CN117372416A - High-robustness digital pathological section diagnosis system and method for countermeasure training - Google Patents
High-robustness digital pathological section diagnosis system and method for countermeasure training Download PDFInfo
- Publication number
- CN117372416A CN117372416A CN202311499700.5A CN202311499700A CN117372416A CN 117372416 A CN117372416 A CN 117372416A CN 202311499700 A CN202311499700 A CN 202311499700A CN 117372416 A CN117372416 A CN 117372416A
- Authority
- CN
- China
- Prior art keywords
- feature extraction
- classification
- feature
- module
- node
- 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
- 230000001575 pathological effect Effects 0.000 title claims abstract description 132
- 238000003745 diagnosis Methods 0.000 title claims abstract description 63
- 238000012549 training Methods 0.000 title claims abstract description 21
- 238000000034 method Methods 0.000 title claims abstract description 14
- 238000000605 extraction Methods 0.000 claims abstract description 188
- 239000013598 vector Substances 0.000 claims abstract description 65
- 238000012216 screening Methods 0.000 claims description 45
- 230000007170 pathology Effects 0.000 claims description 21
- 238000004458 analytical method Methods 0.000 claims description 12
- 206010028980 Neoplasm Diseases 0.000 claims description 9
- 201000011510 cancer Diseases 0.000 claims description 9
- 230000001360 synchronised effect Effects 0.000 claims description 7
- 238000013527 convolutional neural network Methods 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 3
- 238000010219 correlation analysis Methods 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 230000009286 beneficial effect Effects 0.000 description 10
- 238000005070 sampling Methods 0.000 description 8
- 238000012937 correction Methods 0.000 description 6
- 239000000284 extract Substances 0.000 description 6
- 238000011156 evaluation Methods 0.000 description 5
- 239000003086 colorant Substances 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 201000010099 disease Diseases 0.000 description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 3
- 230000006978 adaptation Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 210000000056 organ Anatomy 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 238000003759 clinical diagnosis Methods 0.000 description 1
- 238000002405 diagnostic procedure Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000010827 pathological analysis Methods 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
Classifications
-
- 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
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- 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
- 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
- G06T2207/30024—Cell structures in vitro; Tissue sections in vitro
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Quality & Reliability (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Radiology & Medical Imaging (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a high-robustness digital pathological section diagnosis system and a method for countermeasure training, wherein the system comprises the following steps: the extraction module is used for constructing a feature extraction model and extracting feature vectors from the input first digital pathological image by utilizing feature extraction nodes of the feature extraction model; the first diagnosis module is used for carrying out classification diagnosis on the input first digital pathological image according to the feature vector based on classification nodes in the feature extraction model and calculating classification loss; the prediction module is used for carrying out field prediction according to the feature vector based on the field discrimination node in the feature extraction model and calculating field discrimination loss; the adjusting module is used for synchronously adjusting or singly adjusting parameters of the feature extraction nodes and the classification nodes according to the classification loss and the domain discrimination loss to obtain a converged feature extraction model; and the second diagnosis module is used for diagnosing the subsequent second digital pathological image by utilizing the converged feature extraction model. And higher accuracy and robustness are brought to digital pathological section diagnosis.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a high-robustness digital pathological section diagnosis system and method for countermeasure training.
Background
At present, with the rapid development of artificial intelligence technology, diagnosis of digital pathological sections by using AI assistance has become a research hotspot. The digital pathological section can help doctors to diagnose diseases more accurately, and the accuracy and efficiency of diagnosis are improved. However, in practice, pathological slices from different medical institutions often have different colors or other apparent features. These differences may lead to poor generalization ability of the AI model in the face of new data, thereby affecting its effectiveness in practical applications.
Disclosure of Invention
In view of the above-mentioned problems, the present invention provides a high-robustness digital pathological section diagnosis system and method for countermeasure training, which are used for solving the problem that pathological sections from different medical institutions in the background art often have different colors or other apparent characteristics. These differences may lead to problems of the AI model that its generalization ability is poor in the face of new data, thereby affecting its effectiveness in practical applications.
A high-robustness digital pathological section diagnostic system for countermeasure training, the system comprising:
the extraction module is used for constructing a feature extraction model and extracting feature vectors from the input first digital pathological image by utilizing feature extraction nodes of the feature extraction model;
the first diagnosis module is used for carrying out classification diagnosis on the input first digital pathological image according to the feature vector based on classification nodes in the feature extraction model and calculating classification loss;
the prediction module is used for carrying out field prediction according to the feature vector based on the field discrimination node in the feature extraction model and calculating field discrimination loss;
the adjusting module is used for synchronously adjusting or singly adjusting parameters of the feature extraction nodes and the classification nodes according to the classification loss and the domain discrimination loss, and acquiring a converged feature extraction model according to an adjusting result;
and the second diagnosis module is used for diagnosing the subsequent second digital pathological image by utilizing the converged feature extraction model.
