CN117095815A - System for predicting prostate cancer patient with homologous recombination defect based on magnetic resonance image and pathological panoramic scanning slice - Google Patents

System for predicting prostate cancer patient with homologous recombination defect based on magnetic resonance image and pathological panoramic scanning slice Download PDF

Info

Publication number
CN117095815A
CN117095815A CN202310893003.1A CN202310893003A CN117095815A CN 117095815 A CN117095815 A CN 117095815A CN 202310893003 A CN202310893003 A CN 202310893003A CN 117095815 A CN117095815 A CN 117095815A
Authority
CN
China
Prior art keywords
module
image
hrd
model
prediction
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310893003.1A
Other languages
Chinese (zh)
Inventor
陈锐
王林辉
邓龙昕
塔娜
张倩雯
张文辉
朱焱
吴涵
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
First Affiliated Hospital of Naval Military Medical University of PLA
Original Assignee
First Affiliated Hospital of Naval Military Medical University of PLA
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by First Affiliated Hospital of Naval Military Medical University of PLA filed Critical First Affiliated Hospital of Naval Military Medical University of PLA
Priority to CN202310893003.1A priority Critical patent/CN117095815A/en
Publication of CN117095815A publication Critical patent/CN117095815A/en
Pending legal-status Critical Current

Links

Classifications

    • 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
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video 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
    • G06V10/764—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
    • G06V10/80—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
    • 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
    • G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
    • G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
    • 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
    • G16H30/40—ICT specially adapted for the handling or processing of medical images for processing medical images, e.g. editing
    • 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
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10088—Magnetic resonance imaging [MRI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30081—Prostate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30096—Tumor; Lesion

Abstract

The application provides a system for predicting a prostate cancer patient with a homologous recombination defect based on a magnetic resonance image and a pathological panoramic scanning slice, which comprises: the system comprises an input module, a quality control module and a multi-mode prediction module. The input module collects clinical information, pathological panoramic scan slice information and magnetic resonance image information of a patient. The quality control module preprocesses the received clinical information, pathological information and MRI image information. The multi-mode prediction module inputs the preprocessed prostate cancer data information of the complete three modes of the patient into the risk prediction model to predict the HRD state of the patient. According to the application, the prediction efficiency of the model is enhanced through data expansion by utilizing machine learning and image processing technology, and the problem of multi-mode data fusion is successfully solved by adopting innovative data preprocessing and feature extraction strategies, so that the economic burden of HRD evaluation is obviously reduced, the result can be provided in a shorter time, and the requirements of prostate cancer patients needing emergency treatment are met.

