CN116453114A - Pathological image analysis method, equipment and system based on deep learning - Google Patents
Pathological image analysis method, equipment and system based on deep learning Download PDFInfo
- Publication number
- CN116453114A CN116453114A CN202310280029.9A CN202310280029A CN116453114A CN 116453114 A CN116453114 A CN 116453114A CN 202310280029 A CN202310280029 A CN 202310280029A CN 116453114 A CN116453114 A CN 116453114A
- Authority
- CN
- China
- Prior art keywords
- cells
- image
- cell
- deep learning
- myoepithelial
- 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.)
- Granted
Links
- 238000013135 deep learning Methods 0.000 title claims abstract description 50
- 238000003703 image analysis method Methods 0.000 title claims abstract description 21
- 230000001575 pathological effect Effects 0.000 title claims abstract description 16
- 238000001514 detection method Methods 0.000 claims abstract description 75
- 230000003211 malignant effect Effects 0.000 claims abstract description 52
- 238000000034 method Methods 0.000 claims abstract description 52
- 238000010186 staining Methods 0.000 claims abstract description 52
- 201000011510 cancer Diseases 0.000 claims abstract description 41
- 230000007170 pathology Effects 0.000 claims abstract description 39
- 238000011532 immunohistochemical staining Methods 0.000 claims abstract description 35
- 238000010191 image analysis Methods 0.000 claims abstract description 19
- 210000004027 cell Anatomy 0.000 claims description 272
- 206010028980 Neoplasm Diseases 0.000 claims description 64
- 210000002919 epithelial cell Anatomy 0.000 claims description 37
- 230000004927 fusion Effects 0.000 claims description 34
- 230000000762 glandular Effects 0.000 claims description 33
- 208000003200 Adenoma Diseases 0.000 claims description 28
- 210000001703 glandular epithelial cell Anatomy 0.000 claims description 28
- 238000012549 training Methods 0.000 claims description 26
- 206010001233 Adenoma benign Diseases 0.000 claims description 21
- 230000002710 gonadal effect Effects 0.000 claims description 21
- 238000011176 pooling Methods 0.000 claims description 21
- 238000013136 deep learning model Methods 0.000 claims description 17
- 238000010276 construction Methods 0.000 claims description 14
- 238000004891 communication Methods 0.000 claims description 13
- 230000011218 segmentation Effects 0.000 claims description 12
- 210000000805 cytoplasm Anatomy 0.000 claims description 11
- 230000009467 reduction Effects 0.000 claims description 10
- 238000004422 calculation algorithm Methods 0.000 claims description 9
- 102000006783 calponin Human genes 0.000 claims description 9
- 108010086826 calponin Proteins 0.000 claims description 9
- 210000003855 cell nucleus Anatomy 0.000 claims description 9
- 238000004590 computer program Methods 0.000 claims description 9
- 208000009287 Myoepithelioma Diseases 0.000 claims description 8
- 239000002771 cell marker Substances 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 8
- 206010004453 Benign salivary gland neoplasm Diseases 0.000 claims description 7
- 108010008886 CAM 5.2 antigen Proteins 0.000 claims description 7
- 101000614627 Homo sapiens Keratin, type I cytoskeletal 13 Proteins 0.000 claims description 7
- 102100040487 Keratin, type I cytoskeletal 13 Human genes 0.000 claims description 7
- 208000021146 Warthin tumor Diseases 0.000 claims description 7
- 208000009956 adenocarcinoma Diseases 0.000 claims description 7
- 208000002517 adenoid cystic carcinoma Diseases 0.000 claims description 7
- 238000001914 filtration Methods 0.000 claims description 7
- 201000007416 salivary gland adenoid cystic carcinoma Diseases 0.000 claims description 7
- 208000006336 acinar cell carcinoma Diseases 0.000 claims description 6
- 201000001676 epithelial-myoepithelial carcinoma Diseases 0.000 claims description 6
- 238000011049 filling Methods 0.000 claims description 6
- 208000026226 myoepithelial tumor Diseases 0.000 claims description 6
- 206010041823 squamous cell carcinoma Diseases 0.000 claims description 6
- 230000002055 immunohistochemical effect Effects 0.000 claims description 5
- 208000025278 malignant myoepithelioma Diseases 0.000 claims description 4
- 201000008405 myoepithelial carcinoma Diseases 0.000 claims description 4
- 230000001086 cytosolic effect Effects 0.000 claims description 3
- 238000003066 decision tree Methods 0.000 claims description 3
- 238000003708 edge detection Methods 0.000 claims description 3
- 238000002156 mixing Methods 0.000 claims description 3
- 210000004940 nucleus Anatomy 0.000 claims description 3
- 238000007500 overflow downdraw method Methods 0.000 claims description 3
- 238000007637 random forest analysis Methods 0.000 claims description 3
- 238000012706 support-vector machine Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 2
- 238000004458 analytical method Methods 0.000 abstract description 10
- 239000000523 sample Substances 0.000 description 38
- 210000000981 epithelium Anatomy 0.000 description 25
- 238000013527 convolutional neural network Methods 0.000 description 13
- 230000001613 neoplastic effect Effects 0.000 description 11
- 238000010586 diagram Methods 0.000 description 10
- 239000003550 marker Substances 0.000 description 8
- 210000004907 gland Anatomy 0.000 description 7
- 230000036210 malignancy Effects 0.000 description 6
- 238000011160 research Methods 0.000 description 6
- 238000007490 hematoxylin and eosin (H&E) staining Methods 0.000 description 5
- 238000010606 normalization Methods 0.000 description 5
- 241000321096 Adenoides Species 0.000 description 4
- 201000009030 Carcinoma Diseases 0.000 description 4
- 210000002534 adenoid Anatomy 0.000 description 4
- 238000003748 differential diagnosis Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 210000003205 muscle Anatomy 0.000 description 4
- 239000013610 patient sample Substances 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000004393 prognosis Methods 0.000 description 4
- 208000000583 Adenolymphoma Diseases 0.000 description 3
- 208000012999 benign epithelial neoplasm Diseases 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000003745 diagnosis Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 238000013115 immunohistochemical detection Methods 0.000 description 3
- 238000002372 labelling Methods 0.000 description 3
- 230000003902 lesion Effects 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 208000025444 tumor of salivary gland Diseases 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000015556 catabolic process Effects 0.000 description 2
- 238000002790 cross-validation Methods 0.000 description 2
- 238000013434 data augmentation Methods 0.000 description 2
- 238000006731 degradation reaction Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000013140 knowledge distillation Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000036244 malformation Effects 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000036961 partial effect Effects 0.000 description 2
- 238000010827 pathological analysis Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000002829 reductive effect Effects 0.000 description 2
- 210000003079 salivary gland Anatomy 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 210000001519 tissue Anatomy 0.000 description 2
- 206010006187 Breast cancer Diseases 0.000 description 1
- 208000026310 Breast neoplasm Diseases 0.000 description 1
- 208000005623 Carcinogenesis Diseases 0.000 description 1
- 235000008694 Humulus lupulus Nutrition 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 210000002469 basement membrane Anatomy 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000017531 blood circulation Effects 0.000 description 1
- 210000000481 breast Anatomy 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000036952 cancer formation Effects 0.000 description 1
- 231100000504 carcinogenesis Toxicity 0.000 description 1
- 230000002860 competitive effect Effects 0.000 description 1
- 230000000120 cytopathologic effect Effects 0.000 description 1
- 210000004292 cytoskeleton Anatomy 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000001744 histochemical effect Effects 0.000 description 1
- 230000002962 histologic effect Effects 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000003364 immunohistochemistry Methods 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 210000004324 lymphatic system Anatomy 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 230000009826 neoplastic cell growth Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000035755 proliferation Effects 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 210000000106 sweat gland Anatomy 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
- 231100000588 tumorigenic Toxicity 0.000 description 1
- 230000000381 tumorigenic effect Effects 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
- G06V20/695—Preprocessing, e.g. image segmentation
-
- 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
-
- 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
- G06V10/764—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
- 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
- G06V10/77—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
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- 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
- 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
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H70/00—ICT specially adapted for the handling or processing of medical references
- G16H70/60—ICT specially adapted for the handling or processing of medical references relating to pathologies
-
- 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)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Databases & Information Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Public Health (AREA)
- Molecular Biology (AREA)
- Epidemiology (AREA)
- Data Mining & Analysis (AREA)
- Primary Health Care (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Pathology (AREA)
- Investigating Or Analysing Biological Materials (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a pathology image analysis method, a pathology image analysis system and pathology image analysis equipment based on deep learning. The method comprises the following steps: obtaining a cytopathology staining image of a sample; and detecting cells in the cytopathology staining image by adopting a cell detection model to obtain myoepithelial and/or adenoepithelial cells, wherein the cell detection model is constructed based on an immunohistochemical staining result as a guide. The method aims at obtaining the detection result of the myoepithelial and/or adenoepithelial cells through a cell detection model, and based on the detection result, classification prediction is carried out to obtain the classification result of benign and malignant cells and the prediction result of benign and malignant tumors, so as to discover the intelligent analysis capability and the potential application value of the benign and malignant cells in pathological image data.