Preferably, before the extracting module builds the feature extraction model and extracts the feature vector from the input first digital pathology image using the feature extraction node of the feature extraction model, the system is further configured to:
acquiring a plurality of first pathological section samples with data and labels and constructing a source data field according to the first pathological section samples;
obtaining a plurality of second pathological section samples with data without labels from a database of a new medical institution;
constructing a target data field according to the plurality of second pathological section samples;
and taking the first pathological section sample in the source data domain and the second pathological section sample in the target data domain as countermeasure training data of the feature extraction model.
Preferably, the extraction module comprises:
the construction submodule is used for acquiring a pre-trained convolutional neural network as a feature extraction network and constructing a feature extraction model according to the feature extraction network;
the screening sub-module is used for randomly extracting a pathological section sample from the source data field and the target data field according to preset conditions to serve as a first digital pathological image;
the input sub-module is used for inputting the first digital pathology image into the feature extraction model;
and the extraction submodule is used for starting and controlling a feature extraction node of the feature extraction model to extract feature vectors in the first digital pathological image based on a preset feature extraction rule.
Preferably, the first diagnostic module comprises:
the first determining submodule is used for starting the classification nodes, analyzing the feature vectors by using the classification nodes and acquiring vector representations of the first data pathological images according to analysis results;
a first diagnostic sub-module for performing cancer and non-cancer and cancer subtype classification diagnosis on the first data pathology image according to the vector representation of the first data pathology image;
the first computing sub-module is used for computing the classification loss of the classification nodes when the classification nodes conduct classification diagnosis on the first data pathological images based on the multi-classification cross entropy loss function.
Preferably, the prediction module includes:
the second determining submodule is used for starting the domain judging node and inputting the feature vector into the domain judging node to obtain an output result;
the prediction sub-module is used for determining that the feature vector is a stored feature vector if the output result is 0, determining that the feature vector is a source data field if the output result is 1, determining that the feature vector is not the stored feature vector, and determining that the feature vector is a target data field;
the second computing sub-module is used for computing the domain discrimination loss when the domain discrimination node predicts the domain of the first data pathological image based on the two-classification cross entropy loss function.
Preferably, the adjustment module includes:
the first comparison submodule is used for comparing the classification loss with a first preset threshold value, and generating a synchronous adjustment instruction of the parameters of the feature extraction node and the classification node if the classification loss is larger than the first preset threshold value;
the first adjusting sub-module is used for synchronously adjusting parameters of the feature extraction node and the classification node according to the parameter synchronous adjusting instruction of the feature extraction node and the classification node;
the second comparison submodule is used for comparing the domain discrimination loss with a second preset threshold value, and generating a single adjusting instruction of the feature extraction node if the domain discrimination loss is smaller than the second preset threshold value;
the second adjusting sub-module is used for carrying out single adjustment on the parameters of the feature extraction node according to the single adjusting instruction of the feature extraction node;
and the judging sub-module is used for judging whether the adjusted feature extraction model meets the convergence condition, if so, acquiring the feature extraction model after convergence, and if not, repeatedly adjusting parameters of the feature extraction node and the classification node until the convergence condition is met.
Preferably, the second diagnostic module comprises:
the extraction submodule is used for extracting pathological features irrelevant to the field in the second digital pathological image by utilizing the converged feature extraction model;
the second diagnosis sub-module is used for carrying out case diagnosis on the second digital pathological image according to the pathological characteristics and obtaining a diagnosis result;
the verification sub-module is used for determining the pathological type according to the diagnosis result, acquiring the identification factor corresponding to the pathological type, and verifying whether the diagnosis result is wrong or not based on the identification factor and the analysis index of the second digital pathological image;
and the output sub-module is used for outputting the diagnosis result in a text form if no error exists, and outputting the diagnosis error to the unit where the model is located if the error exists.
Preferably, the second adjustment submodule performs single adjustment on parameters of the feature extraction node according to a single adjustment instruction of the feature extraction node, and the second adjustment submodule includes:
acquiring rule extraction feature parameters of the feature extraction nodes, performing correlation analysis on the rule extraction feature parameters, and acquiring analysis results;
determining the common extraction probability among the extraction characteristic parameters of each rule according to the analysis result;
and adjusting the extracted characteristic parameters of each rule based on the common extraction probability to obtain the adjusted extracted characteristic parameters of the rule.
A high robustness digital pathological section diagnostic method for countermeasure training, comprising the steps of:
constructing a feature extraction model and extracting feature vectors from the input first digital pathological image by utilizing feature extraction nodes of the feature extraction model;
classifying and diagnosing the input first digital pathological image according to the feature vector based on the classification node in the feature extraction model and calculating classification loss;
performing field prediction according to the feature vector based on the field discrimination node in the feature extraction model and calculating field discrimination loss;
synchronously adjusting or singly adjusting parameters of the feature extraction nodes and the classification nodes according to the classification loss and the domain discrimination loss, and acquiring a converged feature extraction model according to an adjustment result;
and diagnosing the subsequent second digital pathological image by utilizing the converged feature extraction model.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention may be realized and attained by the structure particularly pointed out in the written description and drawings.
The technical scheme of the invention is further described in detail through the drawings and the embodiments.
Drawings
The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate the invention and together with the embodiments of the invention, serve to explain the invention.