Description

System for predicting prostate cancer patient with homologous recombination defect based on magnetic resonance image and pathological panoramic scanning slice
Technical Field
The application relates to an artificial intelligence system for predicting the homologous recombination defect condition of prostate cancer by utilizing clinical pathology images based on multiple modes.
Background
Prostate Cancer (PCa) is a malignancy that affects mainly older men. Prostate cancer is highly incidence, advanced prostate cancer treatment and management is a significant medical challenge. According to the report of the united states "Cancer static, 2018", it was expected that in 2018, cases of new prostate Cancer in the united states were 164,690 and cases of death were 29,430. In male malignancies, morbidity and mortality are listed first and second, respectively. In China, the incidence of prostate cancer in China men is rapidly rising with the progression of aging. The annual growth rate is 13% between 2000 and 2016. In particular, once prostate cancer reaches advanced stages, its five-year survival rate is relatively low. This is mainly due to the emergence of endocrine therapy resistance. For such patients, the only targeted therapeutic approved by the FDA and NMPA in the united states in the field of prostate cancer treatment is the PAPR inhibitor o Sha Pali.
However, not all patients can obtain significant efficacy with this drug, and only those with prostate cancer that have homologous recombination defects (Homologous recombination deficiency, HRD) can benefit from it. One problem that is clinically important is how to find those patients with HRD.
The method for detecting prostate cancer HRD is mainly genome sequencing. Genomic sequencing is the process of analyzing the genetic information of tumor tissue by detecting DNA sequences within it. In the context of prostate and other cancers, genomic sequencing is particularly focused on finding variations that may be associated with tumorigenesis, progression, and therapeutic response. Because it can analyze genomic instability while detecting mutations in HRD-related genes, it can evaluate its HRD status in a variety of ways, helping physicians to determine if a patient is suitable for targeted therapy.
However, there are limitations to this approach, firstly, that some patients cannot afford the cost of HRD testing due to its high cost and technical complexity, e.g., the reporting cost of reporting these assays is currently between $ 4800 and $ 5800 as reported in the united states. Furthermore, gene sequencing may take a longer time to obtain results, which is impractical for patients in need of urgent treatment. Analysis of gold standard loss of heterozygosity (LOH) and Copy Number Variation (CNV) requires high quality tumor samples, and large fragment structural variation (LST) and other genomic instability detection require complex data analysis, also limiting its application in clinical practice. Third, accessibility is poor, and in many small urban and rural areas, high-level genetic testing is not readily available due to limited medical resources and lack of necessary expertise and equipment. Patients in remote areas may also need to travel long distances to be tested, increasing time and economic burden. These factors together affect the detection accessibility of the patient. These disorders may limit the patient from missing a potential treatment opportunity. These disorders may limit the patient from missing a potential treatment opportunity.
Other non-invasive, emerging options such as liquid biopsies are increasingly favored, but their sensitivity and specificity are to be validated, e.g. many patients have insufficient amounts of circulating tumor DNA (ctDNA), which may be below detection limits, leading to false negative results, and have not been widely used.
Thus, there is a strong need for a better-accessible HRD detection method to more accurately determine those prostate cancer patients who may benefit from targeted therapy. This will help to improve the therapeutic effect and provide more treatment options and hopes for the patient.
Disclosure of Invention
To overcome the above technical drawbacks, an object of the present application is to provide a system for predicting a prostate cancer patient with a homologous recombination defect based on magnetic resonance images and pathological panoramic scan slices, comprising: the system comprises an input module, a quality control module and a multi-mode prediction module;
the input module is used for collecting clinical information, pathological panoramic scanning slice information and magnetic resonance image information of a patient;
the segmentation quality control module comprises a clinical information quality control sub-module, a pathology panoramic scanning slice quality control sub-module and a magnetic resonance image quality control sub-module;
the clinical information quality control submodule is used for discarding clinical data with more than a certain proportion of (predefined) critical clinical information missing, checking the validity of the clinical data and calculating the complete and valid clinical data;
wherein, pathology panorama scanning section quality control submodule includes pathology section panorama scanning module, image quality evaluationAn estimation module and an image preprocessing module; the pathological section panoramic scanning module is used for scanning and converting sections of a patient into sections with 40 times magnification and 1mm 3 A high resolution digitized panoramic image of resolution; the image quality evaluation module is used for automatically detecting and removing panoramic scanning pathological section images with poor quality; the image preprocessing module is used for digitally capturing the panoramic scanning pathological section image subjected to quality screening, and marking the region of interest (Region of Interests, ROI) in a full-automatic, semi-automatic and purely manual mode; by way of example, the pathological section labeling software combines a self-defined program module to automatically label the tumor region/automatically label the tumor stroma/automatically label the gonad body so as to realize full-automatic labeling; the manual auditing is added on the basis of automatic labeling by utilizing the pathological section labeling and self-defining program module, so that semi-automatic sketching is realized; or marking the ROI by adopting a completely manual marking mode, namely adopting pathological section marking software such as Quaath and the like and adopting a pure manual sketching mode such as a polygonal, circular, magic stick and the like;
the magnetic resonance image quality control submodule comprises an image receiving module, an image quality evaluation module and an image preprocessing and lesion segmentation module; the image receiving module is used for receiving the magnetic resonance image in the DICOM format; the image quality evaluation module is used for automatically detecting and removing the image images with quality defects; the image preprocessing and lesion segmentation module is used for converting the DICOM format into a volume image, namely converting an original magnetic resonance image into a two-dimensional volume image with coordinates, so that subsequent analysis is convenient, and marking the ROI in a full-automatic, semi-automatic and purely manual mode;
the multi-mode prediction module comprises a clinical prediction sub-module, a pathology prediction sub-module, an image prediction sub-module, a multi-mode fusion module and a multi-mode output module;
