatomy 0.000 claims description 30
- 238000004891 communication Methods 0.000 claims description 15
- 238000000605 extraction Methods 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 10
- 238000013136 deep learning model Methods 0.000 claims description 7
- 230000003014 reinforcing effect Effects 0.000 claims description 4
- 238000010586 diagram Methods 0.000 description 15
- 230000011218 segmentation Effects 0.000 description 13
- 238000002604 ultrasonography Methods 0.000 description 8
- 230000011164 ossification Effects 0.000 description 5
- 208000002193 Pain Diseases 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 4
- 208000008035 Back Pain Diseases 0.000 description 3
- 206010028836 Neck pain Diseases 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 201000010099 disease Diseases 0.000 description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 210000001519 tissue Anatomy 0.000 description 3
- 238000012795 verification Methods 0.000 description 3
- 230000003321 amplification Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000001684 chronic effect Effects 0.000 description 2
- 238000004140 cleaning Methods 0.000 description 2
- 238000004590 computer program Methods 0.000 description 2
- 238000013434 data augmentation Methods 0.000 description 2
- 238000013480 data collection Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000001965 increasing effect Effects 0.000 description 2
- 230000001788 irregular Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000003199 nucleic acid amplification method Methods 0.000 description 2
- 230000002285 radioactive effect Effects 0.000 description 2
- 208000011580 syndromic disease Diseases 0.000 description 2
- 208000008930 Low Back Pain Diseases 0.000 description 1
- 210000003484 anatomy Anatomy 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000010420 art technique Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 210000000988 bone and bone Anatomy 0.000 description 1
- 230000002308 calcification Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000015271 coagulation Effects 0.000 description 1
- 238000005345 coagulation Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011109 contamination Methods 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000011022 operating instruction Methods 0.000 description 1
- 208000027753 pain disease Diseases 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 229940126585 therapeutic drug Drugs 0.000 description 1
- 238000002560 therapeutic procedure Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- 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
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- 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
- G06N3/08—Learning methods
-
- 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
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
-
- 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
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a medical image recognition method and a related device based on enhancement annotation and deep learning, wherein the medical image recognition method based on the enhancement annotation and the deep learning comprises the following steps: performing enhanced labeling on a target part in the medical image to obtain a labeling data set; wherein the target site includes a first site as a final target and a second site as an assist; training to obtain a deep learning recognition model of the target part based on the labeling data set; and inputting the medical image to be detected into the target part deep learning recognition model to obtain a recognition result of the medical image to be detected. According to the invention, the target part is enhanced and marked through the first part serving as a final target and the second part serving as an auxiliary target, and the clinical target area is searched through establishing the target part deep learning identification model, so that the accuracy of medical image identification and clinical targets is improved.
Description
Technical Field
The invention relates to the field of medical image recognition, in particular to a medical image recognition method and device based on enhanced annotation and deep learning, computing equipment and a computer storage medium.
Background
The global disease burden published in The journal of medicine "Lancet" (The Lancet for short) 2013 shows that pain diseases rank at position 1 in The global disease burden, wherein chronic lumbago and cervicodynia are respectively ranked at positions 1 and 4. 70% -80% of people suffer chronic neck and low back pain in life with 30% incidence at point. The most common causes of chronic spinal-derived cervical and lumbar pain include radicular kappaphasia pain, lumbar postspinal post-branch kappaphasia syndrome, arthrocele-transverse process pain syndrome, etc. caused by various causes.
Most patients in treatment will be willing to get pain relief in the early stages of the onset of the disease and possibly through minimally invasive interventional therapy means such as nerve block, radio frequency thermal coagulation, etc. In the past, the above treatment is mostly performed by a blind operation or under the guidance of X-rays, however, the blind operation often has poor treatment effect due to insufficient accuracy of the operation, and the X-rays (such as the schematic diagrams of positioning the articular process joint by X-rays shown in fig. 9a to 9 b) are more accurate to position, but require repeated radioactive irradiation, and often put in the threat of radioactive contamination to human health. In recent years, the application of the visual ultrasonic technology (such as the ultrasonic method for positioning the articular process joint and the transverse process shown in fig. 10) can enable a clinician to clearly identify the target structure and the surrounding tissues thereof, design the puncture path in advance, guide the operation in real time and observe the diffusion of the therapeutic drug, thereby improving the accuracy and success rate of treatment and reducing complications. However, a large number of anesthesiologists and pain surgeons are not very familiar with ultrasound images, resulting in long learning curves and severe operator dependence.
The traditional ultrasonic image recognition based on artificial intelligence generally adopts supervised machine learning, a plurality of target images are marked by experienced doctors, a proper model is selected for training, and means such as migration learning are assisted to improve average Accuracy (AP). Because the spinal articular process joints and transverse processes are often embodied as irregular thin strips on an ultrasonic image, when a directly marked and trained model is used for verification and test, a large number of false positives or false negatives appear, and the AP value is low, so that the model is difficult to be used as a proper traction target.
Disclosure of Invention
In view of the foregoing, the present invention has been made in order to provide a medical image recognition method and apparatus, a computing device, and a computer storage medium based on enhancement labeling and deep learning, which overcome the problem of low accuracy in medical image recognition.
According to one aspect of the present invention, there is provided a medical image recognition method based on enhancement tagging and deep learning, comprising:
performing enhanced labeling on a target part in the medical image to obtain a labeling data set; wherein the target site includes a first site as a final target and a second site as an assist;
training to obtain a deep learning recognition model of the target part based on the labeling data set;
And inputting the medical image to be detected into the target part deep learning recognition model to obtain a recognition result of the medical image to be detected.