FIG. 1 is a schematic diagram of a high-robustness digital pathological section diagnosis system for countermeasure training according to the present invention;
FIG. 2 is a schematic diagram of an extraction module in a digital pathological section diagnosis system with high robustness for countermeasure training according to the present invention;
FIG. 3 is a schematic diagram of a first diagnostic module in a high-robustness digital pathological section diagnostic system for countermeasure training according to the present invention;
fig. 4 is a workflow diagram of a high-robustness digital pathological section diagnosis method for countermeasure training according to the present invention.
Detailed Description
Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples are not representative of all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with some aspects of the present disclosure as detailed in the accompanying claims.
At present, with the rapid development of artificial intelligence technology, diagnosis of digital pathological sections by using AI assistance has become a research hotspot. The digital pathological section can help doctors to diagnose diseases more accurately, and the accuracy and efficiency of diagnosis are improved. However, in practice, pathological slices from different medical institutions often have different colors or other apparent features. These differences may lead to poor generalization ability of the AI model in the face of new data, thereby affecting its effectiveness in practical applications. In order to solve the above-mentioned problems, the present embodiment discloses a high-robustness digital pathological section diagnosis system for countermeasure training.
A high-robustness digital pathological section diagnostic system for countermeasure training, as shown in fig. 1, the system comprising:
an extraction module 101, configured to construct a feature extraction model and extract a feature vector from an input first digital pathology image by using feature extraction nodes of the feature extraction model;
a first diagnosis module 102, configured to perform classification diagnosis on the input first digital pathology image according to the feature vector based on the classification node in the feature extraction model and calculate a classification loss;
a prediction module 103, configured to perform domain prediction according to the feature vector based on the domain discrimination node in the feature extraction model and calculate a domain discrimination loss;
the adjustment module 104 is configured to perform synchronous adjustment or single adjustment on parameters of the feature extraction node and the classification node according to the classification loss and the domain discrimination loss, and obtain a converged feature extraction model according to an adjustment result;
a second diagnosis module 105, configured to diagnose a subsequent second digital pathology image using the converged feature extraction model.
In the present embodiment, the feature extraction model is represented as an intelligent model that extracts pathological features in pathological images;
in this embodiment, the feature extraction node is represented as a model work node in the feature extraction model for extracting pathological features in the image;
in this embodiment, the classification node is represented as a model work node for classifying and judging the extracted pathological feature in the feature extraction model;
in the present embodiment, the classification loss is expressed as a recognition error probability loss when performing classification judgment of the pathology type;
in this embodiment, the domain discrimination node is represented as a model work node for performing digital pathological image classification domain discrimination on the extracted pathological features in the feature extraction model;
in the present embodiment, the domain discrimination loss is expressed as a discrimination error probability loss in performing the domain discrimination of the digital pathology image classification.
The working principle of the technical scheme is as follows: firstly, constructing a feature extraction model by utilizing an extraction module, and extracting feature vectors from an input first digital pathological image by utilizing feature extraction nodes of the feature extraction model; secondly, using a first diagnosis module to carry out classification diagnosis on the input first digital pathological image according to the feature vector based on classification nodes in the feature extraction model and calculating classification loss; then, performing field prediction according to the feature vector by a prediction module based on a field discrimination node in the feature extraction model and calculating field discrimination loss; then, synchronously adjusting or singly adjusting parameters of the feature extraction nodes and the classification nodes by using an adjusting module according to the classification loss and the domain discrimination loss, and acquiring a converged feature extraction model according to an adjusting result; and finally, diagnosing the subsequent second digital pathology image by utilizing the converged feature extraction model by utilizing the second diagnosis module.
The beneficial effects of the technical scheme are as follows: the model can extract characteristics irrelevant to the field in the characteristic extraction process by calculating the judgment classification loss and judging the attribute from different fields in the characteristic extraction process of the model, so that the generalization capability of the model on data in different fields is improved, higher accuracy and robustness are brought to digital pathological section diagnosis, more reliable auxiliary tools are provided for clinical diagnosis, and the problem that pathological sections from different medical institutions in the prior art often have different colors or other apparent characteristics is solved. These differences may lead to problems of the AI model that its generalization ability is poor in the face of new data, thereby affecting its effectiveness in practical applications.
In one embodiment, before the extraction module builds a feature extraction model and extracts feature vectors from the input first digital pathology image using feature extraction nodes of the feature extraction model, the system is further configured to:
acquiring a plurality of first pathological section samples with data and labels and constructing a source data field according to the first pathological section samples;
obtaining a plurality of second pathological section samples with data without labels from a database of a new medical institution;
constructing a target data field according to the plurality of second pathological section samples;
and taking the first pathological section sample in the source data domain and the second pathological section sample in the target data domain as countermeasure training data of the feature extraction model.
The beneficial effects of the technical scheme are as follows: the countermeasure training data of the feature extraction model can be generated by constructing pathological section samples in different data fields, so that feature extraction errors of the feature extraction model in the training process can be intuitively determined, model parameters can be adaptively adjusted, and the practicability is improved.