the clinical prediction sub-module comprises a data filling module, a clinical characteristic HRD prediction module and a clinical result display module; the data filling module is used for filling the missing value of the patient data and filling the missing value by using the average value and the median; the clinical feature HRD prediction module is used for inputting the processed clinical information into the logistic regression algorithm to establish a risk assessment model on the basis of preprocessing and feature engineering, and calculating a clinical risk index; the clinical result display module is used for receiving the risk index obtained based on clinic and outputting display in a percentage form;
the pathology prediction sub-module comprises a block dividing module, a data enhancement and standardization module, a block-level HRD state prediction module, a pathology panorama scanning slice-level HRD state prediction module and a pathology result display module; the image block dividing module is used for processing the digitized panoramic scanning pathological section image, finely dividing the digitized panoramic scanning pathological section image into a plurality of image blocks, further screening the image blocks, and only reserving the image blocks with the overlapping degree with a region of interest (ROI) of more than 80 percent so as to improve the calculation efficiency and accuracy; the data enhancement and normalization module is used for adopting an online data enhancement technology of random horizontal and vertical overturn for the image blocks and normalizing the image blocks by performing z-score normalization on RGB channels; the HRD state prediction module of the block level is used for processing the image blocks subjected to data enhancement and standardization by adopting a pretrained convolutional neural network using a ResNet50 method, and calculating the probability of the HRD state of each block; the HRD state prediction module of the pathological panoramic scanning slice level is used for receiving HRD state probability of a picture block level as input, integrating the information by adopting a picture block probability histogram and a word bag model strategy, generating a feature vector of the pathological panoramic scanning slice level, establishing a machine learning classifier model by adopting a LightGBM method, and predicting the HRD state of the whole pathological panoramic scanning slice by using the feature vector generated by a PLH and a BoW method; the pathology result display module is used for receiving HRD state prediction of a pathology panoramic scanning slice level and displaying the HRD state prediction to a user in an intuitive mode in the form of a Grad-CAM graph and a pathology report;
the image prediction sub-module comprises an image standardization module, a feature extraction module, an HRD risk prediction module and an image result display module; the image normalization module is used for firstly sequencing pixel values in the image through the intensity cutting and normalization sub-module, cutting the intensity values to enable the intensity values to fall in the range of 0.5-99.5 percentile, and then resampling the image to the resolution of 1mm x 1mm through the spatial normalization sub-module; the feature extraction module extracts manual features required by HRD (human-machine-tool) system prediction comprising geometric features, strength features and texture features from the segmented and quality-controlled lesion areas by using a PRAD-MRI-OMICS tool constructed by the system; the HRD risk prediction module is used for establishing a machine learning model by using a Support Vector Machine (SVM) based method, predicting the HRD risk according to the selected characteristics of the model, and generating an HRD risk score for each sample by using the trained model; the image result display module is used for receiving HRD state prediction in image histology and outputting HRD risk scores in the form of percentages;
the multi-modal fusion module is used for integrating the output of the clinical result display module, the pathological result display module and the image result display module by adopting the pre-fusion, the middle fusion and the post-fusion strategies, and the comprehensive evaluation result of the HRD state of the prostate cancer is obtained by utilizing the multi-modal fusion technology.
The multi-mode output module is used for summarizing all results, outputting the total predicted HRD risk in a percentage form, simultaneously displaying the comprehensive evaluation result, the prediction results of clinic, pathology and images, and giving reference diagnosis and treatment comments.
Further, the system for predicting a prostate cancer patient with a homologous recombination defect based on the magnetic resonance image and the pathological panoramic scan slice further comprises a high-throughput digital information memory for storing the input clinical information, the pathological image, the magnetic resonance image and the prediction results thereof obtained by the multi-modal prediction module, and a computer program for storing the steps of implementing the method for predicting the HRD state of the prostate cancer, and a high-speed high-frequency data processor for implementing the steps of the method for predicting the HRD state of the prostate cancer when executing the computer management class program stored in the high-throughput digital information memory.
Further, the system for predicting the prostate cancer patient with the homologous recombination defect based on the magnetic resonance image and the pathological panoramic scanning slice further comprises a model evaluation module, a model reconstruction module and a model comparison and self-adaptive updating module; the model evaluation module is used for receiving the HRD evaluation result from the multi-mode prediction module, verifying the prediction result of the model by using known gold standard data, and then calculating various performance indexes so as to evaluate the neural network model obtained through training; the model reconstruction module is used for collecting and integrating the acquired multi-mode data in the storage module, and adjusting the model structure and parameters through complete reconstruction and partial reconstruction or manual reconstruction of the neural network, so that the model with the optimal performance is distinguished and selected as an updated HRD evaluation model; the model comparison and self-adaptive updating module is used for testing the new model by using the verification set and comparing the new model with the performance index of the original model to evaluate the advantages and disadvantages of the two models, and if the performance of the new model does not reach the preset standard or is inferior to that of the original model, the model can be retrained and optimized, or the original model can be reserved.
Further, the pre-fusion strategy is a specification-related analysis strategy, the mid-fusion strategy is a multi-core learning strategy, and the post-fusion strategy is a stacked generalization strategy.
After the technical scheme is adopted, compared with the prior art, the method has the following beneficial effects:
1. the present application provides an enhanced economy HRD assessment scheme: the present application provides an economically enhanced assessment of Homologous Recombination Defects (HRD) of prostate cancer. While conventional methods typically rely on high cost genome sequencing, the present application significantly reduces the economic burden of HRD evaluation by subtly utilizing machine learning and image processing techniques to enable the extraction of key information directly from medical images.
2. The application provides comprehensive utilization optimization of patient data: the present application actively utilizes the clinical information of the patient in optimizing the HRD diagnosis. Conventional diagnostic procedures often do not take full advantage of the potential of such information. By incorporating patient data into the training and verification links of the model, the application increases the information richness of diagnosing the HRD, thereby enhancing the accuracy and reliability of diagnosis.
3. The application enhances the prediction efficiency of the model through data expansion: the method aims at solving the problem of insufficient prediction capability of the existing model caused by insufficient training samples. By adopting the data enhancement and precise feature extraction technology, the application can continuously train the sample and the feature quantity after application, thereby having the prediction precision and the comprehensive performance of the sustainable enhancement model.
4. Precision integration of multimodal data fusion: the application successfully solves the problem of multi-mode data fusion by adopting an innovative data preprocessing and feature extraction strategy. This result is achieved by efficiently integrating data from different modalities into a unified analysis framework, thereby enabling enhancement of model predictive capabilities.
5. Enhancing the interpretability of the multi-modal model: the application obviously enhances the interpretability of the multi-modal model in the pathological layer by adopting a specific feature extraction technology and a model structure. This enhanced interpretability provides a clinician with more insight that helps them understand and trust the predictions of the model more accurately.
6. Cost effectiveness: the present application provides a more economical HRD test method and prescreening strategy that significantly reduces the cost of the test, enabling more patients to afford such tests.
7. Fast results: the application can provide results in a shorter time by optimizing and simplifying the test process, and meets the requirements of prostate cancer patients needing urgent treatment.
8. Compared with the existing genome sequencing method, the method does not need a high-quality tumor sample, so that the HRD detection is more practical and widely feasible.
9. The HRD detection method of the present application enables a physician to more personalize a treatment regimen for a prostate cancer patient because it provides more information about the patient's cancer gene signature. This enables the physician to select more targeted and effective drugs and treatment regimens for the patient.
Drawings
FIG. 1 is a block diagram of a system for predicting a prostate cancer patient with a homologous recombination defect based on magnetic resonance images and pathological panoramic scan slices in accordance with an embodiment of the present application;
FIG. 2 is a flow chart of a method for predicting a prostate cancer patient with a homologous recombination defect using the prediction system of FIG. 1;
FIG. 3 is a block diagram illustrating a multi-modal prediction module according to the present application;
FIG. 4 is a block diagram of the model evaluation and adaptive update module of the present application.
Detailed Description
Advantages of the application are further illustrated in the following description, taken in conjunction with the accompanying drawings and detailed description. It is to be understood by persons skilled in the art that the following detailed description is illustrative and not restrictive, and that this application is not limited to the details given herein.
As shown in fig. 1, the present embodiment provides a system for predicting a prostate cancer patient with a homologous recombination defect based on magnetic resonance images and pathological panoramic scan slices, comprising: the system comprises an input module, a quality control module, a multi-mode prediction module, a model evaluation and self-adaptive updating module and a storage module.
As shown in fig. 2, the method for predicting a prostate cancer patient with a homologous recombination defect using the above prediction system comprises the steps of:
step S1: clinical information, pathology information, and magnetic resonance imaging (Magnetic Resonance Imaging, MRI) image information of the patient are collected.
The input module collects clinical information, pathological panoramic scan slice information and magnetic resonance image information of a patient.
Step S2: and preprocessing the received clinical information, pathology information and MRI image information.
The quality control module preprocesses the received clinical information, pathological information and MRI image information. The quality control module comprises a clinical information quality control sub-module, a pathology panoramic scanning slice quality control sub-module and a magnetic resonance image quality control sub-module.
For clinical data, the clinical information quality control submodule operates based on complete, accurate information, e.g., if the patient's clinical information (e.g., age, PSA, TNM staging, etc.) is missing more than 60%, the modality data may be discarded. At the same time, validity checks are also performed to ensure that the clinical information is within a reasonable range.
The pathological panoramic scanning slice quality control submodule comprises a pathological section panoramic scanning module, an image quality evaluation module and an image preprocessing module aiming at pathological data. The pathological section panoramic scanning module scans the section of the patient in a scanning instrument, and uses a 40-time objective lens to scan and convert the pathological section into a section with 40-time magnification and 1mm 3 High definition digitized panoramic images of resolution. This fine resolution and large data volume allow for deep and accurate pathology analysis. The image quality assessment module automatically detects and eliminates poor quality panoramic scanned pathological slice images, such as images that are significantly folded, contaminated, lack of tumor tissue, or focus-unclear, and prompts rescanning. The image preprocessing module digitally captures the panoramic scanning pathological section image subjected to quality screening, and can perform interested region marking in a full-automatic, semi-automatic and purely manual mode, and by combining a custom program module, pathological section marking software can automatically mark a tumor region/automatically mark a tumor interstitial/automatically mark a good gonad body to realize full-automatic marking; the manual auditing is added on the basis of automatic labeling by utilizing the pathological section labeling and self-defining program module, so that semi-automatic sketching is realized; or a completely manual marking mode is adopted, namely, pathological section marking software such as Quaath and the like is adopted, and the ROI is marked by adopting a pure manual sketching mode by using methods such as polygon, circle, magic stick and the like.