In an optional manner, the enhancement labeling is performed on the target part in the medical image to obtain a labeling data set; wherein the target site includes a first site as a final target and a second site as an assist further includes:
performing enhancement labeling on the lumbar vertebrae ossification mark part in the medical image to obtain a labeling data set; wherein the lumbar vertebrae landmark region includes a first region as a final target and a second region as an assist.
In an optional mode, the lumbar vertebrae bone mark part in the medical image is enhanced and marked to obtain a marked data set; wherein the lumbar vertebrae landmark portion includes a first portion as a final target and a second portion as an assist further includes:
reinforcing and labeling the lumbar articular process joint-transverse process position in the medical image to obtain a labeling data set; wherein the lumbar articular-lateral aspect comprises a lumbar articular-lateral aspect as a final target and a dura mater as an adjunct.
In an alternative, the method further comprises:
and inputting the medical image area of the identification result of the medical image to be detected into a target point searching functional unit to obtain a clinical target point.
In an optional manner, the inputting the medical image to be detected into the target site deep learning recognition model, and obtaining the recognition result of the medical image to be detected further includes:
inputting the medical image to be detected into the target part deep learning identification model to obtain a first part identification result and a second part identification result of the medical image to be detected;
and identifying the medical image to be detected by taking the first part identification result and the second part identification result as a joint metric.
In an optional manner, the identifying the medical image to be detected using the first location identification result and the second location identification result as a joint metric further includes:
judging whether the medical image to be detected comprises the first part identification result and the second part identification result, if yes, judging whether the medical image to be detected comprises the target part.
In an optional manner, training to obtain the deep learning recognition model of the target part based on the labeling data set further includes:
inputting the annotation data set into a mask RCNN model, and training to obtain the target part deep learning recognition model;
the mask RCNN model comprises an input layer, a feature extraction network layer, an RPN region generation network layer, an ROI alignment network layer and at least one fully-connected network layer; wherein, the feature extraction network layer includes: a res net50 residual network layer and a FPN feature network layer, the fully connected network layer comprising: a mask layer, a category layer, and/or a coordinate layer.
In an alternative, the method further comprises:
acquiring a searching inflection point of the identification coordinates according to the identification coordinates of the lumbar articular process joint-transverse process part and the dura mater part;
adding the searching inflection point to the identification coordinate to obtain a target identification coordinate;
taking the target identification coordinates as target detection results;
the acquiring the search inflection point of the identification coordinate further includes:
performing curve fitting on the identification coordinates to obtain a fitted curve function;
calculating each inflection point of the fitting curve function to obtain the searching inflection point;
Or directly calculate the coordinate inflection point.
According to another aspect of the present invention, there is provided a medical image recognition apparatus based on enhanced annotation and deep learning, comprising:
the data labeling module is used for carrying out enhancement labeling on the target part in the medical image to obtain a labeling data set; wherein the target site includes a first site as a final target and a second site as an assist;
the model training module is used for training to obtain a target part deep learning recognition model based on the labeling data set;
the identification module is used for inputting the medical image to be detected into the target part deep learning identification model to obtain the identification result of the medical image to be detected.
According to yet another aspect of the present invention, there is provided a computing device comprising: the device comprises a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface complete communication with each other through the communication bus;
the memory is used for storing at least one executable instruction, and the executable instruction enables the processor to execute the operation corresponding to the medical image identification method based on the enhancement annotation and the deep learning.
According to still another aspect of the present invention, there is provided a computer storage medium having stored therein at least one executable instruction for causing a processor to perform operations corresponding to the medical image recognition method based on enhancement tagging and deep learning as described above.
According to the scheme provided by the invention, the target part in the medical image is enhanced and marked to obtain a marked data set; wherein the target site includes a first site as a final target and a second site as an assist; training to obtain a deep learning recognition model of the target part based on the labeling data set; and inputting the medical image to be detected into the target part deep learning recognition model to obtain a recognition result of the medical image to be detected. According to the invention, the target part is enhanced and marked through the first part serving as a final target and the second part serving as an auxiliary target, and the accuracy of medical image recognition is improved by establishing the target part deep learning recognition model, so that the accuracy of clinical targets is also improved.
The foregoing description is only an overview of the present invention, and is intended to be implemented in accordance with the teachings of the present invention in order that the same may be more clearly understood and to make the same and other objects, features and advantages of the present invention more readily apparent.
Drawings
Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to designate like parts throughout the figures. In the drawings:
FIG. 1 shows a flow diagram of a medical image recognition method based on enhanced annotation and deep learning according to an embodiment of the invention;
FIG. 2 shows a flow diagram of a medical image recognition method based on enhanced annotation and deep learning according to another embodiment of the invention;
FIG. 3 illustrates a schematic diagram of a maskRCNN ultrasound image segmentation process in accordance with an embodiment of the present invention;
FIG. 4 illustrates a schematic view of a target site of an embodiment of the present invention;
fig. 5a to 5b show the recognition result and original schematic diagrams according to the embodiment of the present invention;
FIGS. 6 a-6 c show raw ultrasound image schematics of an embodiment of the present invention;
FIGS. 7 a-7 b are schematic diagrams of data annotation according to embodiments of the present invention;
FIG. 8 is a schematic diagram showing the target detection result according to an embodiment of the present invention;
FIGS. 9 a-9 b illustrate a prior art technique for locating the articular process joint-transverse process using X-rays;
FIG. 10 is a schematic diagram showing the prior art positioning of the articular process joint-transverse process using ultrasound;
FIG. 11 shows a schematic structural diagram of a medical image recognition device based on enhanced annotation and deep learning according to an embodiment of the present invention;
FIG. 12 illustrates a schematic diagram of a computing device in accordance with an embodiment of the invention.