In one embodiment, as shown in fig. 2, the extraction module 101 includes:
a construction submodule 1011 for acquiring a pre-trained convolutional neural network as a feature extraction network and constructing a feature extraction model according to the feature extraction network;
a screening submodule 1012 for randomly extracting a pathological section sample from the source data field and the target data field as a first digital pathological image according to a preset condition;
an input submodule 1013 for inputting the first digital pathology image into the feature extraction model;
an extraction submodule 1014 is used for starting and controlling a feature extraction node of the feature extraction model to extract feature vectors in the first digital pathology image based on preset feature extraction rules.
In this embodiment, the preset condition may be that the sharpness and brightness are greater than or equal to a preset threshold, and sample extraction is performed on this condition.
The beneficial effects of the technical scheme are as follows: the stability and reliability of the model can be guaranteed by using the pre-trained convolutional neural network as the feature extraction network and constructing the feature extraction model according to the feature extraction network, and further, the feature extraction nodes of the feature extraction model are started and controlled to extract feature vectors in the first digital pathological image based on the preset feature extraction rules, so that the precision control of each node of the model can be performed for related functions, the whole model is not required to be scheduled, and convenience is improved.
In this embodiment, randomly extracting a pathological section sample from a source data domain and a target data domain according to a preset condition as a first digital pathological image includes:
clustering all pathological section samples in a source data domain and a target data domain according to preset characteristic attributes to obtain clustering results;
determining the dividing weight of each characteristic attribute according to the clustering result;
constructing a sample screening model, and setting the bias weight of the sample screening model for each characteristic attribute based on the partition weight of each characteristic attribute;
acquiring general track characteristics of all pathological section samples in a source data field and a target data field and independent track characteristics of each case section sample;
generating screening basic feature set parameters of a sample screening model according to the general track features, and simultaneously generating screening auxiliary feature set parameters of the sample screening model according to the independent track features of each case slice sample;
setting sampling definition and sampling brightness as preset conditions, and carrying out random sample screening by using a sample screening model based on the preset conditions, screening basic feature group parameters and screening auxiliary feature group parameters of the sample screening model to obtain a screened sample;
performing machine vision evaluation on the screened samples, acquiring sampling characteristics according to an evaluation result, determining whether the sampling characteristics meet preset sampling requirements, if so, determining that the sample screening model is qualified, and if not, determining that the sample screening model is unqualified;
analyzing the screening sample to obtain characteristic distribution parameters, and determining random characteristic preference gain rate of the sample screening model according to the characteristic distribution parameters;
determining correction parameters of each of screening basic feature group parameters and screening auxiliary feature group parameters of the sample screening model according to the random feature preference gain rate of the sample screening model;
correcting the screening basic feature set parameters and the screening auxiliary feature set parameters of the sample screening model through the correction parameters to obtain corrected screening basic feature set parameters and screening auxiliary feature set parameters;
and randomly extracting a pathological section sample from the source data field and the target data field by using the sample screening model to serve as a first digital pathological image based on the screening basic feature set parameters of the preset condition and the adjusted sample screening model and the screening auxiliary feature set parameters.
In this embodiment, the preset feature attributes are expressed as group feature attributes of respective display contents of the digital pathology images;
in this embodiment, the dividing weight is represented as a proportion weight occupied by the group feature of the same display content in all the pathological section samples;
in the present embodiment, the sample screening model is represented as a model for intelligently screening pathological section samples;
in this embodiment, the general track features are represented as peripheral descriptive track features of diseased organs in all pathological section samples;
in this embodiment, the independent trajectory features are represented as intrinsic descriptive trajectory features of the diseased organ in each pathological section sample;
in the present embodiment, the visual evaluation is represented as an evaluation on visual effects such as image content deviation on the screening sample;
in this embodiment, the sampling feature is represented as a display feature of the screening sample;
in this embodiment, the preset sampling requirement is expressed as whether the image displayed by sampling is a complete pathological image;
in this embodiment, the feature distribution parameter is expressed as a distribution parameter of each element feature in the sampled image;
in this embodiment, the random feature preference gain rate is expressed as a screening preference gain of each feature of the sample screening model for the sample screening model when the random sample screening is performed;
in the present embodiment, the correction parameters are expressed as correction parameters for the ranges and shapes of the common track feature and the independent track feature.
The technical scheme has the beneficial effects that; through setting the characteristic deflection weight parameters and the screening characteristic group parameters of the sample screening model, the objectivity and stability of the sample screening model in random sample screening can be ensured, repeated screening of the same pathological section image due to characteristic preference is avoided, and the practicability, reliability and reference of screening data are improved.
In one embodiment, as shown in FIG. 3, the first diagnostic module 102 includes:
the first determination submodule 1021 is used for starting the classification node and analyzing the feature vector by utilizing the classification node, and acquiring the vector representation of the first data pathological image according to the analysis result;
a first diagnosis sub-module 1022 for performing cancer and non-cancer and cancer subtype classification diagnosis on the first data pathology image according to the vector representation of the first data pathology image;
the first calculation submodule 1023 is used for calculating the classification loss when the classification node performs classification diagnosis on the first data pathological image based on the multi-classification cross entropy loss function.