The magnetic resonance image quality control submodule comprises an image receiving module, an image quality evaluation module and an image preprocessing and lesion segmentation module aiming at image data. The image receiving module is used for receiving magnetic resonance images in DICOM format from different devices. The image quality evaluation module automatically detects and eliminates poor quality image, such as blurring, excessive noise, insufficient contrast and brightness, artifacts, insufficient spatial resolution, or incomplete important structure and shielding. The image preprocessing and lesion segmentation module is used for converting the DICOM format into a volume image, namely converting an original magnetic resonance image into a two-dimensional volume image with coordinates, so that subsequent analysis is facilitated, and a full-automatic, semi-automatic and purely manual mode can be adopted for marking the interested ROI, so that subsequent processing is facilitated.
The complete three modalities of prostate cancer data information (i.e., pre-processed clinical information, pathology images, and MRI images) of the pre-processed patient is input to the multi-modality prediction module.
Step S3: and inputting the complete three-mode prostate cancer data information of the preprocessed patient into a risk prediction model to predict the HRD state of the patient.
In the multi-mode prediction module, clinical information is processed through a logistic regression model to obtain clinical risk indexes, pathology images and MRI images are processed through a series of processes (such as image segmentation, data enhancement, standardization and the like) and are processed through a deep learning model to obtain HRD state predictions on pathology and images respectively, and then outputs of the three sub-modules of the clinic, the pathology and the images are combined with each other through a post fusion technology by utilizing a fusion output module to obtain a final HRD overall evaluation result and are output in a percentage form.
As shown in fig. 3, the multi-modal prediction module includes a clinical prediction sub-module, a pathology prediction sub-module, an image prediction sub-module, a multi-modal fusion module, and a multi-modal output module;
the clinical prediction sub-module comprises a data filling module, a clinical characteristic HRD prediction module and a clinical result display module. The data filling module is used for filling the missing value of the patient data, and the average value and the median are used for filling. The clinical feature HRD prediction module is used for inputting the processed clinical information into a logistic regression (Logistic Regression, LR) algorithm to establish a risk assessment model on the basis of preprocessing and feature engineering, and calculating a clinical risk index. The clinical outcome presentation module is configured to receive a clinically derived risk index and output a presentation in percent form.
The pathology prediction sub-module comprises a block dividing module, a data enhancement and standardization module, a block-level HRD state prediction module, a pathology panorama scanning slice-level HRD state prediction module and a pathology result display module. The block dividing module is used for processing the digitized panoramic scanning pathological section image and finely dividing the digitized panoramic scanning pathological section image into a plurality of blocks, and the configurable block size comprises 256x256 pixels, 512x512 pixels and the like so as to meet different computing requirements and resolution preferences. The module further screens the tiles, retaining only tiles that overlap with the region of interest (ROI) by more than 80% to increase computational efficiency and accuracy. The data enhancement and normalization module is used to employ on-line data enhancement techniques on tiles, including random horizontal and vertical flipping. In addition, image blocks are normalized by performing z-score normalization on the RGB channels, thereby increasing the generalization ability of the model. The HRD state prediction module of the block level adopts a pre-training convolutional neural network of the ResNet50 method to process the image blocks subjected to data enhancement and standardization, and calculates the probability of the HRD state of each block. A variety of pre-trained convolutional neural networks, including res net50, VGG16, conceptionv 3, densnet, or Xception, etc., can be used to flexibly select according to different needs and preferences. These networks have different structural and performance characteristics and can be used to achieve efficient and accurate tile-level HRD state prediction. The HRD state prediction module of the pathological panoramic scan slice level is used for receiving HRD state probability of a block level as input, integrating the information by adopting a block probability histogram (Patch Likelihood Histogram, PLH) and a Bag of Words (BoW) strategy, generating a feature vector of the pathological panoramic scan slice level, establishing a machine learning classifier model by adopting a LightGBM method, and predicting the HRD state of the whole pathological panoramic scan slice by using the feature vector generated by the PLH and the BoW method. The pathology result display module is used for receiving HRD state prediction of a pathology panoramic scanning slice level and displaying the HRD state prediction to a user in an intuitive mode in the form of a Grad-CAM graph and a pathology report;
the image prediction sub-module comprises an image standardization module, a feature extraction module, an HRD risk prediction module and an image result display module. The image normalization module is used for firstly sequencing the pixel values in the image through the intensity cutting and normalization sub-module, and cutting the intensity values to enable the intensity values to fall in the range of 0.5-99.5 percentile so as to reduce the influence of abnormal points of the pixel values on the result. Then, the image is resampled to 1mm x 1mm resolution by the spatial normalization sub-module to reduce the effect of changes in voxel spacing on the results, thereby increasing the generalization ability of the model. The feature extraction module extracts the manual features required by the HRD for system prediction comprising geometric features, intensity features and texture features from the segmented and quality-controlled lesion area by using a PRAD-MRI-OMICS tool constructed by the system. The HRD risk prediction module is used for building a machine learning model by using a Support Vector Machine (SVM) based method, and predicting the HRD risk according to the selected characteristics of the model. A HRD risk score may be generated for each sample using the trained model. The image result display module is used for receiving HRD state prediction in image histology and outputting HRD risk scores in the form of percentages.
The multi-modal fusion module is used for integrating the output of the clinical result display module, the pathological result display module and the image result display module by adopting the pre-fusion, the middle fusion and the post-fusion strategies, and the comprehensive evaluation result of the HRD state of the prostate cancer is obtained by utilizing the multi-modal fusion technology.