Description
Technical Field
The present invention relates to the field of computer vision and pathology image analysis, and more particularly, to a deep learning-based pathology image analysis method, apparatus, system, computer-readable storage medium, and application thereof.
Background
In recent years, with the gradual application of computer vision and pattern recognition technology in medical auxiliary diagnosis, medical image processing and pathological analysis become a popular research and application field. Myoepithelial cells are the only tumor component in myoepithelial tumors and malignant myoepithelial tumors, and as a special type of cells, appear as different cell morphologies, between the epithelial cells of the salivary gland acini and leap ducts, breast and sweat gland acini and the basement membrane, the most remarkable feature of which is the presence of cytoplasmic filaments on the basal side, together with the structure and function of epithelial and mesenchymal cells, and the altered myoepithelial cells (neoplastic proliferation) may appear as epithelial and/or mesenchymal features. The gland epithelial cells are normal cells of a human body and are an important source cell for generating malignant tumors, and the cancerous gland epithelial cells can invade surrounding normal tissues, a lymphatic system and a blood circulation system and transfer to other parts of the body, so that the damage to the body is caused.
Currently, in conventional sections, myoepithelial cells and glandular epithelial cells are illegible. The morphological diversity of neoplastic myoepithelial cells, which makes it quite difficult to diagnose salivary gland tumors, is particularly difficult to identify when a salivary gland tumor is encountered with simultaneous neoplastic gonadal epithelial cell components. When the epithelium grows or forms tumors, the markers in the immunohistochemical detection can provide important help for the differential diagnosis of lesion types, are important auxiliary methods for pathological diagnosis and differential diagnosis, are also main means for cancer typing and accurate treatment marker screening, and have high price in the immunohistochemical detection.
Disclosure of Invention
The embodiment of the application provides a cell detection model construction method, a pathology image analysis system, a pathology image analysis device, a computer readable storage medium and application thereof, which aim to help to conduct classification prediction of benign and malignant epithelial tumors based on specific classification and identification of glandular epithelial cells and myoepithelial cells, and conduct identification research of a whole sample on a pathology staining image by taking an immunohistochemical staining result as guide training to construct a cell detection model, wherein the identification research comprises glandular epithelial cells and myoepithelial cells prediction analysis, benign and malignant classification and benign and malignant tumor prediction result, so as to provide more sufficient support and potential application value for doctors in the aspect of auxiliary diagnosis analysis of tumor lesion types.
According to a first aspect of the present application, an embodiment of the present application provides a method for constructing a cell detection model, where a construction process of the model includes: obtaining a cytopathology staining image and a corresponding immunohistochemical staining image of a sample, constructing a training set, and obtaining target cells of the sample based on an immunohistochemical result of target cell markers in the immunohistochemical staining image; inputting the cytopathology staining image of the sample into a deep learning model to obtain predicted target cells, comparing the predicted target cells with target cells in a training set, and optimizing the deep learning model to obtain a trained cell detection model.
Further, the target cells include myoepithelial cells and/or adenoepithelial cells.
Further, the selectable target cell markers include any one or more of the following: p63, calponin, SMA, S-100, CK13, CK14, CK7, EMA, CAM5.2, P53.
Still further, the markers for myoepithelial cells include P63, calponin, SMA, S-100, CK13, CK14, CK7.
Still further, the markers of the glandular epithelial cells include EMA, CK7, CAM5.2, P53.
Still further, the target cells filter the corresponding immunohistochemical staining image according to the pixel value range in the immunohistochemical staining result of the target cell marker to obtain the pixel value of positive staining in the immunohistochemical staining image, and threshold segmentation is carried out based on the pixel value to obtain a cell mask image; filling the closed region in the cell mask map to obtain a communication region of each cell; calculating the area and the aspect ratio of the connected region of each cell, and filtering the connected region with the area and the aspect ratio outside a preset threshold value.
Further, the optimized deep learning model is a trained cell detection model obtained by optimizing according to the predicted loss value between the target cells and the target cells in the training set. Wherein the deep learning model comprises an instance segmentation model and/or a target detection model, and the optimization comprises data preprocessing, data augmentation, random weight attenuation, learning rate, countermeasure training, regularization, adamW optimizer and self-knowledge distillation.
Still further, the deep learning model includes any one or more of the following models: YOLO, swinTransformer, MT-UNet, FA-SSD, FF-SSD, R-CNN, R-FCN, refineDet.
In a preferred embodiment, the cell detection model is constructed based on a modified YOLOV5 deep learning model, the modified YOLOV5 introducing a lightweight U-Net in the original YOLOV5 to achieve segmentation of target cells, the lightweight U-Net subtracting a set of up-sampling and down-sampling modules from the original U-Net, and reducing the number of all convolutions to 0.5 times.
According to a second aspect of the present application, an embodiment of the present application provides a pathology image analysis method based on deep learning, including: obtaining a cytopathology staining image of a sample; and detecting the cells in the cytopathology staining image by adopting the cell detection model to obtain the myoepithelial cells and/or the adenoepithelial cells.
Further, the pathological image analysis method based on deep learning further comprises the following steps: obtaining a cytopathology staining image of a sample; detecting cells in the cytopathology staining image by adopting the cell detection model to obtain myoepithelial cells and/or glandular epithelial cells; classifying the detected myoepithelial cells and/or glandular epithelial cells to obtain classification results of benign gonadal epithelial cells, malignant gonadal epithelial cells, benign myoepithelial cells and/or malignant myoepithelial cells.
Still further, the pathological image analysis method based on deep learning further comprises the following steps: obtaining a cytopathology staining image of a sample; detecting cells in the cytopathology staining image by adopting the cell detection model to obtain myoepithelial cells and/or glandular epithelial cells; classifying the detected myoepithelial cells and/or glandular epithelial cells to obtain classification results of benign gonadal epithelial cells, malignant gonadal epithelial cells, benign myoepithelial cells and/or malignant myoepithelial cells; prediction results of malignant or benign tumors are obtained based on the classification results, including malignant or benign tumors of the glandular epithelial and/or myoepithelial cell types.
Further, the benign tumor comprises any one or more of the following benign tumors: basal cell adenoma, warthin tumor, cystic adenoma, myoepithelial tumor, polymorphous adenoma, papillary salivary adenoma, said malignant tumor comprising any one or more of the following malignant tumors: mucinous epidermoid carcinoma, acinar cell carcinoma, myoepithelial carcinoma, adenoid cystic carcinoma, polymorphous adenocarcinomas, and epithelial-myoepithelial carcinoma.
Still further, for obtaining a cytopathology stained image of the sample, the image containing glandular epithelial cells is glandular epithelial type image, the image containing myoepithelial cells is myoepithelial type image, and the image containing glandular epithelial cells and myoepithelial cells is glandular epithelial and myoepithelial type image. Wherein, the prediction result of the malignant tumor or benign tumor specifically comprises: for benign images of the glandular epithelium, the prediction results include any one or more of the following benign tumors: basal cell adenoma, warthin tumor and cystic adenoma; for adenoepithelial malignancy images, the prediction result includes any one or more of the following malignant tumors: mucinous epidermoid carcinoma and acinar cell carcinoma; for benign images of the myoepithelial class, the predicted outcome is myoepithelial neoplasia; for myoepithelial malignancy images, the prediction result is myoepithelial cancer; for benign images of the glandular and myoepithelial types, the prediction results include any one or more of the following benign tumors: polymorphic adenoma, papillary salivary adenoma; for glandular and myoepithelial malignancy images, the prediction results include one or more of the following malignancies: adenoid cystic carcinoma, polymorphous adenocarcinoma, and epithelial-myoepithelial carcinoma.