In this embodiment, the vector representation is an image representation of a case feature vector in the first data pathology image.
The beneficial effects of the technical scheme are as follows: the method can accurately carry out pathological diagnosis on the pathological image according to the feature vector, and simultaneously can carry out classification loss calculation in the diagnosis process, so that reference conditions are laid for subsequent model parameter correction and adjustment, and the practicability and stability are further improved.
In one embodiment, a prediction module includes:
the second determining submodule is used for starting the domain judging node and inputting the feature vector into the domain judging node to obtain an output result;
the prediction sub-module is used for determining that the feature vector is a stored feature vector if the output result is 0, determining that the feature vector is a source data field if the output result is 1, determining that the feature vector is not the stored feature vector, and determining that the feature vector is a target data field;
the second computing sub-module is used for computing the domain discrimination loss when the domain discrimination node predicts the domain of the first data pathological image based on the two-classification cross entropy loss function.
The beneficial effects of the technical scheme are as follows: the method can accurately judge the field of the pathological image according to the feature vector, and meanwhile, can judge the loss calculation in the judging process, lays reference conditions for the follow-up correction and adjustment of the model parameters, and further improves the practicability and stability.
In one embodiment, the adjustment module includes:
the first comparison submodule is used for comparing the classification loss with a first preset threshold value, and generating a synchronous adjustment instruction of the parameters of the feature extraction node and the classification node if the classification loss is larger than the first preset threshold value;
the first adjusting sub-module is used for synchronously adjusting parameters of the feature extraction node and the classification node according to the parameter synchronous adjusting instruction of the feature extraction node and the classification node;
the second comparison submodule is used for comparing the domain discrimination loss with a second preset threshold value, and generating a single adjusting instruction of the feature extraction node if the domain discrimination loss is smaller than the second preset threshold value;
the second adjusting sub-module is used for carrying out single adjustment on the parameters of the feature extraction node according to the single adjusting instruction of the feature extraction node;
and the judging sub-module is used for judging whether the adjusted feature extraction model meets the convergence condition, if so, acquiring the feature extraction model after convergence, and if not, repeatedly adjusting parameters of the feature extraction node and the classification node until the convergence condition is met.
The beneficial effects of the technical scheme are as follows: the classification loss can be minimized by synchronously adjusting the parameters of the feature extraction nodes and the classification nodes, and the field discrimination loss can be maximized by singly adjusting the parameters of the feature extraction nodes, so that the model extracts the features irrelevant to the fields on the data of different fields in the subsequent identification and feature extraction processes to ensure the lowest assimilation, and the robustness of the model is further improved.
In one embodiment, a second diagnostic module includes:
the extraction submodule is used for extracting pathological features irrelevant to the field in the second digital pathological image by utilizing the converged feature extraction model;
the second diagnosis sub-module is used for carrying out case diagnosis on the second digital pathological image according to the pathological characteristics and obtaining a diagnosis result;
the verification sub-module is used for determining the pathological type according to the diagnosis result, acquiring the identification factor corresponding to the pathological type, and verifying whether the diagnosis result is wrong or not based on the identification factor and the analysis index of the second digital pathological image;
and the output sub-module is used for outputting the diagnosis result in a text form if no error exists, and outputting the diagnosis error to the unit where the model is located if the error exists.
The beneficial effects of the technical scheme are as follows: the generalization effect of the model can be further ensured, the practicability is improved, and furthermore, the accuracy and the precision of the diagnosis result can be ensured by verifying the diagnosis result, so that the stability and the reliability are laid for disease evaluation of pathological images.
In one embodiment, the second adjustment submodule performs single adjustment on parameters of the feature extraction node according to a single adjustment instruction of the feature extraction node, including:
acquiring rule extraction feature parameters of the feature extraction nodes, performing correlation analysis on the rule extraction feature parameters, and acquiring analysis results;
determining the common extraction probability among the extraction characteristic parameters of each rule according to the analysis result;
and adjusting the extracted characteristic parameters of each rule based on the common extraction probability to obtain the adjusted extracted characteristic parameters of the rule.
The beneficial effects of the technical scheme are as follows: the self-adaptive parameter adjustment can be performed intuitively by extracting the relevance between the characteristic parameters according to various rules, so that the maximized parameter adjustment effect is ensured, and the practicability and stability are further improved.
In one embodiment, the present embodiment further discloses a high-robustness digital pathological section diagnosis method for countermeasure training, as shown in fig. 4, comprising the following steps:
s401, constructing a feature extraction model and extracting feature vectors from an input first digital pathological image by utilizing feature extraction nodes of the feature extraction model;
step S402, classifying and diagnosing the input first digital pathological image according to the feature vector based on classification nodes in the feature extraction model, and calculating classification loss;
step S403, performing field prediction according to the feature vector based on the field discrimination node in the feature extraction model and calculating field discrimination loss;
step S404, synchronously adjusting or singly adjusting parameters of the feature extraction nodes and the classification nodes according to the classification loss and the domain discrimination loss, and acquiring a converged feature extraction model according to an adjustment result;
and step S405, diagnosing the subsequent second digital pathology image by utilizing the converged feature extraction model.