The module is used for integrating the output of the three sub-modules and obtaining the comprehensive evaluation result of the HRD state of the prostate cancer by utilizing a multi-mode fusion technology. This module allows three main fusion strategies to be employed: pre-fusion, mid-fusion, and post-fusion. Various fusion methods are available in the fusion strategy, including "weighted average", "voting mechanism", "stacked generalization (Stacking)", "canonical correlation analysis (Canonical Correlation Analysis, CCA)", and "multi-kernel learning (Multiple Kernel Learning, MKL)". The module aims to provide a flexible and powerful framework to fully utilize information of different modalities so as to realize accurate and comprehensive evaluation of the HRD state. The user can select and adjust the proper fusion strategy and method according to the specific application scene and the data characteristics. In addition, the following recommended combination options are included in the module for selection:
pre-fusion: under this strategy it is recommended to use a "canonical correlation analysis (Canonical Correlation Analysis, CCA)" to find and integrate correlated features from different modalities. This strategy integrates the data prior to model training and is particularly useful for processing multi-modal data with high correlation.
Fusion of (2): under this strategy, "multi-kernel learning (Multiple Kernel Learning, MKL)" is recommended as a fusion method. By linearly combining the data kernels of different modalities, MKL allows the model to dynamically integrate information from different modalities during training.
Post fusion: under this strategy, it is recommended to use "stacked generalization" (Stacking) to integrate the output from each sub-module. By training a metamodel, the stack generalization can learn how to effectively combine the output of sub-modules to optimize the prediction of HRD states.
The multi-mode output module is used for outputting the fused prediction result, outputting the overall prediction HRD risk by utilizing a group of predetermined score cut-off values, and giving reference diagnosis and treatment comments. For example, when the composite score is below 0.4, the system classifies the HRD risk as very low; when the composite score is between 0.4 and 0.7, it is classified as medium risk; when the score is higher than 0.7, it is classified as high risk.
For example, three cases using a weighted average method under post fusion policy: patient a, patient B and patient C. Patient a provided only pathological section data that was analyzed to generate a score of 0.76. Since this score exceeds a cutoff value of 0.70, the system classifies patient a as HRD high risk according to the criteria described above.
Patient B provides pathology and clinical data. His pathology score was 0.65 and clinical score was 0.5. The system derives a composite score, i.e., 0.58, by calculating a simple average of these scores. Since this score lies between 0.4 and 0.7, the system classifies patient B as a medium risk in HRD.
Patient C provides pathology, clinical data, and image data. The pathology score was 0.6, the clinical score was 0.7, and the image score was 0.8. The system calculates a simple average of these scores, resulting in a composite score of 0.7. Since this score is equal to the cutoff value of 0.7, the system classifies patient C as HRD high risk.
Preferably, the storage module is a high-throughput digital information memory, and stores input clinical information, pathology images, MRI images, prediction results obtained by the multi-mode prediction module according to three features, and a computer program, and the computer program when executed by the processor realizes the steps of the method for predicting the state of the HRD for predicting the prostate cancer. Preferably, the processor is a high-speed high-frequency data processor.
Step S4: model evaluation and adaptive updating.
The model evaluation and self-adaptive updating module receives the prediction result of the multi-mode prediction module, and performs verification with known gold standard data to calculate performance indexes such as accuracy rate. The model may be reconstructed using a reconstruction module if a certain amount of patient data is collected. If the accuracy of the model does not meet the predetermined criteria, then a retraining and optimization process may be performed or the improved model may be rejected.
Specifically, as shown in fig. 4, the model evaluation and adaptive updating module includes a model evaluation module, a model reconstruction module, and a model comparison and adaptive updating module.
Model evaluation module this module is mainly responsible for evaluating the performance of the model. It uses HRD assessment results from the multimodal prediction module and uses known gold standard data such as loss of identity rearrangement (LOH), copy Number Variation (CNV) and large scale genomic instability (LST) to validate the prediction results of the model. It then calculates various performance metrics such as accuracy, recall, F1 score and area under the curve (Area Under the Curve, AUC). While including analysis of confusion matrices to gain insight into the performance of the model across different categories, as well as the assessment of sensitivity and specificity to particular problems.
Model reconstruction module this module is mainly concerned with the adjustment of model structure and parameters. The aim is to ensure that the accuracy and reliability of HRD assessment increases over time and accumulation of data.
According to different needs and conditions, different layers of reconstruction can be realized:
A. and (3) complete reconstruction: the module can acquire and integrate new multi-mode data from the storage module, then re-extract and select the characteristics, and comprehensively construct a new neural network model by using unchanged super parameters.
B. Partial reconstruction: in this case, only the end nodes of the neural network are tuned, while the previous model is taken as a pre-trained model, leaving the super-parameters unchanged. This is a faster adjustment method.
C. Manual reconstruction: on the basis of complete reconstruction, the module allows for manual adjustment of the super parameters, structural options and other settings of the model construction, enabling finer optimization adjustments. Super parameters include altering learning rate, optimizers, activation functions, etc.
The model comparison and self-adaptive updating module is used for testing the new model by using the verification set in the module and comparing the new model with performance indexes such as accuracy, AUC and the like of the original model. In this way, the merits of both models can be evaluated. If the performance of the new model does not meet the predetermined criteria or is inferior to the original model, the model may be selected for retraining and optimization or the original model may be selected for retention. The method aims at distinguishing and selecting the model with the optimal performance, and taking the model as an updated HRD evaluation model to improve the prediction effect of the model.
It should be noted that the embodiments of the present application are preferred and not limited in any way, and any person skilled in the art may make use of the above-disclosed technical content to change or modify the same into equivalent effective embodiments without departing from the technical scope of the present application, and any modification or equivalent change and modification of the above-described embodiments according to the technical substance of the present application still falls within the scope of the technical scope of the present application.