Detailed Description
Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
FIG. 1 shows a flow diagram of a medical image recognition method based on enhanced annotation and deep learning according to an embodiment of the invention. The method carries out enhancement labeling on the target part based on the first part serving as a final target and the second part serving as an auxiliary target, and identifies the medical image by establishing a target part deep learning identification model. Specifically, as shown in fig. 1, the method comprises the following steps:
Step S101, performing enhancement labeling on a target part in a medical image to obtain a labeling data set; wherein the target site includes a first site as a final target and a second site as an assist.
In this embodiment, the target site is enhanced and labeled according to the anatomical features of the target site, so that the joint measurement of clinician experience is more satisfied, the detection rate of the target site can be improved, the probability of false positive and false negative is reduced, and a proper measurement index is provided as the judgment basis of final detection, so that the actual clinical operation can be guided.
In an optional mode, performing enhancement labeling on a target part in the medical image to obtain a labeling data set; wherein the target site includes a first site as a final target and a second site as an assist further includes:
performing enhancement labeling on the lumbar vertebrae ossification mark part in the medical image to obtain a labeling data set; wherein the lumbar vertebrae landmark region includes a first region as a final target and a second region as an assist.
Step S102, training to obtain a target part deep learning recognition model based on the labeling data set.
The labeling data set is input to the target site deep learning recognition model for training, optionally, before the labeling data set is input to the target site deep learning recognition model, data preprocessing is performed on the labeling data set, for example, data preparation works such as scaling, data augmentation, data enhancement and the like are performed on the sampled data.
Step S103, inputting the medical image to be detected into a target part deep learning recognition model to obtain a recognition result of the medical image to be detected.
And inputting the medical image to be detected into the target part deep learning recognition model to obtain a recognition result of the medical image to be detected.
In this embodiment, because the detection target portion F is often represented as an irregular thin strip on the ultrasound image, the AP value of the detection target portion F is directly adopted as a measure, which will cause serious deviation between the segmentation presentation result and the actual experience, and the true value cannot be represented correctly. By improving the detection metric, in addition to the AP value of the detection target site F, the joint detection and segmentation accuracy of the first site as the final target and the second site as the assist is used as a reference standard, so that the target detection result is consistent with the experience of the doctor. Based on the same model or algorithm, the second part (such as the dura mater E) serving as an auxiliary is adopted to assist the first part (such as the articular process joint-transverse process F) serving as a final target to detect and divide, and compared with the training of only marking the detection target part F, the method has the beneficial effects (such as improvement of the accuracy by about 10 percent and reduction of the false positive rate by about 20 percent).
In an alternative mode, the image area of the identification result of the medical image to be detected is input into a target point searching functional unit to obtain a clinical target point.
In an optional manner, inputting the medical image to be detected into the target site deep learning recognition model, and obtaining the recognition result of the medical image to be detected further includes:
inputting the medical image to be detected into a target part deep learning identification model to obtain a first part identification result and a second part identification result of the medical image to be detected;
and identifying the medical image to be detected by taking the first part identification result and the second part identification result as joint metrics.
In an alternative way, identifying the medical image to be detected using the first and second region identification results as a joint metric further comprises:
judging whether the medical image to be detected comprises a first part identification result and a second part identification result, if so, judging that the medical image to be detected comprises a target part.
In an alternative manner, the identification result of the medical image to be detected further includes:
the segmentation result, the recognition coordinates and/or the category of the first part as the final target and the second part as the auxiliary.
In an alternative manner, the identification result of the medical image to be detected further includes:
the segmentation results, recognition coordinates and/or categories of the lumbar articular process-transverse process site as a final target and the dura mater site as an adjunct.
According to the scheme provided by the embodiment of the invention, the target part in the medical image is enhanced and marked to obtain the marked data set; wherein the target site includes a first site as a final target and a second site as an assist; training to obtain a deep learning recognition model of the target part based on the labeling data set; and inputting the medical image to be detected into the target part deep learning recognition model to obtain a recognition result of the medical image to be detected. According to the invention, the target part is enhanced and marked through the first part serving as a final target and the second part serving as an auxiliary target, and the accuracy of medical image recognition is improved by establishing the target part deep learning recognition model, so that the accuracy of clinical targets is also improved.
FIG. 2 shows a flow diagram of a medical image recognition method based on enhanced annotation and deep learning according to another embodiment of the invention. The method inputs the labeling data set into a mask RCNN model or other deep learning models for training to obtain a target position deep learning identification model, and further judges whether a target detection result is a true lumbar articular process joint-transverse process. Specifically, as shown in fig. 2, the method comprises the following steps:
Step S201, enhancing and labeling the lumbar articular process joint-transverse process position in the medical image to obtain a labeling data set; among them, the lumbar articular-lateral protrusion site includes a lumbar articular-lateral protrusion site as a final target and a dura mater site as an auxiliary.
Specifically, a target part in a medical image (such as an ultrasonic image and the like) is enhanced and marked to obtain a marked data set; among them, the lumbar articular-lateral protrusion site includes a lumbar articular-lateral protrusion site as a final target and a dura mater site as an auxiliary.