The working principle and the beneficial effects of the above technical solution are described in the system claims, and are not repeated here.
It will be appreciated by those skilled in the art that the first and second aspects of the present invention refer to different phases of application.
Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any adaptations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.
It is to be understood that the present disclosure is not limited to the precise arrangements and instrumentalities shown in the drawings, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims (9)
1. A high robustness digital pathological section diagnostic system for countermeasure training, the system comprising:
the extraction module is used for constructing a feature extraction model and extracting feature vectors from the input first digital pathological image by utilizing feature extraction nodes of the feature extraction model;
the first diagnosis module is used for carrying out classification diagnosis on the input first digital pathological image according to the feature vector based on classification nodes in the feature extraction model and calculating classification loss;
the prediction module is used for carrying out field prediction according to the feature vector based on the field discrimination node in the feature extraction model and calculating field discrimination loss;
the adjusting module is used for synchronously adjusting or singly adjusting parameters of the feature extraction nodes and the classification nodes according to the classification loss and the domain discrimination loss, and acquiring a converged feature extraction model according to an adjusting result;
and the second diagnosis module is used for diagnosing the subsequent second digital pathological image by utilizing the converged feature extraction model.
2. The high robustness digital pathological section diagnostic system of claim 1, wherein prior to the extraction module constructing a feature extraction model and extracting feature vectors from the input first digital pathological image using feature extraction nodes of the feature extraction model, the system is further configured to:
acquiring a plurality of first pathological section samples with data and labels and constructing a source data field according to the first pathological section samples;
obtaining a plurality of second pathological section samples with data without labels from a database of a new medical institution;
constructing a target data field according to the plurality of second pathological section samples;
and taking the first pathological section sample in the source data domain and the second pathological section sample in the target data domain as countermeasure training data of the feature extraction model.
3. The high-robustness digital pathological section diagnostic system of claim 2, wherein the extraction module comprises:
the construction submodule is used for acquiring a pre-trained convolutional neural network as a feature extraction network and constructing a feature extraction model according to the feature extraction network;
the screening sub-module is used for randomly extracting a pathological section sample from the source data field and the target data field according to preset conditions to serve as a first digital pathological image;
the input sub-module is used for inputting the first digital pathology image into the feature extraction model;
and the extraction submodule is used for starting and controlling a feature extraction node of the feature extraction model to extract feature vectors in the first digital pathological image based on a preset feature extraction rule.
4. The high robustness digital pathological section diagnostic system of claim 1, wherein the first diagnostic module comprises:
the first determining submodule is used for starting the classification nodes, analyzing the feature vectors by using the classification nodes and acquiring vector representations of the first data pathological images according to analysis results;
a first diagnostic sub-module for performing cancer and non-cancer and cancer subtype classification diagnosis on the first data pathology image according to the vector representation of the first data pathology image;
the first computing sub-module is used for computing the classification loss of the classification nodes when the classification nodes conduct classification diagnosis on the first data pathological images based on the multi-classification cross entropy loss function.
5. The high-robustness digital pathological section diagnostic system of claim 2, wherein the prediction module comprises:
the second determining submodule is used for starting the domain judging node and inputting the feature vector into the domain judging node to obtain an output result;
the prediction sub-module is used for determining that the feature vector is a stored feature vector if the output result is 0, determining that the feature vector is a source data field if the output result is 1, determining that the feature vector is not the stored feature vector, and determining that the feature vector is a target data field;
the second computing sub-module is used for computing the domain discrimination loss when the domain discrimination node predicts the domain of the first data pathological image based on the two-classification cross entropy loss function.
6. The high robustness digital pathological section diagnostic system of claim 1, wherein the adjustment module comprises:
the first comparison submodule is used for comparing the classification loss with a first preset threshold value, and generating a synchronous adjustment instruction of the parameters of the feature extraction node and the classification node if the classification loss is larger than the first preset threshold value;
the first adjusting sub-module is used for synchronously adjusting parameters of the feature extraction node and the classification node according to the parameter synchronous adjusting instruction of the feature extraction node and the classification node;
the second comparison submodule is used for comparing the domain discrimination loss with a second preset threshold value, and generating a single adjusting instruction of the feature extraction node if the domain discrimination loss is smaller than the second preset threshold value;
the second adjusting sub-module is used for carrying out single adjustment on the parameters of the feature extraction node according to the single adjusting instruction of the feature extraction node;
and the judging sub-module is used for judging whether the adjusted feature extraction model meets the convergence condition, if so, acquiring the feature extraction model after convergence, and if not, repeatedly adjusting parameters of the feature extraction node and the classification node until the convergence condition is met.