Claims (4)

1. A system for predicting a prostate cancer patient with a homologous recombination defect based on magnetic resonance images and pathological panoramic scan slices, comprising: the system comprises an input module, a quality control module and a multi-mode prediction module;
the input module is used for collecting clinical information, pathological panoramic scanning slice information and magnetic resonance image information of a patient;
the quality control module comprises a clinical information quality control sub-module, a pathology panoramic scanning slice quality control sub-module and a magnetic resonance image quality control sub-module;
the clinical information quality control submodule is used for discarding clinical data with the critical clinical information missing exceeding a certain proportion, checking the validity of the clinical data and calculating the complete and valid clinical data;
the pathological panoramic scanning slice quality control submodule comprises a pathological section panoramic scanning module, an image quality evaluation module and an image preprocessing module; the pathological section panoramic scanning module is used for scanning and converting sections of a patient into sections with 40 times magnification and 1mm 3 A high resolution digitized panoramic image of resolution; the image quality evaluation module is used for automatically detecting and removing panoramic scanning pathological section images with poor quality; the image preprocessing module is used for digitally capturing the panoramic scanning pathological section image subjected to quality screening, and marking the ROI in a full-automatic, semi-automatic and purely manual mode;
the magnetic resonance image quality control submodule comprises an image receiving module, an image quality evaluation module and an image preprocessing and lesion segmentation module; the image receiving module is used for receiving the magnetic resonance image in the DICOM format; the image quality evaluation module is used for automatically detecting and removing the image images with quality defects; the image preprocessing and lesion segmentation module is used for converting the DICOM format into a volume image, namely converting an original magnetic resonance image into a two-dimensional volume image with coordinates, and marking the ROI in a full-automatic, semi-automatic and purely manual mode;
the multi-mode prediction module comprises a clinical prediction sub-module, a pathology prediction sub-module, an image prediction sub-module, a multi-mode fusion module and a multi-mode output module;
the clinical prediction sub-module comprises a data filling module, a clinical characteristic HRD prediction module and a clinical result display module; the data filling module is used for filling the missing value of the patient data and filling the missing value by using the average value and the median; the clinical feature HRD prediction module is used for inputting the processed clinical information into the logistic regression algorithm to establish a risk assessment model on the basis of preprocessing and feature engineering, and calculating a clinical risk index; the clinical result display module is used for receiving the risk index obtained based on clinic and outputting display in a percentage form;
the pathology prediction sub-module comprises a block dividing module, a data enhancement and standardization module, a block-level HRD state prediction module, a pathology panorama scanning slice-level HRD state prediction module and a pathology result display module; the image block dividing module is used for processing the digitized panoramic scanning pathological section image, finely dividing the digitized panoramic scanning pathological section image into a plurality of image blocks, further screening the image blocks, and only reserving the image blocks with the overlapping degree with the ROI exceeding 80 percent so as to improve the calculation efficiency and accuracy; the data enhancement and normalization module is used for adopting an online data enhancement technology of random horizontal and vertical overturn for the image blocks and normalizing the image blocks by performing z-score normalization on RGB channels; the HRD state prediction module of the block level is used for processing the image blocks subjected to data enhancement and standardization by adopting a pretrained convolutional neural network using a ResNet50 method, and calculating the probability of the HRD state of each block; the HRD state prediction module of the pathological panoramic scanning slice level is used for receiving HRD state probability of a picture block level as input, integrating the information by adopting a picture block probability histogram and a word bag model strategy, generating a feature vector of the pathological panoramic scanning slice level, establishing a machine learning classifier model by adopting a LightGBM method, and predicting the HRD state of the whole pathological panoramic scanning slice by using the feature vector generated by a PLH and a BoW method; the pathology result display module is used for receiving HRD state prediction of a pathology panoramic scanning slice level and displaying the HRD state prediction to a user in an intuitive mode in the form of a Grad-CAM graph and a pathology report;
the image prediction sub-module comprises an image standardization module, a feature extraction module, an HRD risk prediction module and an image result display module; the image normalization module is used for firstly sequencing pixel values in the image through the intensity cutting and normalization sub-module, cutting the intensity values to enable the intensity values to fall in the range of 0.5-99.5 percentile, and then resampling the image to the resolution of 1mm x 1mm through the spatial normalization sub-module; the feature extraction module is used for extracting manual features required by the HRD for system prediction comprising geometric features, intensity features and texture features from the segmented and quality-controlled lesion areas by using a PRAD-MRI-OMICS tool; the HRD risk prediction module is used for establishing a machine learning model based on a support vector machine method, predicting HRD risk according to selected characteristics in the system, and generating an HRD risk score for each sample by using the trained model; the image result display module is used for receiving HRD state prediction in image histology and outputting HRD risk scores in the form of percentages;
the multi-modal fusion module is used for integrating the output of the clinical result display module, the pathological result display module and the image result display module by adopting the pre-fusion, the middle fusion and the post-fusion strategies, and the comprehensive evaluation result of the HRD state of the prostate cancer is obtained by utilizing the multi-modal fusion technology.
The multi-mode output module is used for summarizing all results, outputting the total predicted HRD risk in a percentage form, simultaneously displaying the comprehensive evaluation result, the prediction results of clinic, pathology and images, and giving reference diagnosis and treatment comments.
2. The system for predicting a prostate cancer patient with a homologous recombination defect based on magnetic resonance images and pathological panoramic scan slices as recited in claim 1, further comprising a high-throughput digital information storage for storing input clinical information, pathology images, magnetic resonance images and their predictions from the multi-modality prediction module, and a high-speed high-frequency data processor for storing a computer program for implementing the steps of the prostate cancer prediction HRD state prediction method when executing the computer management class program stored in the high-throughput digital information storage.
3. The system for predicting a prostate cancer patient with a homologous recombination defect based on magnetic resonance images and pathological panoramic scan slices of claim 2, further comprising a model evaluation module, a model reconstruction module, and a model contrast and adaptive update module; the model evaluation module is used for receiving the HRD evaluation result from the multi-mode prediction module, verifying the prediction result of the model by using known gold standard data, and then calculating various performance indexes so as to evaluate the neural network model obtained through training; the model reconstruction module is used for collecting and integrating the acquired multi-mode data in the storage module, and adjusting the model structure and parameters through complete reconstruction and partial reconstruction or manual reconstruction of the neural network, so that the model with the optimal performance is distinguished and selected as an updated HRD evaluation model; the model comparison and self-adaptive updating module is used for testing the new model by using the verification set and comparing the new model with the performance index of the original model to evaluate the advantages and disadvantages of the two models, and if the performance of the new model does not reach the preset standard or is inferior to that of the original model, the model can be retrained and optimized, or the original model can be reserved.
4. The system for predicting a prostate cancer patient with a homologous recombination defect based on magnetic resonance images and pathological panoramic scan slices as recited in claim 1, wherein the pre-fusion strategy is a canonical correlation analysis strategy, the mid-fusion strategy is a multi-kernel learning strategy, and the post-fusion strategy is a stacked generalization strategy.
CN202310893003.1A 2023-07-20 2023-07-20 System for predicting prostate cancer patient with homologous recombination defect based on magnetic resonance image and pathological panoramic scanning slice Pending CN117095815A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310893003.1A CN117095815A (en) 2023-07-20 2023-07-20 System for predicting prostate cancer patient with homologous recombination defect based on magnetic resonance image and pathological panoramic scanning slice