Further, in some embodiments, the method of classifying includes image histology and/or deep learning.
Preferably, the classification method is a fusion of image histology and deep learning methods.
Still further, the image histology performs benign and malignant taxonomic learning of myoepithelial cells and/or glandular epithelial cells by extracting nuclei, cytoplasm, and related features therebetween.
Preferably, the image histology adopts an edge detection method to separate cell nuclei and cytoplasms, respectively extracts cell, cell nuclei and cytoplasms to obtain image histology characteristics, and then inputs the image histology characteristics into a classifier to obtain a classification result that the cells are benign cells or malignant cells. Wherein the image histology features comprise any one or more of the following features: size, shape, shade of color, uniformity, presence or absence of cavitation, and nuclear cytoplasmic ratio of the cytoplasm; the classifier comprises any one or more of the following algorithms: support vector machines, decision trees, random forests, XGBoost, lightGBM.
Still further, the deep learning is implemented based on one or more of the following models: resNet, resNetXt, mobileNet, shuffleNet, squeezeNet, efficientNet, mnasNet, NFNet, alexNet, VGG, googleNet, viT, EVA.
In a preferred embodiment, the deep learning uses modified mobilenet v3 to classify cells for benign and malignant classification, the modified mobilenet v3 performing the following operations on the original mobilenet v3-Small model: deleting a first downsampling module and a first set of bnecks of the original model; introducing deformable convolution modules sensitive to shape characteristics into the last two groups of bneck of the original model; introducing a pooling pyramid structure before pooling 7x7 of the original model, and extracting to obtain information features with different scales; and performing dimension reduction on the output 1024-dimension features through a group of 1x1 dimension reduction convolution layers to obtain multi-dimension features. Preferably, the dimension reduction range of dimension reduction comprises 10-50, and the pooling pyramid structure comprises a cavity convolution layer, a deformable convolution layer and a global pooling module, wherein the step sizes of the cavity convolution layer, the deformable convolution layer and the global pooling module are respectively 1, 3 and 5; the deformable convolution module comprises any one or more of the following sub-modules: DCNV1, DCNV2, DCNV3.
Further optionally, the global pooling module includes a global pooling layer, an upsampling layer, and a convolution layer.
In some alternative schemes, the fusion comprises result fusion and feature fusion, the result fusion is to fuse the results predicted by the image histology and the deep learning method, the feature fusion is to fuse the histology features obtained by the image histology method and the multi-scale features obtained by the deep learning method, and the result fusion method comprises hard voting fusion, weighting fusion, stacking fusion and blending fusion.
According to a third aspect of the present application, an embodiment of the present application provides a pathology image analysis system based on deep learning, which includes a computer program, and when the computer program is executed, implements the pathology image analysis method based on deep learning or implements the cell detection model construction method described above.
Further, from the view of the module structure, the pathology image analysis system based on deep learning further comprises an acquisition module, a cell detection module, a cell classification module and a tumor prediction module.
Still further, an acquisition module for acquiring a cytopathology staining image of the sample;
the cell detection module is used for detecting cells in the cell pathology staining image by adopting the cell detection model to obtain myoepithelial cells and/or glandular epithelial cells;
the cell classification module is used for classifying the detected myoepithelial cells and/or glandular epithelial cells to obtain classification results of benign gonadal epithelial cells, malignant gonadal epithelial cells, benign myoepithelial cells and/or malignant myoepithelial cells;
a tumor prediction module predicts a prediction of a malignant tumor or a benign tumor based on the classification results, the prediction comprising a malignant or benign tumor of the glandular epithelial and/or myoepithelial cell type.
Further, the benign tumor comprises any one or more of the following benign tumors: basal cell adenoma, warthin's tumor, cystic adenoma, myoepithelial tumor, polymorphic adenoma, papillary salivary adenoma; the malignant tumor comprises any one or more of the following malignant tumors: mucinous epidermoid carcinoma, acinar cell carcinoma, myoepithelial carcinoma, adenoid cystic carcinoma, polymorphous adenocarcinomas, and epithelial-myoepithelial carcinoma.
According to a fourth aspect of the present application, an embodiment of the present application provides a pathology image analysis apparatus based on deep learning, mainly including: a memory and a processor; the memory is used for storing program instructions; the processor is used for calling program instructions, and when the program instructions are executed, the processor is used for executing the pathological image analysis method or the cell detection model construction method based on the deep learning.
According to a fifth aspect of the present application, an embodiment of the present application provides a computer-readable storage medium having stored thereon a computer program for performing pathological image analysis, which when executed by a processor, implements the above-described pathological image analysis method or cell detection model construction method based on deep learning.
According to a sixth aspect of the present application, an embodiment of the present application provides related applications thereof, mainly including:
the application of the device or the system in intelligent analysis of the degree of glandular epithelial and myoepithelial cytopathy or sample canceration;
the application of the device or the system in carrying out auxiliary prediction of malignant tumors and benign tumors on patients; optionally, the prediction comprises a malignant tumor, a benign tumor prediction of the adeno-epithelial class, a malignant tumor, a benign tumor prediction of the myo-epithelial class, a malignant tumor, a benign tumor prediction of the adeno-myo-epithelial class;
the equipment or the system provides effectiveness suggestions in the application of assisting tumor lesion diagnosis and differential diagnosis and prognosis judgment and curative effect prediction; optionally, the differential diagnosis includes identifying important markers for benign and malignant breast cancer by detecting the integrity of the muscle epithelium, and predicting salivary gland tumor (a characteristic tumor of the oromaxillofacial region) by the muscle epithelium; optionally, the prognosis and the efficacy prediction include detection and classification learning of cells in the cytopathology staining image through a cell detection model, so as to obtain a cell classification result and a tumor prediction result, and the method has positive influence and promotion effect on the research of prognosis and efficacy prediction.
According to the invention, the immunohistochemical staining result is taken as a guide training to construct a cell detection model, and the machine learning algorithm and the medical image processing technology are applied to realize cell detection and classification tasks in the cytopathology staining image, so that the influence of subjective judgment of the experience of a doctor and expensive immunohistochemical detection is overcome, the automatic classification of the glandular epithelial cells and the myoepithelial cells and the tumor prediction of the corresponding whole sample are realized, the innovation is very strong, and a beneficial pushing effect is generated on the analysis and research of pathological image data.
The application has the advantages that:
1. the application creatively discloses a cell detection model, wherein a training set is constructed by acquiring a cell pathology staining image and a corresponding immunohistochemical staining image of a sample, and muscle epithelial cells and/or gland epithelial cells obtained based on the immunohistochemical result in the training set are used as a result label to train a pathology image, so that the trained model can detect the muscle epithelial cells and/or gland epithelial cells in the pathology image, and the aging is obvious;
2. the innovative pathological image analysis method based on deep learning applies a cell detection model constructed by a machine learning algorithm and a medical image processing technology, outputs a cell detection result, and simultaneously carries out deep analysis based on the cell detection result to obtain benign and malignant classification of glandular epithelial cells and/or myoepithelial cells, so as to predict and obtain a prediction result of benign and malignant tumors, and objectively improves the accuracy and depth of data analysis;
3. The application creatively discloses a cell detection model constructed based on immunohistochemistry and pathology staining images, which is used for analyzing adeno epithelial cells and/or myo-epithelial cells in the cytopathology staining images, and meanwhile, can realize deep benign and malignant cell classification, so that a benign and malignant tumor prediction result of a whole sample is obtained, and has important research significance on prognosis and prevention and control in view of the cell classification and tumor prediction result, so that the method is more accurately applied to auxiliary analysis of tumorigenesis development related to cytopathology image data.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram of a use scenario of a cell detection model according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a flow chart for constructing a cell detection model according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of sample cell classification and prediction results according to an embodiment of the present invention;
Fig. 4 is a schematic diagram of a pathological image analysis flow based on deep learning according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of an immunohistochemical staining image and a sample prediction result provided by the embodiment of the invention;
fig. 6 is a schematic diagram of a pathology image analysis apparatus based on deep learning according to an embodiment of the present invention.