As shown in fig. 4, in addition to labeling the articular process joint-transverse process (in the upper square area, denoted as F) as a final target, a dura mater (in the lower square area, denoted as E) as an auxiliary is added and labeled according to anatomical features and clinical experience, and the target dura mater E is labeled in addition to labeling the articular process joint-transverse process F site.
Optionally, according to the characteristics of the articular process and the transverse process in the anatomy, the method is often represented as symmetrical echo areas on the ultrasonic image, the target part in the medical image is labeled on one side and two sides, and the labeled data set is amplified by horizontally overturning and rotating the target part in the medical image, so that the beneficial effect (such as improving the detection accuracy by about 8%) can be achieved.
And S202, inputting the labeling data set into a mask RCNN model, and training to obtain a target part deep learning recognition model.
MaskRCNN is an instance segmentation (Instance segmentation) algorithm, which is a multi-tasking network that can be used for tasks such as object detection, object instance segmentation, object keypoint detection, etc.
In this embodiment, the target portion deep learning recognition model is obtained through training based on the maskrnn model or other deep learning models.
In an alternative manner, training to obtain the target site deep learning recognition model based on the labeling dataset further includes:
inputting the annotation data set into a mask RCNN model or other deep learning models, and training to obtain a target part deep learning recognition model;
the mask RCNN model comprises an input layer, a feature extraction network layer, an RPN region generation network layer, an ROI alignment network layer and at least one fully-connected network layer; wherein, the feature extraction network layer includes: the ResNet50 residual network layer and the FPN characteristic network layer, the full connection network layer comprises: a mask layer, a category layer, and/or a coordinate layer.
As shown in fig. 3, the FPN (Feature Pyramid Networks, i.e., feature pyramid network) is a backbone network of the maskrnn model for extracting features of an image.
The RPN (Region Proposal Network, i.e. the region generation network) is used to find the region of interest, and the RPN can be understood as two tasks, one is a classification task and the other is a regression task, wherein the regression task is to give coordinates of the candidate frame (coordinates of the top left and bottom right of the candidate frame), and the classification task is to determine whether there is a target in the candidate frame (probability of having a target). When both tasks are completed, the candidate box for which the probability score for an object is greater than the threshold (e.g., 0.7) is retained as proposals (i.e., an area that is likely to contain a target).
The ROI alignment is mainly used for converting all the Proposed ROIs generated in the RPN process into feature maps with the same size, and then converting the feature maps into one-dimensional vectors.
The Mask rcnn model obtains the segmentation result (Mask), the recognition result (coordinate) and the category of the image through 3 independent fully-connected networks respectively.
Optionally, the annotation dataset is subjected to data collection, data cleansing, data annotation, data augmentation and data classification prior to input into the MaskRCNN model.
Data collection refers to the acquisition of an original ultrasound image, which requires a consistent viewing angle and an approximate field of view, as shown in fig. 6 a-6 c.
Because the data sample volume is huge and is usually manufactured manually, the correctness of the data is ensured by data cleaning, and further, the smooth implementation of the subsequent steps is ensured, and the data cleaning process mainly comprises the steps of: whether the image and the marked data are damaged, whether the data naming format meets the requirements, whether the marking is correct, whether the original image and the marked image are correctly associated, etc.
For data labeling of an ultrasonic image, a professional ultrasonic doctor is usually required to label the part needing to be identified and segmented, and labelme and other software can be adopted for labeling. As shown in fig. 7a to 7b, although it is generally considered that giving more features reduces the false detection rate of the target and can improve the recognition accuracy of the target, in this embodiment, the experimental result shows that the larger the labeling range is, the better, and the requirements of medical scenes and computer vision algorithms are met.
The data amplification is carried out by horizontally overturning and rotating the ultrasonic image, so that the detection accuracy can be improved.
Data classification is primarily used to divide data into training and validation sets required for deep learning. Based on the specificity of the object, the object is divided into a negative set, namely an image set without any object in the image, besides a traditional training set and a verification set.
Step S203, inputting a medical image to be detected into a target position deep learning recognition model to obtain a lumbar articular process joint-transverse process recognition result and a dura mater recognition result; and inputting the medical image area of the identification result into a target point searching functional unit to obtain a clinical target point.
The medical image to be detected is input into a target position deep learning recognition model, a lumbar articular process-transverse process recognition result and a dura mater recognition result are obtained as shown in fig. 8, and fig. 5a to 5b show recognition results and original diagrams.
And inputting the medical image area of the identification result into a target point searching functional unit to obtain a clinical target point.
Step S204, acquiring a searching inflection point of the identification coordinates according to the identification coordinates of the lumbar articular process joint-transverse process part and the dura mater part; adding the searching inflection point to the identification coordinate to obtain a target identification coordinate; and taking the target identification coordinates as target detection results.
The segmentation of the ultrasonic image is not a final target, and the doctor focuses on the puncture accuracy of the target part, and based on the segmentation result of machine learning, in this example, the search inflection point of the segmented part (such as a binary image) is added (such as the search inflection point is calculated through an inflection point calculation formula), so that the positioning error of the articular process joint-transverse process and the transverse process can be controlled within a usable range (typical value is 5 mm) as a final detection result of the target, and the puncture accuracy can be effectively improved.
Specifically, after a target detection result is obtained, acquiring a searching inflection point of the identification coordinates according to the identification coordinates of the lumbar articular process joint-transverse process part and the dura mater part; adding the searching inflection point to the identification coordinate to obtain a target identification coordinate; and taking the target identification coordinates as target detection results.