7. The high robustness digital pathological section diagnostic system of claim 1, wherein the second diagnostic module comprises:
the extraction submodule is used for extracting pathological features irrelevant to the field in the second digital pathological image by utilizing the converged feature extraction model;
the second diagnosis sub-module is used for carrying out case diagnosis on the second digital pathological image according to the pathological characteristics and obtaining a diagnosis result;
the verification sub-module is used for determining the pathological type according to the diagnosis result, acquiring the identification factor corresponding to the pathological type, and verifying whether the diagnosis result is wrong or not based on the identification factor and the analysis index of the second digital pathological image;
and the output sub-module is used for outputting the diagnosis result in a text form if no error exists, and outputting the diagnosis error to the unit where the model is located if the error exists.
8. The high robustness digital pathological section diagnostic system of claim 1, wherein the second adjustment submodule performs a single adjustment of parameters of the feature extraction node according to a single adjustment instruction of the feature extraction node, comprising:
acquiring rule extraction feature parameters of the feature extraction nodes, performing correlation analysis on the rule extraction feature parameters, and acquiring analysis results;
determining the common extraction probability among the extraction characteristic parameters of each rule according to the analysis result;
and adjusting the extracted characteristic parameters of each rule based on the common extraction probability to obtain the adjusted extracted characteristic parameters of the rule.
9. A high robustness digital pathological section diagnosis method for countermeasure training, characterized by comprising the steps of:
constructing a feature extraction model and extracting feature vectors from the input first digital pathological image by utilizing feature extraction nodes of the feature extraction model;
classifying and diagnosing the input first digital pathological image according to the feature vector based on the classification node in the feature extraction model and calculating classification loss;
performing field prediction according to the feature vector based on the field discrimination node in the feature extraction model and calculating field discrimination loss;
synchronously adjusting or singly adjusting parameters of the feature extraction nodes and the classification nodes according to the classification loss and the domain discrimination loss, and acquiring a converged feature extraction model according to an adjustment result;
and diagnosing the subsequent second digital pathological image by utilizing the converged feature extraction model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311499700.5A CN117372416B (en) | 2023-11-13 | 2023-11-13 | High-robustness digital pathological section diagnosis system and method for countermeasure training |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311499700.5A CN117372416B (en) | 2023-11-13 | 2023-11-13 | High-robustness digital pathological section diagnosis system and method for countermeasure training |
Publications (2)
Publication Number | Publication Date |
---|---|
CN117372416A true CN117372416A (en) | 2024-01-09 |
CN117372416B CN117372416B (en) | 2024-06-14 |
Family
ID=89407709
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311499700.5A Active CN117372416B (en) | 2023-11-13 | 2023-11-13 | High-robustness digital pathological section diagnosis system and method for countermeasure training |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117372416B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN118097093A (en) * | 2024-01-29 | 2024-05-28 | 北京透彻未来科技有限公司 | System for searching images on digital pathological section data set based on pathological large model |
Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110186680A (en) * | 2019-05-30 | 2019-08-30 | 盐城工学院 | A kind of confrontation differentiation domain adapts to one-dimensional convolutional neural networks intelligent failure diagnosis method |
CN111368886A (en) * | 2020-02-25 | 2020-07-03 | 华南理工大学 | Sample screening-based label-free vehicle picture classification method |
US20200285896A1 (en) * | 2019-03-09 | 2020-09-10 | Tongji University | Method for person re-identification based on deep model with multi-loss fusion training strategy |
CN111695596A (en) * | 2020-04-30 | 2020-09-22 | 华为技术有限公司 | Neural network for image processing and related equipment |
CN113807353A (en) * | 2021-09-29 | 2021-12-17 | 中国平安人寿保险股份有限公司 | Image conversion model training method, device, equipment and storage medium |
CN114299324A (en) * | 2021-12-01 | 2022-04-08 | 万达信息股份有限公司 | Pathological image classification method and system based on multi-scale domain confrontation network |
WO2022188489A1 (en) * | 2021-03-12 | 2022-09-15 | 北京邮电大学 | Training method and apparatus for multi-mode multi-disease long-tail distribution ophthalmic disease classification model |
CN115345255A (en) * | 2022-08-31 | 2022-11-15 | 北京理工大学 | Fault diagnosis method, control device, terminal and storage medium |
CN115880538A (en) * | 2023-02-17 | 2023-03-31 | 阿里巴巴达摩院(杭州)科技有限公司 | Method and equipment for domain generalization of image processing model and image processing |
US20230281805A1 (en) * | 2022-02-18 | 2023-09-07 | Arizona Board Of Regents On Behalf Of Arizona State University | SYSTEMS, METHODS, AND APPARATUSES FOR IMPLEMENTING DISCRIMINATIVE, RESTORATIVE, AND ADVERSARIAL (DiRA) LEARNING FOR SELF-SUPERVISED MEDICAL IMAGE ANALYSIS |