Detailed Description
In order to enable those skilled in the art to better understand the present invention, the following description will make clear and complete descriptions of the technical solutions according to the embodiments of the present invention with reference to the accompanying drawings.
In some of the flows described in the specification and claims of the present invention and in the above figures, a plurality of operations appearing in a particular order are included, but it should be clearly understood that the operations may be performed in other than the order in which they appear herein or in parallel, the sequence numbers of the operations such as S101, S102, etc. are merely used to distinguish between the various operations, and the sequence numbers themselves do not represent any order of execution. In addition, the flows may include more or fewer operations, and the operations may be performed sequentially or in parallel.
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments according to the invention without any creative effort, are within the protection scope of the invention.
Embodiments of the present application provide a cell detection model construction method, a XXXX computer device, and a computer-readable storage medium. The method can be integrated into computer equipment, and the computer equipment can be a terminal or a server and other equipment. The terminal can be terminal equipment such as a smart phone, a tablet personal computer, a notebook computer, a personal computer and the like. The server may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content Delivery Networks (CDNs), basic cloud computing services such as big data and artificial intelligent platforms, and the like. The terminal and the server may be directly or indirectly connected through wired or wireless communication, which is not limited herein.
Referring to fig. 1, fig. 1 is a schematic view of a usage scenario of a cell detection model according to an embodiment of the present invention, and specifically, a computer device shown in fig. 1 is a terminal device, and the terminal device may obtain a cytopathology staining image of a sample; and detecting the cells in the cytopathology staining image by adopting a cell detection model to obtain the myoepithelial cells and/or the adenoepithelial cells. The cytopathology staining image also comprises a corresponding immunohistochemical staining image and a corresponding result.
Further, as shown in fig. 2, the construction of the cell detection model specifically includes the following steps:
s101: and obtaining a cytopathology staining image and a corresponding immunohistochemical staining image of the sample, constructing a training set, and obtaining target cells of the sample based on an immunohistochemical result of target cell markers in the immunohistochemical staining image.
In one embodiment, the cytopathology stain image of the obtained sample includes an HE stain pathology image of the patient's epithelial cell sample, a portion of the corresponding immunohistochemical staining image, and the corresponding immunohistochemical staining result.
Further, the target cells include myoepithelial cells and/or adenoepithelial cells.
Still further, markers for labeling myoepithelial cells include P63, calponin, SMA, S-100, CK13, CK14, CK7.
Still further, markers for labeling glandular epithelial cells include EMA, CK7, CAM5.2, P53.
Still further, the marker for simultaneous labeling of glandular and myoepithelial cells was CK7.
In a preferred embodiment, the target cells of the sample are obtained by filtering the corresponding immunohistochemical staining image according to the pixel value range in the immunohistochemical staining result of the target cell marker to obtain the pixel value of positive staining in the immunohistochemical staining image, and performing threshold segmentation based on the pixel value to obtain a cytomask image; filling the closed region in the cell mask map to obtain a communication region of each cell; the area and aspect ratio of the connected region of each cell were calculated, and the connected region where the filter area and aspect ratio were outside the preset threshold was obtained.
Further, the obtained cytopathology staining image and the corresponding immunohistochemical staining image comprise sample types shown in fig. 3, and the immunohistochemical staining result of the sample comprises the prediction results of benign and malignant tumors of three types, namely, glandular epithelium, myoepithelium and glandular epithelium-myoepithelium, which are formed by two types of cells, namely glandular epithelium and myoepithelium. FIG. 3 is a schematic diagram showing the classification and prediction results of sample cells according to an embodiment of the present invention. Among them, malignant epithelial tumors shown in fig. 3 include six malignant epithelial tumor subtypes of myoepithelial cancer (myoepithelial cells), mucous epidermoid cancer (adenoepithelial cells), acinar cell cancer (adenoepithelial cells), adenoid cystic cancer (adenoepithelial cells+myoepithelial cells), polymorphous adenocarcinomas (adenoepithelial cells+myoepithelial cells) and epithelial-myoepithelial cancer (adenoepithelial cells+myoepithelial cells), and benign epithelial tumors may be classified into six benign tumor subtypes of polymorphous adenoma (adenoepithelial cells+myoepithelial cells), papillary salivary adenoma (adenoepithelial cells+myoepithelial cells), myoepithelial tumor (myoepithelial cells), basal cell adenoma (adenoepithelial cells), warthin tumor (adenoepithelial cells), and cystic adenoma (adenoepithelial cells).
Further, in a specific embodiment, obtaining a myoepithelial cell region and/or obtaining a glandular epithelial cell region based on the immunohistochemical result of the myoepithelial cell marker and/or the glandular epithelial cell marker in the immunohistochemical staining image of the sample, performing preliminary filtration on the corresponding immunohistochemical staining image according to the pixel value range in the immunohistochemical staining result generated by the myoepithelial cell marker and/or the glandular epithelial cell marker to obtain the pixel value of positive staining in the immunohistochemical staining image, and then performing threshold segmentation based on the pixel value to obtain a cell binary mask map; filling the closed region in the cell binary mask map to obtain a communication region of each cell; and finally, calculating the area and the aspect ratio of the communication area of each cell, and filtering the communication areas with the area and the aspect ratio outside a preset threshold value to obtain the corresponding myoepithelial cells and/or adenoepithelial cells. Wherein, in the immunohistochemical staining image of the sample, the tumor myoepithelial cells express myoepithelial markers: p63, calponin, SMA, S-100, CK13, CK14, CK7, positive; the luminal surface epithelium expression gland epithelium markers of the tumor gland duct-like structure are: EMA, CK7, CAM5.2, P53 positive; markers for simultaneous expression of neoplastic myoepithelial cells and glandular epithelium are: CK7, positive.
In a more specific embodiment, the construction of the training set for epithelial cell detection is divided into five steps altogether. The first step, performing preliminary filtration according to the pixel value range, and filtering partial background cells and other tissues in the immunohistochemical staining image, wherein the filtered pixel values are pixel values except positive staining such as yellow, brown and the like; secondly, searching a segmentation threshold value by adopting an OSTU algorithm, and further generating a cell binary mask; thirdly, filling the closed area by adopting a filling algorithm, inhibiting the influence of the cavity in the cell, and further obtaining the communication area of each cell; a fourth step of calculating the area and aspect ratio of each communication area, and the communication areas with the filter area and aspect ratio outside a preset threshold; and fifthly, mapping the connected region to a cytopathology staining image (such as an HE staining image), generating a target mask image of each cell, and constructing a training set of a cell detection model.
S102: inputting the cytopathology staining image of the sample into a deep learning model to obtain predicted target cells, comparing the predicted target cells with target cells in a training set, and optimizing the deep learning model to obtain a trained cell detection model.
In some embodiments, the cell detection model detects cells in a cytopathology staining image of a sample based on the deep learning model to obtain predicted target cells, calculates a loss value between the predicted target cells and target cells in a training set, and optimizes the model according to the loss value to obtain a trained cell detection model.
In one embodiment, the cell detection model is trained from a training set, and the trained cell detection model automatically detects glandular epithelial cells and/or myoepithelial cells in the HE stained image. The training set includes HE images and target mask maps, wherein the target mask maps are automatically generated from corresponding immunohistochemical staining images.
Further, the deep learning model includes an instance segmentation model and/or a target detection model.
Still further, the deep learning model includes any one or more of the following model frameworks: YOLO, swinTransformer, MT-UNet, FA-SSD, FF-SSD, R-CNN, R-FCN, refineDet.
YOLO (YouOnlyLookOnce) is a model family, and specific position information and category classification information of target detection are obtained in a direct regression mode, so that a plurality of tasks are solved while training is performed, the calculated amount is greatly reduced, and the detection speed is remarkably improved. Evolution of YOLO family series model: from v1 to v8, YOLOv8 currently contains five models for detection, segmentation and classification tasks.
SwinTransformer is a hierarchical visual transducer using a moving window, features with different sizes in an image are learned by designing the hierarchical transducer, and interaction is performed between windows by using the moving window, so that the image is globally modeled, and higher calculation efficiency and model performance are presented.