In an alternative manner, acquiring the search inflection point of the identification coordinates further includes:
performing curve fitting on the identification coordinates to obtain a fitted curve function;
calculating each inflection point of the fitted curve function to obtain a search inflection point;
or directly search or calculate the inflection point on the identified coordinates.
Step S205, it is determined whether the target detection result includes both the lumbar articular process-transverse process portion and the dura mater portion.
The dura mater E can assist in the evidence that the articular process joint-transverse process F is a true articular process joint-transverse process, which may be a true articular process joint-transverse process or a false articular process joint-transverse process if the articular process joint-transverse process F is present but the dura mater E is not present; if both the articular process joint-transverse process F and the dura mater E are present, the articular process joint-transverse process F must be the true articular process joint-transverse process.
The dura mater E is an intra-spinal structure, and if the dura mater E appears in the image, it means that the ultrasound section position is necessarily at the level of the intervertebral space, and the articular process joint-transverse process F is also at the level of the intervertebral space, so that the articular process joint-transverse process F appearing at the level where the dura mater E can appear is necessarily the articular process joint-transverse process. However, the articular process joint-transverse process F occurs without the dura mater E, and may be a structural tissue in which the dura mater E cannot be recognized under ultrasound due to ossification and calcification of the intervertebral space tissue, and may be in other portions in a morphology similar to that of the articular process joint-transverse process F.
In an optional manner, inputting the medical image to be detected into the target part deep learning recognition model to obtain a first part recognition result and a second part recognition result of the medical image to be detected;
and identifying the medical image to be detected by taking the first part identification result and the second part identification result as a joint metric.
In an optional manner, whether the medical image to be detected includes the first part identification result and the second part identification result at the same time is judged, if yes, the medical image to be detected includes the target part is judged.
Specifically, it is determined whether the target detection result includes both the lumbar articular process-transverse process portion and the dura mater portion, and if so, step S206 is performed.
Step S206, judging the lumbar articular process-transverse process.
The lumbar articular process-transverse process was judged.
According to the scheme provided by the embodiment of the invention, the single-side and double-side labeling and the data amplification are carried out on the target part in the medical image, so that the detection accuracy is further improved. And a target part deep learning recognition model is obtained through mask RCNN model training, so that the detection accuracy and the segmentation performance are improved. The accuracy of the target detection result is further increased by the auxiliary verification of the fact that the articular process joint-transverse process is the true articular process joint-transverse process or not through the dura mater. And based on the machine learning segmentation result, controlling the positioning errors of the articular process joint-transverse process and transverse process within a usable range by increasing the search inflection point of the segmentation part as a target final detection result.
FIG. 11 shows a schematic structural diagram of a medical image recognition device based on enhanced annotation and deep learning according to an embodiment of the invention. The medical image recognition device based on the enhancement annotation and the deep learning comprises: a data annotation module 1110, a model training module 1120, and an identification module 1130.
The data labeling module 1110 is configured to perform enhancement labeling on a target location in a medical image to obtain a labeled dataset; wherein the target site includes a first site as a final target and a second site as an assist;
the model training module 1120 is configured to train to obtain a deep learning recognition model of the target part based on the labeling data set;
the recognition module 1130 is configured to input a medical image to be detected to the target site deep learning recognition model, so as to obtain a recognition result of the medical image to be detected.
In an alternative manner, the data annotation module 1110 is further configured to:
performing enhancement labeling on the lumbar vertebrae ossification mark part in the medical image to obtain a labeling data set; wherein the lumbar vertebrae landmark region includes a first region as a final target and a second region as an assist.
In an alternative manner, the data annotation module 1110 is further configured to:
Reinforcing and labeling the lumbar articular process joint-transverse process position in the medical image to obtain a labeling data set; wherein the lumbar articular-lateral aspect comprises a lumbar articular-lateral aspect as a final target and a dura mater as an adjunct.
In an alternative manner, the identification module 1130 is further configured to:
and inputting the medical image area of the identification result of the medical image to be detected into a target point searching functional unit to obtain a clinical target point.
In an alternative manner, the identification module 1130 is further configured to:
inputting the medical image to be detected into the target part deep learning identification model to obtain a first part identification result and a second part identification result of the medical image to be detected;
and identifying the medical image to be detected by taking the first part identification result and the second part identification result as a joint metric.
In an alternative manner, the identification module 1130 is further configured to:
judging whether the medical image to be detected comprises the first part identification result and the second part identification result at the same time, if yes, judging that the medical image to be detected comprises the target part.
In an alternative manner, the model training module 1120 is further configured to:
inputting the annotation data set into a mask RCNN model, and training to obtain the target part deep learning recognition model;
the mask RCNN model comprises an input layer, a feature extraction network layer, an RPN region generation network layer, an ROI alignment network layer and at least one fully-connected network layer; wherein, the feature extraction network layer includes: a res net50 residual network layer and a FPN feature network layer, the fully connected network layer comprising: a mask layer, a category layer, and/or a coordinate layer.
In an alternative approach, the model training module 1130 is further configured to:
acquiring a searching inflection point of the identification coordinates according to the identification coordinates of the lumbar articular process joint-transverse process part and the dura mater part;
adding the searching inflection point to the identification coordinate to obtain a target identification coordinate;
taking the target identification coordinates as target detection results;
the acquiring the search inflection point of the identification coordinate further includes:
performing curve fitting on the identification coordinates to obtain a fitted curve function;
calculating each inflection point of the fitting curve function to obtain the searching inflection point;
Or directly calculate the coordinate inflection point.