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310893003.1A CN117095815A (en) 2023-07-20 2023-07-20 System for predicting prostate cancer patient with homologous recombination defect based on magnetic resonance image and pathological panoramic scanning slice

Publications (1)

Publication Number Publication Date
CN117095815A true CN117095815A (en) 2023-11-21

Family

ID=88772595

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310893003.1A Pending CN117095815A (en) 2023-07-20 2023-07-20 System for predicting prostate cancer patient with homologous recombination defect based on magnetic resonance image and pathological panoramic scanning slice

Country Status (1)

Country Link
CN (1) CN117095815A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117542528A (en) * 2024-01-10 2024-02-09 中国人民解放军总医院第一医学中心 Ankylosing spondylitis hip joint affected risk marking system based on image histology
CN117612711A (en) * 2024-01-22 2024-02-27 神州医疗科技股份有限公司 Multi-mode prediction model construction method and system for analyzing liver cancer recurrence data

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117542528A (en) * 2024-01-10 2024-02-09 中国人民解放军总医院第一医学中心 Ankylosing spondylitis hip joint affected risk marking system based on image histology
CN117542528B (en) * 2024-01-10 2024-03-12 中国人民解放军总医院第一医学中心 Ankylosing spondylitis hip joint affected risk marking system based on image histology
CN117612711A (en) * 2024-01-22 2024-02-27 神州医疗科技股份有限公司 Multi-mode prediction model construction method and system for analyzing liver cancer recurrence data

Similar Documents

Publication Publication Date Title
Murugesan et al. A hybrid deep learning model for effective segmentation and classification of lung nodules from CT images
CN110197493B (en) Fundus image blood vessel segmentation method
CN104751178B (en) Lung neoplasm detection means and method based on shape template matching combining classification device
WO2018120942A1 (en) System and method for automatically detecting lesions in medical image by means of multi-model fusion
CN113011485B (en) Multi-mode multi-disease long-tail distribution ophthalmic disease classification model training method and device
CN112101451B (en) Breast cancer tissue pathological type classification method based on generation of antagonism network screening image block
CN111524137B (en) Cell identification counting method and device based on image identification and computer equipment
CN117095815A (en) System for predicting prostate cancer patient with homologous recombination defect based on magnetic resonance image and pathological panoramic scanning slice
US11257210B2 (en) Method and system of performing medical treatment outcome assessment or medical condition diagnostic
CN110348477B (en) Medical image processing method, storage medium, and computer device
WO2022110525A1 (en) Comprehensive detection apparatus and method for cancerous region
CN112071418B (en) Gastric cancer peritoneal metastasis prediction system and method based on enhanced CT image histology
CN114332132A (en) Image segmentation method and device and computer equipment
US20220148164A1 (en) System and method for automatic detection of vertebral fractures on imaging scans using deep networks
Lei et al. Automated detection of retinopathy of prematurity by deep attention network
Xiao et al. A cascade and heterogeneous neural network for CT pulmonary nodule detection and its evaluation on both phantom and patient data
US20210209755A1 (en) Automatic lesion border selection based on morphology and color features
Pavani et al. Simultaneous multiclass retinal lesion segmentation using fully automated RILBP-YNet in diabetic retinopathy
KR20220144687A (en) Dual attention multiple instance learning method
Zhang et al. Evaluation of a new dataset for visual detection of cervical precancerous lesions
WO2021183765A1 (en) Automated detection of tumors based on image processing
KR20210050790A (en) Apparatus and methods for classifying neurodegenerative diseases image of amyloid-positive based on deep-learning
JP2024027078A (en) Multi-scale whole slide pathological feature fusion extraction method, system, electronic equipment and storage medium
Mudeng et al. Domain and histopathology adaptations–based classification for malignancy grading system
CN115880245A (en) Self-supervision-based breast cancer disease classification method

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