MT-UNet, a new hybrid transducer module for medical image segmentation, is capable of inter-affinities and intra-affinities learning simultaneously, by first efficiently computing window internal affinities by local-global Gaussian weighted self-attention, and then mining the links between data samples by external attention.
The FA-SSD builds a network frame based on SSD (singleshotmultiboxdetector) algorithm, combines the feature fusion and attention module, connects the target feature and the context feature by superposing the features, places the attention module of onestage on the target feature, and realizes the feature map fusion of higher layers with the target feature layer.
The FF-SSD (featurefusionbasedSSD) is a target detection algorithm, can obtain higher detection precision for various targets under a complex background, introduces modulation factors into the original loss function, and realizes the semantic information fusion enhancement of a low-level feature map by constructing a feature pyramid and fusing a multi-level feature map so as to improve the precision of small target detection and the overall detection precision.
R-CNN, which is known as Region-CNN, is used as a strategy of R-CNN in a Region selection stage by a selective search method, and the R-CNN model which is mature and universal at present is FasterR-CNN, mask-CNN and CascadeR-CNN.
R-FCN, collectively referred to as Region-based FullyConvolitionalNet, is a Region-based, fully convoluted network capable of handling both position variability (locationvariance) and position invariance (locationvariance) simultaneously through a position sensitive score map, with all the computations in the network being shared.
The refinished, which consists of two interconnected modules, namely an anchor refinement module (Anchor RefinementModule) and an object detection module (ObjectDetectionModule), achieves better accuracy than the two-stage approach and maintains the equivalent efficiency of the one-stage approach.
In a preferred embodiment, the cell detection model is constructed based on a modified YOLOV5, the modified YOLOV5 introducing a lightweight U-Net in the original YOLOV5 to achieve segmentation of the target cells, the lightweight U-Net subtracting a set of up-sampling and down-sampling modules from the original U-Net, and reducing the number of all convolutions to 0.5 times.
Further, in the training process, the optimization method of the model comprises data preprocessing, data augmentation, random weight attenuation, learning rate, countermeasure training, regularization, adamW optimizer and self-knowledge distillation.
In a more specific embodiment, the corresponding immunohistochemical staining results are obtained as model-trained or model-validated test criteria.
Furthermore, the computer device shown in fig. 1 is a terminal device, and can also learn the classification of the myoepithelial cells and/or the glandular epithelial cells detected by the cell detection model, so as to obtain the classification results of the benign gonadal epithelial cells, the malignant gonadal epithelial cells, the benign myoepithelial cells and/or the malignant myoepithelial cells.
And further, carrying out further tumor prediction analysis on the cytopathology staining image based on the classification result to obtain a prediction result of benign and malignant tumors. The essence of the classification learning is to perform classification learning according to the heterogeneity characteristics of benign and malignant cells based on the detected myoepithelial cells and/or adenoepithelial cells, so as to obtain classification results of the benign gonadal epithelial cells, malignant gonadal epithelial cells, benign myoepithelial cells and/or malignant myoepithelial cells.
Specifically, fig. 4 is a schematic diagram of a pathological image analysis flow provided in an embodiment of the present invention.
S401: a cytopathology staining image of the sample is obtained.
In one embodiment, the cytopathology stain image of the acquired sample includes pathology features based on the HE staining pathology image, predominantly in two sample types, malignant epithelial tumors, benign epithelial tumors.
Further, the histochemical staining results corresponding to the cytopathological staining image of the obtained sample include the corresponding results obtained by expressing the myoepithelial markers by the above-mentioned neoplastic myoepithelial cells, expressing the adenoepithelial markers by the luminal surface epithelium of the neoplastic adenoid structures, and simultaneously expressing the markers by the neoplastic myoepithelial cells and the adenoepithelium, as shown in fig. 5, which is a schematic diagram of the partial immunohistochemical staining image obtained under a x 200 microscope and the sample prediction results.
Specifically, FIG. 5A shows that patient samples marked by neoplastic myoepithelial cells expressing the myoepithelial marker S-100 are positive, i.e., S-100 positive; FIG. 5B shows a sample of a patient labeled with CK7, a marker expressed simultaneously by neoplastic myoepithelial cells and glandular epithelium, resulting in positive for glandular and myoepithelial cells, i.e., positive for CK 7; FIG. 5C shows that samples of patients marked with the tumor adenoid structures by the luminal surface epithelium expression gland epithelium marker EMA are positive, i.e., EMA positive; FIG. 5D shows that patient samples marked by the tumor adenoid structures for the luminal surface epithelium expression of the adenoid marker P53 are positive, i.e., P53 positive; FIG. 5E shows that patient samples labeled with the tumor myoepithelial cells expressing the myoepithelial marker Calponin were positive, i.e., calponin positive; FIG. 5F shows that patient samples labeled with the myoepithelial marker P63 expressed by neoplastic myoepithelial cells were positive, i.e., P63 positive.
More specifically, the markers selected in the present patent are directed to clinical significance of the predicted outcome:
myoepithelial markers positive for tumor myoepithelial cell expression are: p63, calponin, SMA, S-100, CK13, CK14, CK7;
the luminal epithelial markers positive for expression of tumor adeno-tubular structures are: EMA, CK7, CAM5.2, P53;
markers positive for simultaneous expression of neoplastic myoepithelial cells and glandular epithelium are: CK7.
S402: and detecting cells in the cytopathology staining image by adopting a cell detection model to obtain myoepithelial cells and/or adenoepithelial cells.
Further, the cytopathology staining image for cell detection by using the cell detection model mainly comprises a sample type shown in fig. 3, and a malignant tumor and benign tumor sample of three combined types, namely, glandular epithelium, myoepithelium and glandular epithelium-myoepithelium, which are mainly composed of two types of cells, namely glandular epithelium and myoepithelium.
S403: classifying the detected myoepithelial cells and/or glandular epithelial cells to obtain classification results of benign gonadal epithelial cells, malignant gonadal epithelial cells, benign myoepithelial cells and/or malignant myoepithelial cells.
In some embodiments, the classification result also includes abnormal epithelial cells, such as cancerous glandular epithelial cells and/or myoepithelial cells (malignant epithelial cells) in the salivary glands, as well as suspicious diseased cells. The characteristics of malignant epithelial cells on HE staining images are mainly represented by enlargement of cell nuclei, malformation, deepening of staining, cytoplasmia reduction, malformation, deepening of staining and the like.
Further, the method of classifying includes image histology and/or deep learning.
In a preferred embodiment, the method for performing a classification study of benign and malignant myoepithelial cells and/or adenoepithelial cells is a fusion of an image histology method and a depth model. The measure of fusion comprises result fusion and feature fusion.
Further, the result fusion is to fuse the result predicted by the image histology and the deep learning method, and the feature fusion is to fuse the histology feature obtained by the image histology method and the multi-scale feature obtained by the deep learning method.
Further, the features extracted by image histology are mainly focused on the nucleus, cytoplasm and the association between the two. Firstly, separating cell nucleus and cytoplasm by adopting methods such as edge detection and the like; then, respectively extracting the characteristics of cells, cell nuclei and cytoplasm, wherein the characteristics mainly comprise the size, shape and color shade of the cell nuclei, the size, shape, color shade, uniformity, presence or absence of cavitation bubbles, nuclear cytoplasmic ratio and the like; and finally, sending the features into a classifier to generate a benign and malignant classification result. Wherein, the classifier can adopt any one or more of the following algorithms: support vector machines, decision trees, random forests, XGBoost, lightGBM, etc.
Further, the method of deep learning is based on one or more of the following models: resNet, resNetXt, mobileNet, shuffleNet, squeezeNet, efficientNet, mnasNet, NFNet, alexNet, VGG, googleNet, viT, EVA.
ResNet is a framework for learning residuals, which solves the problems of gradient explosion and convergence of the extremely deep network to a great extent through a normalization initialization layer and an intermediate normalization layer, but causes degradation problems, and the idea is to solve the degradation problems by introducing a deep residual learning framework, so that each stacked layer fits a residual mapping instead of fitting the layers to an implicit mapping.