According to the scheme provided by the embodiment of the invention, the target part in the medical image is enhanced and marked to obtain the marked data set; wherein the target site includes a first site as a final target and a second site as an assist; training to obtain a deep learning recognition model of the target part based on the labeling data set; and inputting the medical image to be detected into the target part deep learning recognition model to obtain a recognition result of the medical image to be detected. According to the invention, the target part is enhanced and marked through the first part serving as a final target and the second part serving as an auxiliary target, and the accuracy of medical image recognition is improved by establishing the target part deep learning recognition model, so that the accuracy of clinical targets is also improved.
FIG. 12 illustrates a schematic diagram of an embodiment of a computing device of the present invention, and the embodiments of the present invention are not limited to a particular implementation of the computing device.
As shown in fig. 12, the computing device may include: a processor 1202, a communication interface 1204, a memory 1206, and a communication bus 1208.
Wherein: the processor 1202, the communication interface 1204, and the memory 1206 communicate with each other via a communication bus 1208. A communication interface 1204 for communicating with network elements of other devices, such as clients or other servers, etc. The processor 1202 is configured to execute the program 1210 and may specifically perform relevant steps in the above-described medical image identification method embodiment based on enhancement tagging and deep learning.
In particular, program 1210 may include program code including computer operating instructions.
The processor 1202 may be a central processing unit CPU, or a graphics processor GPU, or an application specific integrated circuit ASIC, or one or more integrated circuits configured to implement embodiments of the present invention. The one or more processors included by the computing device may be the same type of processor, such as one or more CPU/GPUs; but may also be different types of processors such as one or more CPUs/GPUs and one or more ASICs.
Memory 1206 for storing program 1210. The memory 1206 may comprise high-speed RAM memory, and may also comprise non-volatile memory, such as at least one disk memory.
The program 1210 may be used, inter alia, to cause the processor 1202 to:
performing enhanced labeling on a target part in the medical image to obtain a labeling data set; wherein the target site includes a first site as a final target and a second site as an assist;
training to obtain a deep learning recognition model of the target part based on the labeling data set;
and inputting the medical image to be detected into the target part deep learning recognition model to obtain a recognition result of the medical image to be detected.
In an alternative, the program 1210 causes the processor to:
performing enhancement labeling on the lumbar vertebrae ossification mark part in the medical image to obtain a labeling data set; wherein the lumbar vertebrae landmark region includes a first region as a final target and a second region as an assist.
In an alternative, the program 1210 causes the processor to:
reinforcing and labeling the lumbar articular process joint-transverse process position in the medical image to obtain a labeling data set; wherein the lumbar articular-lateral aspect comprises a lumbar articular-lateral aspect as a final target and a dura mater as an adjunct.
In an alternative, the program 1210 causes the processor to:
and inputting the medical image area of the identification result of the medical image to be detected into a target point searching functional unit to obtain a clinical target point.
In an alternative, the program 1210 causes the processor to:
inputting the medical image to be detected into the target part deep learning identification model to obtain a first part identification result and a second part identification result of the medical image to be detected;
And identifying the medical image to be detected by taking the first part identification result and the second part identification result as a joint metric.
In an alternative, the program 1210 causes the processor to:
judging whether the medical image to be detected comprises the first part identification result and the second part identification result at the same time, if yes, judging that the medical image to be detected comprises the target part.
In an alternative, the program 1210 causes the processor to:
inputting the annotation data set into a mask RCNN model, and training to obtain the target part deep learning recognition model;
the mask RCNN model comprises an input layer, a feature extraction network layer, an RPN region generation network layer, an ROI alignment network layer and at least one fully-connected network layer; wherein, the feature extraction network layer includes: a res net50 residual network layer and a FPN feature network layer, the fully connected network layer comprising: a mask layer, a category layer, and/or a coordinate layer.
In an alternative, the program 1210 causes the processor to:
acquiring a searching inflection point of the identification coordinates according to the identification coordinates of the lumbar articular process joint-transverse process part and the dura mater part;
Adding the searching inflection point to the identification coordinate to obtain a target identification coordinate;
taking the target identification coordinates as target detection results;
the acquiring the search inflection point of the identification coordinate further includes:
performing curve fitting on the identification coordinates to obtain a fitted curve function;
calculating each inflection point of the fitting curve function to obtain the searching inflection point;
or directly calculate the coordinate inflection point.
According to the scheme provided by the embodiment of the invention, the target part in the medical image is enhanced and marked to obtain the marked data set; wherein the target site includes a first site as a final target and a second site as an assist; training to obtain a deep learning recognition model of the target part based on the labeling data set; and inputting the medical image to be detected into the target part deep learning recognition model to obtain a recognition result of the medical image to be detected. According to the invention, the target part is enhanced and marked through the first part serving as a final target and the second part serving as an auxiliary target, and the accuracy of medical image recognition is improved by establishing the target part deep learning recognition model, so that the accuracy of clinical targets is also improved.
Embodiments of the present invention provide a non-volatile computer storage medium storing at least one executable instruction that may perform the medical image recognition method based on enhanced annotation and deep learning in any of the above method embodiments.
The algorithms or displays presented herein are not inherently related to any particular computer, virtual system, or other apparatus. Various general-purpose systems may also be used with the teachings herein. The required structure for a construction of such a system is apparent from the description above. In addition, embodiments of the present invention are not directed to any particular programming language. It will be appreciated that the teachings of the present invention described herein may be implemented in a variety of programming languages, and the above description of specific languages is provided for disclosure of enablement and best mode of the present invention.