CN116721343A (en) * | 2023-06-06 | 2023-09-08 | 武汉工程大学 | Cross-domain field cotton boll recognition method based on deep convolutional neural network |
-
2023
- 2023-11-13 CN CN202311499700.5A patent/CN117372416B/en active Active
Patent Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20200285896A1 (en) * | 2019-03-09 | 2020-09-10 | Tongji University | Method for person re-identification based on deep model with multi-loss fusion training strategy |
CN110186680A (en) * | 2019-05-30 | 2019-08-30 | 盐城工学院 | A kind of confrontation differentiation domain adapts to one-dimensional convolutional neural networks intelligent failure diagnosis method |
CN111368886A (en) * | 2020-02-25 | 2020-07-03 | 华南理工大学 | Sample screening-based label-free vehicle picture classification method |
CN111695596A (en) * | 2020-04-30 | 2020-09-22 | 华为技术有限公司 | Neural network for image processing and related equipment |
WO2022188489A1 (en) * | 2021-03-12 | 2022-09-15 | 北京邮电大学 | Training method and apparatus for multi-mode multi-disease long-tail distribution ophthalmic disease classification model |
CN113807353A (en) * | 2021-09-29 | 2021-12-17 | 中国平安人寿保险股份有限公司 | Image conversion model training method, device, equipment and storage medium |
CN114299324A (en) * | 2021-12-01 | 2022-04-08 | 万达信息股份有限公司 | Pathological image classification method and system based on multi-scale domain confrontation network |
US20230281805A1 (en) * | 2022-02-18 | 2023-09-07 | Arizona Board Of Regents On Behalf Of Arizona State University | SYSTEMS, METHODS, AND APPARATUSES FOR IMPLEMENTING DISCRIMINATIVE, RESTORATIVE, AND ADVERSARIAL (DiRA) LEARNING FOR SELF-SUPERVISED MEDICAL IMAGE ANALYSIS |
CN115345255A (en) * | 2022-08-31 | 2022-11-15 | 北京理工大学 | Fault diagnosis method, control device, terminal and storage medium |
CN115880538A (en) * | 2023-02-17 | 2023-03-31 | 阿里巴巴达摩院(杭州)科技有限公司 | Method and equipment for domain generalization of image processing model and image processing |
CN116721343A (en) * | 2023-06-06 | 2023-09-08 | 武汉工程大学 | Cross-domain field cotton boll recognition method based on deep convolutional neural network |
Non-Patent Citations (3)
Title |
---|
YU ZHE等: "Domain Generalization via Adversarially Learned Novel Domains", 《 IEEE ACCESS 》, 26 September 2022 (2022-09-26) * |
李琼;柏正尧;刘莹芳;: "糖尿病性视网膜图像的深度学习分类方法", 中国图象图形学报, no. 10, 16 October 2018 (2018-10-16) * |
蔡国永;林强;任凯琪;: "基于域对抗网络和BERT的跨领域文本情感分析", 山东大学学报(工学版), no. 01, 5 November 2019 (2019-11-05) * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN118097093A (en) * | 2024-01-29 | 2024-05-28 | 北京透彻未来科技有限公司 | System for searching images on digital pathological section data set based on pathological large model |
Also Published As
Publication number | Publication date |
---|---|
CN117372416B (en) | 2024-06-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11681418B2 (en) | Multi-sample whole slide image processing in digital pathology via multi-resolution registration and machine learning | |
WO2020253629A1 (en) | Detection model training method and apparatus, computer device, and storage medium | |
US11288550B2 (en) | Data processing apparatus and method, recognition apparatus, learning data storage apparatus, machine learning apparatus, and program | |
CN111814741B (en) | Method for detecting embryo-sheltered pronucleus and blastomere based on attention mechanism | |
CN109145921A (en) | A kind of image partition method based on improved intuitionistic fuzzy C mean cluster | |
CN111512322B (en) | Using neural networks | |
CN109509551A (en) | A kind of common disease intelligent diagnosing method and system | |
CN117372416B (en) | High-robustness digital pathological section diagnosis system and method for countermeasure training | |
CN111445440B (en) | Medical image analysis method, device and storage medium | |
CN111524137A (en) | Cell identification counting method and device based on image identification and computer equipment | |
CN111956212B (en) | Inter-group atrial fibrillation recognition method based on frequency domain filtering-multi-mode deep neural network | |
US9092851B2 (en) | Information processing system, information processing method, information processing apparatus, control method therefor, and storage medium storing control program therefor | |
US7606779B2 (en) | Methods and system for data aggregation of physical samples | |
WO2022057309A1 (en) | Lung feature recognition method and apparatus, computer device, and storage medium | |
CN115359873B (en) | Control method for operation quality | |
WO2020161481A1 (en) | Method and apparatus for quality prediction | |
CN106529470A (en) | Gesture recognition method based on multistage depth convolution neural network | |
CN117975170A (en) | Medical information processing method and system based on big data | |
CN117371511A (en) | Training method, device, equipment and storage medium for image classification model | |
GB2604706A (en) | System and method for diagnosing small bowel cleanliness | |
CN111598144A (en) | Training method and device of image recognition model | |
CN116805522A (en) | Diagnostic report output method, device, terminal and storage medium | |
CN115424081A (en) | Interpretable cancer pathological section image identification method and device and computer equipment | |
US11798268B2 (en) | Method for improving reliability of artificial intelligence-based object recognition using collective intelligence-based mutual verification | |
CN112785559A (en) | Bone age prediction method based on deep learning and formed by mutually combining multiple heterogeneous models |
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 |