ResNetXt aggregates multiple transformations (e.g., finding the best structure in the indication) by repeating multiple blocks (e.g., stacks in VGG) while introducing cross-layer connections (hops in ResNet).
The MobileNet series is representative of a lightweight network, so that the light weight and mobile end deployment of CNNs are possible, and three versions of MobileNet v1, mobileNet v2 and MobileNet v3 are available at present.
The SheffeNet combines GroupConv and ChannelSheffe, improves the ResNet basic module bottleck, and the SheffeNet V2 basic module is cascaded and matched with convolution, pooling and other links.
The SquezeNet replaces more parameters through deeper networks, has the same precision as AlexNet, but only uses the parameter quantity of AlexNet1/50, and the core is to reduce the parameter quantity by adopting a convolution mode different from a conventional convolution mode, and reduce the channel quantity and increase the channel quantity by using convolution for multiple times through FireModule.
EfficientNet is a lightweight neural convolutional network, and the importance of the improvement of the convolutional network precision is three dimensions of network depth, width and input image resolution.
MnasNet is a terminal lightweight model, and is also an automatic neural structure searching method of a terminal CNN model with resource constraint, and the main idea is to use reinforcement learning.
NFNet is a network without batch normalization, and is realized through a training strategy of residual branches without batch normalization and adaptive gradient clipping, so that compared with an optimal network with a normalization layer, the NFNet is faster in training speed and more competitive in performance.
AlexNet opens the application history of deep learning models in image classification, which contains 8 learning layers: the 5 convolution layers and the 3 full connection layers lead the response value to be relatively larger and inhibit other neurons with smaller feedback by introducing a competition mechanism to the activity of local neurons, so that the generalization capability is strong.
VGG refers to a very deep convolutional network for large-scale image recognition, the improvement of the prior art can be realized by pushing the depth to 16-19 layers, and the core idea is to use an architecture with a very small convolutional kernel, increase the depth of the network and then perform comprehensive evaluation.
The concept of GoogleNet and AlexNet/VGGNet that relies on deepening the depth of the network structure is not exactly the same, but makes structural innovations while deepening the depth, and introduces a structure called concept to replace the classical components of convolution and activation before.
ViT (VisionTransformer) is a powerful competitor to convolutional neural networks, which are currently the most advanced technology in the field of computer vision and are widely used in many image recognition applications. The ViT model is nearly four times more than convolutional neural networks in terms of computational efficiency and accuracy.
EVA is a simple and powerful visual basic model with 10 hundred million parameters, and has multiple SOTAs as a visual pre-training method, so that mask image modeling can be realized.
In a preferred embodiment, deep learning employs modified MobileNetV3 for benign and malignant classification learning of cells. Considering the characteristics of Small size, large number and the like of the cell images to be classified, the improved MobileNet V3 deletes a first downsampling module and a first group of bneck on the basis of the original MobileNet V3-Small; in consideration of the characteristic that the shape of malignant cells is changed, a deformable convolution module sensitive to shape characteristics is introduced into the last two groups of bneck of the original model, and the deformable convolution modules can be DCNV1, DCNV2 and DCNV3; in consideration of the importance of malignant cell size information, an optimized pooling pyramid structure is introduced before 7x7 pooling of an original model, the optimized pooling pyramid structure mainly comprises a cavity convolution layer, a deformable convolution layer and a global pooling module with steps of 1, 3 and 5 respectively, and the modules extracting different scale information can better utilize multi-scale characteristics of malignant cells, wherein the global pooling module mainly comprises a global pooling layer, an up-sampling layer and a convolution layer.
For the modified MobileNetV3, the 1024-dimensional features it outputs will be reduced again by a set of 1x1 convolutional layers to ensure matching with the lower dimensional histologic features, which can be from 10 to 50.
Still further, the result fusion method comprises hard voting fusion, weighted fusion, stacking fusion and blending fusion.
In a specific embodiment, the malignant epithelial cells are classified according to their characteristics on the HE staining image, and the classification method may be any of the following: image histology, deep learning and fusion of the two, the characteristics of malignant epithelial cells on HE staining images are mainly represented by enlargement of cell nuclei, deformity, deepening of staining, cytoplasmia reduction, deformity, deepening of staining and the like.
S404: and predicting based on the classification result to obtain the prediction result of the malignant tumor or the benign tumor.
Further, the predicted outcome includes malignant or benign tumors of the glandular epithelial and/or myoepithelial cell types.
In one embodiment, the prediction result of the malignant tumor or the benign tumor is obtained by predicting by a classifier based on the types of the epithelial cells and the benign and malignant cells contained in the classification result. The prediction results are shown in fig. 3, and include classification results of three combination types of glandular epithelium, myoepithelium, glandular epithelium and myoepithelium based on the obtained glandular epithelium and myoepithelium, and prediction results of malignant tumors and benign tumors of the three combination types of glandular epithelium, myoepithelium, glandular epithelium and myoepithelium are obtained through a classifier respectively. In the method, in the cell pathology staining image to be treated, the image containing the glandular epithelial cells is an glandular epithelial image, the image containing the myoepithelial cells is a myoepithelial image, and the image containing the glandular epithelium and the myoepithelium is an glandular epithelium and a myoepithelial image. For benign images of the glandular epithelium, the predicted outcome is basal cell adenoma, warthin's tumor and cystic adenoma; for adenoepithelial malignancy images, the predicted outcome is mucinous epidermoid carcinoma and acinar cell carcinoma; for benign images of myoepithelial type, the predicted outcome is myoepithelial tumor; for myoepithelial malignant images, the predicted outcome is myoepithelial cancer; for benign images of glandular and myoepithelial types, the predicted outcome is polymorphous adenoma, papillary salivary adenoma; for adeno-and myo-epithelial malignancy images, the predicted outcome is adenoid cystic carcinoma, polymorphous adenocarcinoma, and epithelial-myo-epithelial carcinoma.
In a specific embodiment, random oversampling and cross-validation were performed taking into account the imbalance of the three data types of glandular, myoepithelial, glandular-myoepithelial cells, oversampling was performed only in training folds, while the optimal number of features was determined by maximizing the macroscopic area under the subject's operating feature curve of the cross-validation results. Once the selected features are determined, the entire training set is used to retrain the model and the test set is used to evaluate the model performance.
The method is feasible for analysis and prediction of the cytopathology staining image of the sample, shows that the cell prediction of myoepithelial cells and/or adenoepithelial cells in the cytopathology staining image is realized based on a deep learning target detection model and a cell detection model constructed by taking an immunohistochemical staining result as a training guide, so that classification results of benign gonadal epithelial cells, malignant gonadal epithelial cells, benign myoepithelial cells and/or malignant myoepithelial cells are obtained by classification, a prediction result of benign malignant tumors is obtained based on the classification result prediction, more accurate layer-by-layer pathology information is provided for image evaluation, and the model prediction better reflects specific classification prediction efficiency, so that the method is more beneficial in the aspect of being applied to auxiliary analysis of tumorigenic development prediction related to pathology image data.
The pathology image analysis system based on the deep learning provided by the embodiment of the invention comprises a computer program, and when the computer program is executed, the pathology image analysis method based on the deep learning or the construction method of the cell detection model are realized.
Further, from a modular structure, the deep learning-based pathology image analysis system can perform cell detection (myoepithelial cells and/or adenoepithelial cells), cell classification (benign gonadal epithelial cells, malignant gonadal epithelial cells, benign myoepithelial cells and/or malignant myoepithelial cells), and tumor prediction tasks. Specifically, the pathological image analysis system comprises any one or more of the following modules: the system comprises an acquisition module, a cell detection module, a cell classification module and a tumor prediction module.
Still further, an acquisition module for acquiring a cytopathology staining image of the sample.
And the cell detection module is used for detecting cells in the cell pathology staining image by adopting the cell detection model to obtain myoepithelial cells and/or glandular epithelial cells.
Still further, the cell classification module is used for classifying the detected myoepithelial cells and/or glandular epithelial cells to obtain classification results of benign gonadal epithelial cells, malignant gonadal epithelial cells, benign myoepithelial cells and/or malignant myoepithelial cells.
Still further, the tumor prediction module predicts a prediction of a malignant tumor or a benign tumor based on the classification result, wherein the prediction comprises a malignant tumor or a benign tumor of the glandular epithelial and/or myoepithelial cell type.