In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.
Similarly, it should be appreciated that in the above description of exemplary embodiments of the invention, various features of the embodiments of the invention are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure and aiding in the understanding of one or more of the various inventive aspects. However, the disclosed method should not be construed as reflecting the intention that: i.e., the claimed invention requires more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of this invention.
Those skilled in the art will appreciate that the modules in the apparatus of the embodiments may be adaptively changed and disposed in one or more apparatuses different from the embodiments. The modules or units or components of the embodiments may be combined into one module or unit or component and, furthermore, they may be divided into a plurality of sub-modules or sub-units or sub-components. Any combination of all features disclosed in this specification (including any accompanying claims, abstract and drawings), and all of the processes or units of any method or apparatus so disclosed, may be used in combination, except insofar as at least some of such features and/or processes or units are mutually exclusive. Each feature disclosed in this specification (including any accompanying claims, abstract and drawings), may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.
Furthermore, those skilled in the art will appreciate that while some embodiments herein include some features but not others included in other embodiments, combinations of features of different embodiments are meant to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments can be used in any combination.
Various component embodiments of the invention may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. Those skilled in the art will appreciate that some or all of the functionality of some or all of the components according to embodiments of the present invention may be implemented in practice using microprocessors or CPU/GPUs or Digital Signal Processors (DSPs). The present invention can also be implemented as an apparatus or device program (e.g., a computer program and a computer program product) for performing a portion or all of the methods described herein. Such a program embodying the present invention may be stored on a computer readable medium, or may have the form of one or more signals. Such signals may be downloaded from an internet website, provided on a carrier signal, or provided in any other form.
It should be noted that the above-mentioned embodiments illustrate rather than limit the invention, and that those skilled in the art will be able to design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention may be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by one and the same item of hardware. The use of the words first, second, third, etc. do not denote any order. These words may be interpreted as names. The steps in the above embodiments should not be construed as limiting the order of execution unless specifically stated.
Claims (7)
1. A medical image recognition method based on enhancement labeling and deep learning, comprising:
reinforcing and labeling the lumbar articular process joint-transverse process position in the medical image to obtain a labeling data set; wherein the lumbar articular-lateral aspect comprises a lumbar articular-lateral aspect as a final target and a dura mater as an adjunct;
Inputting the labeling data set into a deep learning model, and training to obtain a target part deep learning recognition model; the deep learning model comprises an input layer, a feature extraction network layer, a region generation network layer, an ROI alignment network layer and at least one fully connected network layer; wherein, the feature extraction network layer includes: a residual network layer and a feature network layer, the fully connected network layer comprising: a mask layer, a category layer, and/or a coordinate layer;
and inputting the medical image to be detected into the target part deep learning recognition model to obtain a recognition result of the medical image to be detected.
2. The enhanced annotation and deep learning based medical image recognition method of claim 1, further comprising:
and inputting the medical image area of the identification result of the medical image to be detected into a target point searching functional unit to obtain a clinical target point.
3. The medical image recognition method based on enhancement labeling and deep learning according to any one of claims 1-2, wherein the inputting the medical image to be detected into the target site deep learning recognition model, obtaining the recognition result of the medical image to be detected further comprises:
Inputting the medical image to be detected into the target part deep learning identification model to obtain a first part identification result and a second part identification result of the medical image to be detected;
and identifying the medical image to be detected by taking the first part identification result and the second part identification result as a joint metric.
4. The medical image recognition method based on enhanced labeling and deep learning of claim 3, wherein the recognizing the medical image to be detected using the first and second region recognition results as a joint metric further comprises:
judging whether the medical image to be detected comprises the first part identification result and the second part identification result at the same time, if yes, judging that the medical image to be detected comprises a target part.
5. The enhanced annotation and deep learning based medical image recognition method of claim 1, further comprising:
acquiring a searching inflection point of the identification coordinates according to the identification coordinates of the lumbar articular process joint-transverse process part and the dura mater part;
adding the searching inflection point to the identification coordinate to obtain a target identification coordinate;
Taking the target identification coordinates as target detection results;
the acquiring the search inflection point of the identification coordinate further includes:
performing curve fitting on the identification coordinates to obtain a fitted curve function;
calculating each inflection point of the fitting curve function to obtain the searching inflection point;
or directly calculate the coordinate inflection point.
6. A medical image recognition device based on enhanced annotation and deep learning, comprising:
the data labeling module is used for carrying out enhancement labeling on the lumbar articular process joint-transverse projection position in the medical image to obtain a labeling data set; wherein the lumbar articular-lateral aspect comprises a lumbar articular-lateral aspect as a final target and a dura mater as an adjunct;
the model training module is used for inputting the labeling data set into a deep learning model and training to obtain a target part deep learning recognition model; the deep learning model comprises an input layer, a feature extraction network layer, a region generation network layer, an ROI alignment network layer and at least one fully connected network layer; wherein, the feature extraction network layer includes: a residual network layer and a feature network layer, the fully connected network layer comprising: a mask layer, a category layer, and/or a coordinate layer;
The identification module is used for inputting the medical image to be detected into the target part deep learning identification model to obtain the identification result of the medical image to be detected.