Still further, the benign tumor comprises any one or more of the following benign tumors: basal cell adenoma, warthin's tumor, cystic adenoma, myoepithelial tumor, polymorphic adenoma, papillary salivary adenoma; the malignant tumor comprises any one or more of the following malignant tumors: mucinous epidermoid carcinoma, acinar cell carcinoma, myoepithelial carcinoma, adenoid cystic carcinoma, polymorphous adenocarcinomas, and epithelial-myoepithelial carcinoma.
Fig. 6 is a schematic diagram of a pathology image analysis apparatus based on deep learning according to an embodiment of the present invention, including: a memory and a processor; the apparatus may further include: input means and output means.
The memory, processor, input device, and output device may be connected by a bus or other means, as illustrated by way of example in fig. 6; wherein the memory is used for storing program instructions; the processor is used for calling program instructions, and when the program instructions are executed, the processor is used for executing the pathological image analysis method based on deep learning or realizing the cell detection model construction method.
The invention provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor, implements the above-mentioned pathology image analysis method based on deep learning or implements the above-mentioned cell detection model construction method.
It will be clearly understood by those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described system, apparatus and module may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.
In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods may be implemented in other ways. For example, the device embodiments described above are merely illustrative; for another example, the division of the modules is just one logic function division, and other division modes can be adopted in actual implementation; as another example, multiple modules or components may be combined or may be integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices, or modules, which may be in electrical, mechanical, or other forms.
The modules illustrated as separate components may or may not be physically separate, and components shown as modules may or may not be physical modules, i.e., may be located in one place, or may be distributed over a plurality of network modules. Specifically, some or all modules in the system may be selected according to actual needs to achieve the purpose of the embodiment.
In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist alone physically, or two or more modules may be integrated into one module. The integrated module can be realized in a form of hardware or a form of a software functional module.
Those of ordinary skill in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be implemented by a program to instruct related hardware, the program may be stored in a computer readable storage medium, and the storage medium may include: read-only memory (ROM, readOnlyMemory), random access memory (RAM, randomAccessMemory), magnetic or optical disk, and the like.
Those of ordinary skill in the art will appreciate that all or part of the steps in the methods of the above embodiments may be implemented by a program, and the program may be stored in a computer readable storage medium, where the storage medium may be a read only memory, a magnetic disk, or an optical disk.
While the invention has been described in detail with respect to a computer device, those skilled in the art will appreciate that they can readily use the disclosed embodiments as a basis for the teaching of the present invention. In summary, the present description should not be construed as limiting the invention.
Claims (10)
1. A method for constructing a cell detection model, which comprises the following steps:
obtaining a cytopathology staining image and a corresponding immunohistochemical staining image of a sample, constructing a training set, and obtaining target cells of the sample based on an immunohistochemical result of target cell markers in the immunohistochemical staining image; inputting the cytopathology staining image of the sample into a deep learning model to obtain predicted target cells, comparing the predicted target cells with target cells in a training set, and optimizing the deep learning model to obtain a trained cell detection model.
2. The method according to claim 1, wherein the target cells include myoepithelial cells and/or adenoepithelial cells;
optionally, the target cell marker comprises any one or more of the following markers: p63, calponin, SMA, S-100, CK13, CK14, CK7, EMA, CAM5.2, P53;
Preferably, the markers of myoepithelial cells include P63, calponin, SMA, S-100, CK13, CK14, CK7;
preferably, the markers of the glandular epithelial cells comprise EMA, CK7, CAM5.2, P53;
preferably, the target cells filter the corresponding immunohistochemical staining image according to the pixel value range in the immunohistochemical staining result of the target cell marker to obtain the pixel value of positive staining in the immunohistochemical staining image, and threshold segmentation is performed based on the pixel value to obtain a cell mask image; filling the closed region in the cell mask map to obtain a communication region of each cell; calculating the area and the aspect ratio of the connected region of each cell, and filtering the connected region with the area and the aspect ratio outside a preset threshold value.
3. The method according to claim 1, wherein the deep learning model comprises an instance segmentation model and/or a target detection model;
preferably, the deep learning model comprises any one or more of the following models: YOLO, swinTransformer, MT-UNet, FA-SSD, FF-SSD, R-CNN, R-FCN, refineDet; preferably, the cell detection model is constructed based on a modified YOLOV5 deep learning model, the modified YOLOV5 introducing a lightweight U-Net in the original YOLOV5 to achieve segmentation of target cells, the lightweight U-Net subtracting a set of up-sampling and down-sampling modules from the original U-Net, and reducing the number of all convolutions to 0.5 times.
4. A deep learning-based pathology image analysis method, the method comprising:
obtaining a cytopathology staining image of a sample;
the use of the cell detection model of claims 1-3 to detect cells in said cytopathology staining image to obtain myoepithelial cells and/or adenoepithelial cells.
5. A deep learning-based pathology image analysis method, the method further comprising:
obtaining a cytopathology staining image of a sample;
detecting cells in the cytopathology staining image by using the cell detection model according to the claims 1-3 to obtain myoepithelial cells and/or adenoepithelial cells;
classifying the detected myoepithelial cells and/or glandular epithelial cells to obtain classification results of benign gonadal epithelial cells, malignant gonadal epithelial cells, benign myoepithelial cells and/or malignant myoepithelial cells;
optionally, the method further comprises predicting a result of prediction of a malignant or benign tumor based on the classification result, the result of prediction comprising a malignant or benign tumor of the glandular epithelial and/or myoepithelial cell type; optionally, the benign tumor comprises any one or more of the following benign tumors: basal cell adenoma, warthin's tumor, cystic adenoma, myoepithelial tumor, polymorphic adenoma, papillary salivary adenoma;
Optionally, the malignant tumor comprises any one or more of the following malignant tumors: mucinous epidermoid carcinoma, acinar cell carcinoma, myoepithelial carcinoma, adenoid cystic carcinoma, polymorphous adenocarcinomas, and epithelial-myoepithelial carcinoma.
6. The deep learning based pathology image analysis method according to claim 5, wherein the classification method comprises image histology and/or deep learning;
preferably, the classification method is a fusion of image histology and a deep learning method;
optionally, the fusion comprises result fusion and feature fusion, wherein the result fusion is to fuse results predicted by an image histology method and a deep learning method, and the feature fusion is to fuse histology features obtained by the image histology method and multi-scale features obtained by the deep learning method;
preferably, the result fusion method comprises hard voting fusion, weighted fusion, stacking fusion and blending fusion.
7. The deep learning based pathology image analysis method according to claim 6, wherein the image histology performs benign and malignant classification learning of myoepithelial cells and/or adenoepithelial cells by extracting nuclei, cytoplasm and related features therebetween;
Preferably, the image histology adopts an edge detection method to separate cell nuclei and cytoplasms, respectively extracts cell, cell nuclei and cytoplasms to obtain image histology characteristics, and then inputs the image histology characteristics into a classifier to obtain a classification result that the cells are benign cells or malignant cells; optionally, the image histology features include any one or more of the following features: size, shape, shade of color, uniformity, presence or absence of cavitation, and nuclear cytoplasmic ratio of the cytoplasm; optionally, the classifier includes any one or several of the following algorithms: support vector machines, decision trees, random forests, XGBoost, lightGBM;
optionally, the deep learning is implemented based on one or several of the following models: resNet, resNetXt, mobileNet, shuffleNet, squeezeNet, efficientNet, mnasNet, NFNet, alexNet, VGG, googleNet, viT, EVA;
preferably, the deep learning uses modified mobilenet v3 to learn benign and malignant classification of cells, the modified mobilenet v3 performing the following operations on the original mobilenet v3-Small model: deleting a first downsampling module and a first set of bnecks of the original model;
Introducing deformable convolution modules sensitive to shape characteristics into the last two groups of bneck of the original model;
introducing a pooling pyramid structure before pooling 7x7 of the original model, and extracting to obtain information features with different scales;
the output 1024-dimensional features are subjected to dimension reduction through a group of 1x1 dimension reduction convolution layers, so that multi-scale features are obtained; preferably, the pooling pyramid structure comprises a cavity convolution layer, a deformable convolution layer and a global pooling module, wherein the step sizes of the cavity convolution layer, the deformable convolution layer and the global pooling module are respectively 1, 3 and 5;
optionally, the deformable convolution module includes any one or several of the following sub-modules: DCNV1, DCNV2, DCNV3;
optionally, the global pooling module includes a global pooling layer, an up-sampling layer and a convolution layer;
preferably, the dimension reduction range of the dimension reduction includes 10 to 50.
8. A deep learning based pathology image analysis system, characterized in that the system comprises a computer program which, when executed, implements the deep learning based pathology image analysis method according to any one of claims 4 to 7 or implements the cell detection model construction method according to any one of claims 1 to 3.
9. A deep learning-based pathology image analysis apparatus, the apparatus comprising: a memory and a processor; the memory is used for storing program instructions; the processor is configured to invoke program instructions, which when executed, implement the deep learning-based pathology image analysis method according to any one of claims 4 to 7 or implement the cell detection model construction method according to any one of claims 1 to 3.
10. A computer-readable storage medium, on which a computer program for performing pathological image analysis is stored, characterized in that the computer program, when executed by a processor, implements the deep learning-based pathological image analysis method according to any one of claims 4 to 7 or implements the construction method of the cell detection model according to any one of claims 1 to 3.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211610039 | 2022-12-14 | ||
| CN2022116100396 | 2022-12-14 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116453114A true CN116453114A (en) | 2023-07-18 |
| CN116453114B CN116453114B (en) | 2024-03-05 |
Family
ID=87122975
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310280029.9A Active CN116453114B (en) | 2022-12-14 | 2023-03-22 | Pathological image analysis method, equipment and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116453114B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109259869A (en) * | 2018-10-12 | 2019-01-25 | 西南医科大学附属口腔医院 | A kind of CTA body surface marking auxiliary device and its application method |
| US20200258223A1 (en) * | 2018-05-14 | 2020-08-13 | Tempus Labs, Inc. | Determining biomarkers from histopathology slide images |
| CN112215801A (en) * | 2020-09-14 | 2021-01-12 | 北京航空航天大学 | Pathological image classification method and system based on deep learning and machine learning |
| CN112419295A (en) * | 2020-12-03 | 2021-02-26 | 腾讯科技（深圳）有限公司 | Medical image processing method, apparatus, computer device and storage medium |
| CN114118123A (en) * | 2021-09-18 | 2022-03-01 | 上海申挚医疗科技有限公司 | Fluorescence-stained urine exfoliated cell identification method and system |
| WO2022100034A1 (en) * | 2020-11-10 | 2022-05-19 | 广州柏视医疗科技有限公司 | Detection method for malignant region of thyroid cell pathological section based on deep learning |
| CN114565761A (en) * | 2022-02-25 | 2022-05-31 | 无锡市第二人民医院 | Deep learning-based method for segmenting tumor region of renal clear cell carcinoma pathological image |
-
2023
- 2023-03-22 CN CN202310280029.9A patent/CN116453114B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200258223A1 (en) * | 2018-05-14 | 2020-08-13 | Tempus Labs, Inc. | Determining biomarkers from histopathology slide images |
| CN109259869A (en) * | 2018-10-12 | 2019-01-25 | 西南医科大学附属口腔医院 | A kind of CTA body surface marking auxiliary device and its application method |
| CN112215801A (en) * | 2020-09-14 | 2021-01-12 | 北京航空航天大学 | Pathological image classification method and system based on deep learning and machine learning |
| WO2022100034A1 (en) * | 2020-11-10 | 2022-05-19 | 广州柏视医疗科技有限公司 | Detection method for malignant region of thyroid cell pathological section based on deep learning |
| CN112419295A (en) * | 2020-12-03 | 2021-02-26 | 腾讯科技（深圳）有限公司 | Medical image processing method, apparatus, computer device and storage medium |
| CN114118123A (en) * | 2021-09-18 | 2022-03-01 | 上海申挚医疗科技有限公司 | Fluorescence-stained urine exfoliated cell identification method and system |
| CN114565761A (en) * | 2022-02-25 | 2022-05-31 | 无锡市第二人民医院 | Deep learning-based method for segmenting tumor region of renal clear cell carcinoma pathological image |
Non-Patent Citations (5)
| Title |
|---|
| BAU PHAM ET AL.: "Cell Counting and Segmentation of Immunohistochemical Images in the Spinal Cord: Comparing Deep Learning and Traditional Approaches", 《2018 40TH ANNUAL INTERNATIONAL CONFERENCE OF THE IEEE ENGINEERING IN MEDICINE AND BIOLOGY SOCIETY (EMBC)》, 28 October 2018 (2018-10-28), pages 842 - 845 * |
| KOMURA D. ET AL.: "Machine Learning Methods for Histopathological Image Analysis", 《COMPUTATIONAL AND STRUCTURAL BIOTECHNOLOGY JOURNAL》, vol. 16, 9 February 2018 (2018-02-09), pages 34 - 42, XP055713525, DOI: 10.1016/j.csbj.2018.01.001 * |
| PARMIDA GHAHREMANI ET AL.: "Deep learning-inferred multiplex immunofluorescence for immunohistochemical image quantification", 《NATURE MACHINE INTELLIGENCE》, vol. 4, 7 April 2022 (2022-04-07), pages 401 * |
| 王荃等: "基于深度学习和组织形态分析的肺癌基因突变预测", 《生物医学工程学杂志》, vol. 37, no. 01, pages 10 - 18 * |
| 银温社等: "基于深度学习的细胞癌恶化程度预测方法研究", 《软件导刊》, vol. 17, no. 03, pages 11 - 14 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116453114B (en) | 2024-03-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Gecer et al. | Detection and classification of cancer in whole slide breast histopathology images using deep convolutional networks | |
| CN107274386B (en) | artificial intelligent auxiliary cervical cell fluid-based smear reading system | |
| Man et al. | Classification of breast cancer histopathological images using discriminative patches screened by generative adversarial networks | |
| Dundar et al. | Computerized classification of intraductal breast lesions using histopathological images | |
| Alzubaidi et al. | Robust application of new deep learning tools: an experimental study in medical imaging | |
| Kumar et al. | Convolutional neural networks for prostate cancer recurrence prediction | |
| Rachapudi et al. | Improved convolutional neural network based histopathological image classification | |
| Xu et al. | Deep learning for histopathological image analysis: Towards computerized diagnosis on cancers | |
| CN110009600A (en) | A kind of medical image area filter method, apparatus and storage medium | |
| Yu et al. | Breast cancer classification in pathological images based on hybrid features | |
| CN109389129A (en) | A kind of image processing method, electronic equipment and storage medium | |
| Pal et al. | Enhanced bag of features using alexnet and improved biogeography-based optimization for histopathological image analysis | |
| Krithiga et al. | Deep learning based breast cancer detection and classification using fuzzy merging techniques | |
| Jia et al. | Multi-layer segmentation framework for cell nuclei using improved GVF Snake model, Watershed, and ellipse fitting | |
| Zormpas-Petridis et al. | Superpixel-based conditional random fields (SuperCRF): incorporating global and local context for enhanced deep learning in melanoma histopathology | |
| Sornapudi et al. | EpithNet: Deep regression for epithelium segmentation in cervical histology images | |
| Dogar et al. | Attention augmented distance regression and classification network for nuclei instance segmentation and type classification in histology images | |
| Albayrak et al. | A hybrid method of superpixel segmentation algorithm and deep learning method in histopathological image segmentation | |
| Wang et al. | SC-dynamic R-CNN: A self-calibrated dynamic R-CNN model for lung cancer lesion detection | |
| CN114119525A (en) | Method and system for segmenting cell medical image | |
| BenTaieb et al. | Automatic diagnosis of ovarian carcinomas via sparse multiresolution tissue representation | |
| Hassan et al. | A dilated residual hierarchically fashioned segmentation framework for extracting gleason tissues and grading prostate cancer from whole slide images | |
| CN111326238A (en) | Cancer cell detection device based on sliding window | |
| Huang et al. | Recent advances in medical image processing | |
| Sambyal et al. | Recent developments in cervical cancer diagnosis using deep learning on whole slide images: An Overview of models, techniques, challenges and future directions |
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 |