7. A computing device, comprising: the device comprises a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface complete communication with each other through the communication bus;
the memory is configured to store at least one executable instruction that causes the processor to perform operations corresponding to the enhanced annotation and deep learning based medical image recognition method according to any one of claims 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211634149.6A CN116385756B (en) | 2022-12-19 | 2022-12-19 | Medical image recognition method and related device based on enhancement annotation and deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211634149.6A CN116385756B (en) | 2022-12-19 | 2022-12-19 | Medical image recognition method and related device based on enhancement annotation and deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116385756A CN116385756A (en) | 2023-07-04 |
| CN116385756B true CN116385756B (en) | 2023-10-17 |
Family
ID=86962094
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211634149.6A Active CN116385756B (en) | 2022-12-19 | 2022-12-19 | Medical image recognition method and related device based on enhancement annotation and deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116385756B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3553743A2 (en) * | 2018-04-11 | 2019-10-16 | Shanghai United Imaging Healthcare Co., Ltd. | Systems and methods for image processing |
| CN112699869A (en) * | 2020-12-17 | 2021-04-23 | 深圳视见医疗科技有限公司 | Rib fracture auxiliary detection method based on deep learning and image identification method |
| CN113537408A (en) * | 2021-09-08 | 2021-10-22 | 深圳开立生物医疗科技股份有限公司 | Ultrasonic image processing method, device and equipment and storage medium |
| CN114693719A (en) * | 2022-03-30 | 2022-07-01 | 南京航空航天大学 | Spine image segmentation method and system based on 3D-SE-Vnet |
| CN115393314A (en) * | 2022-08-23 | 2022-11-25 | 北京雅德嘉企业管理有限公司 | Deep learning-based oral medical image identification method and system |
-
2022
- 2022-12-19 CN CN202211634149.6A patent/CN116385756B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3553743A2 (en) * | 2018-04-11 | 2019-10-16 | Shanghai United Imaging Healthcare Co., Ltd. | Systems and methods for image processing |
| CN112699869A (en) * | 2020-12-17 | 2021-04-23 | 深圳视见医疗科技有限公司 | Rib fracture auxiliary detection method based on deep learning and image identification method |
| CN113537408A (en) * | 2021-09-08 | 2021-10-22 | 深圳开立生物医疗科技股份有限公司 | Ultrasonic image processing method, device and equipment and storage medium |
| CN114693719A (en) * | 2022-03-30 | 2022-07-01 | 南京航空航天大学 | Spine image segmentation method and system based on 3D-SE-Vnet |
| CN115393314A (en) * | 2022-08-23 | 2022-11-25 | 北京雅德嘉企业管理有限公司 | Deep learning-based oral medical image identification method and system |
Non-Patent Citations (3)
| Title |
|---|
| 基于深度学习的2D/3D医学图像配准研究;陈向前;郭小青;周钢;樊瑜波;王豫;;中国生物医学工程学报(04);全文 * |
| 基于深度学习的医疗影像识别技术研究综述;张琦;张荣梅;陈彬;;河北省科学院学报(03);全文 * |
| 深度迁移学习辅助的阿尔兹海默氏症早期诊断;金祝新;秦飞巍;方美娥;;计算机应用与软件(05);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116385756A (en) | 2023-07-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US8150132B2 (en) | Image analysis apparatus, image analysis method, and computer-readable recording medium storing image analysis program | |
| CN105513077A (en) | System for screening diabetic retinopathy | |
| CN110796636A (en) | CT image bone condition detection method and device based on convolutional neural network | |
| EP3722996A2 (en) | Systems and methods for processing 3d anatomical volumes based on localization of 2d slices thereof | |
| CN111462049A (en) | Automatic lesion area form labeling method in mammary gland ultrasonic radiography video | |
| CN109360213A (en) | A kind of automation centrum recognition methods based on backbone ultrasound coronal image | |
| Lin et al. | Using deep learning in ultrasound imaging of bicipital peritendinous effusion to grade inflammation severity | |
| CN112927213B (en) | Medical image segmentation method, medium and electronic device | |
| Fatima et al. | Vertebrae localization and spine segmentation on radiographic images for feature‐based curvature classification for scoliosis | |
| CN116309647B (en) | Method for constructing craniocerebral lesion image segmentation model, image segmentation method and device | |
| CN117409002A (en) | Visual identification detection system for wounds and detection method thereof | |
| CN116385756B (en) | Medical image recognition method and related device based on enhancement annotation and deep learning | |
| Radlak et al. | Automatic detection of bones based on the confidence map for rheumatoid arthritis analysis | |
| Zhang et al. | Automatic rib segmentation in chest CT volume data | |
| Yan et al. | Segmentation of pulmonary parenchyma from pulmonary CT based on ResU-Net++ model | |
| Shankar et al. | Leveraging clinically relevant biometric constraints to supervise a deep learning model for the accurate caliper placement to obtain sonographic measurements of the fetal brain | |
| CN114332858A (en) | Focus detection method and device and focus detection model acquisition method | |
| CN113808105A (en) | Focus detection method based on ultrasonic scanning | |
| CN112085698A (en) | Method and device for automatically analyzing left and right breast ultrasonic images | |
| JP5403431B2 (en) | Tomographic image processing method and apparatus | |
| Ebrahimi et al. | Automatic Segmentation and Identification of Spinous Processes on Sagittal X-Rays Based on Random Forest Classification and Dedicated Contextual Features | |
| Jain et al. | A novel strategy for automatic localization of cephalometric landmarks | |
| Zair et al. | An automated segmentation of coronary artery calcification using deep learning in specific region limitation | |
| Pandey | Real-time ultrasound bone segmentation and robust us-ct registration for surgical navigation of pelvic fractures | |
| Cai et al. | Deep‐learning based segmentation of ultrasound adipose image for liposuction |
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 |