CN113160135A - Intelligent colon lesion identification method, system and medium based on unsupervised migration image classification - Google Patents

Intelligent colon lesion identification method, system and medium based on unsupervised migration image classification Download PDF

Info

Publication number
CN113160135A
CN113160135A CN202110274834.1A CN202110274834A CN113160135A CN 113160135 A CN113160135 A CN 113160135A CN 202110274834 A CN202110274834 A CN 202110274834A CN 113160135 A CN113160135 A CN 113160135A
Authority
CN
China
Prior art keywords
colon
module
intelligent
model
sample
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
CN202110274834.1A
Other languages
Chinese (zh)
Inventor
吴庆耀
张一帆
赖吕龙
谭明奎
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.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
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 South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to CN202110274834.1A priority Critical patent/CN113160135A/en
Publication of CN113160135A publication Critical patent/CN113160135A/en
Priority to PCT/CN2021/123276 priority patent/WO2022193628A1/en
Pending legal-status Critical Current

Links

Images

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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10056—Microscopic image
    • 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
    • G06T2207/30028—Colon; Small intestine
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images

Abstract

The invention discloses a colon lesion intelligent identification method, a system and a medium based on unsupervised migration picture classification, which can be used for intelligently identifying colon lesions by constructing a model comprising two sub-network modules with the same structure, a difficulty quantification module, a domain alignment module, a noise adaptability module and a diversity module, does not need a labeled colon microscopic image sample, has high robustness to wrong labeling, and overcomes the defect that the cost is very high because the number of difficultly obtained colon microscopic images and the labeling quality are very depended in the existing colon lesion intelligent identification technology. Meanwhile, the intelligent colon lesion identification method provided by the invention is based on unsupervised transfer learning, and has the advantages of low cost, strong robustness and high flexibility.

Description

Intelligent colon lesion identification method, system and medium based on unsupervised migration image classification
Technical Field
The invention belongs to the technical field of unsupervised migration learning and intelligent medical image classification, and particularly relates to a colon lesion intelligent identification method, a colon lesion intelligent identification system and a colon lesion intelligent identification medium based on unsupervised migration image classification.
Background
In recent years, artificial intelligence and related industries have been rapidly developed and become a focus of attention in academic and industrial fields. In the field of colon pathologic change intelligent identification, colon microscopic image samples are not easy to obtain, and the difficulty in labeling the samples is very high, professional and skilled doctors are required to manually label the samples, and labeling errors inevitably exist. The existing method is very dependent on a high-quality marked colon microscopic image sample, so that the cost is huge and the method is difficult to apply to the field of real medical treatment. Therefore, how to reduce the dependence on the labeled colon microscopic image is a difficult problem to be solved urgently by intelligent colon lesion identification.
Disclosure of Invention
The invention mainly aims to overcome the defects in the prior art and provides a colon lesion intelligent identification method, a colon lesion intelligent identification system and a colon lesion intelligent identification medium based on unsupervised migration picture classification.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention provides an intelligent colon lesion identification method based on unsupervised migration picture classification, which comprises the following steps:
defining the category of the colon microscopic image of the target field; collecting and processing a source field colon digital slice image to enable the label of the source field colon digital slice image to be consistent with the category of a target field colon microscopic image;
constructing an intelligent colon lesion recognition model, which comprises the following steps: the system comprises two sub-network modules with the same structure, a difficulty quantification module, a domain alignment module, a noise adaptability module and a diversity module;
the method for training the intelligent colon lesion recognition model by using the processed source field colon digital slice image as a sample specifically comprises the following steps:
inputting the samples into the two sub-network modules to obtain classification prediction results and feature vectors of the samples;
inputting the classification prediction result of the sample into a difficulty quantization module to obtain a difficulty coefficient of the sample;
the domain alignment module, the noise adaptability module and the diversity module are used for constructing a final loss function of the intelligent colon lesion recognition model; the domain alignment module constructs domain alignment loss by using the feature vector and the difficulty coefficient of the sample; the noise adaptability module processes the prediction result by adopting a modeling artificial labeling error probability method and constructs classification loss; the diversity module adopts KL divergence to measure the similarity between the two sub-network modules and construct diversity loss; the final loss function is used for iteratively optimizing the intelligent colon lesion recognition model;
and (3) model deployment and prediction, namely inputting the target field colon microscopic image into the trained colon lesion intelligent recognition model, and predicting whether lesion occurs according to a model output result.
As a preferred technical solution, the defining the category of the target area colon microscopic image includes: normal, adenoma, adenocarcinoma, and mucinous adenocarcinoma.
As a preferred technical scheme, in the training process, the ith training sample is set as xi；
The sample passes through feature extractors of two sub-network modules to obtain a feature vector Pτ(xi) Where τ ═ {1,2} represents two subnetworks; the feature vector Pτ(xi) Obtaining classification prediction results through the classifiers of the two sub-network modules
Figure BDA00029761870700000210
As a preferred technical solution, the difficulty quantization module obtains the training sample x by using a quantization formulaiIs a difficulty coefficient λ (x)i) Specifically, the following formula:
Figure BDA0002976187070000021
wherein the content of the first and second substances,
Figure BDA0002976187070000022
the result is predicted for the ith class of both sub-network modules.
As a preferred technical solution, the domain alignment module obtains the domain alignment loss by using a re-weighting method, specifically as follows:
Figure BDA0002976187070000023
wherein d isτ(. for a domain alignment module, the probability prediction of a sample from a source domain or a target domain, S is a source domain dataset, T is a target domain dataset, nsIs the number of source domain samples, ntThe target domain sample number.
As a preferred technical solution, the processing of the prediction result by using the modeling artificial labeling error probability method specifically includes: when the model of the training stage predicts correctly and labels wrongly, the converted prediction result is consistent with the label, and the prediction stage uses the non-converted prediction result, wherein the model of the artificial labeling error probability method has the following formula:
Figure BDA0002976187070000024
Figure BDA0002976187070000025
wherein, { wkm，bkmF is a prediction result of the model on the sample;
the classification loss is specifically as follows:
Figure BDA0002976187070000026
wherein the content of the first and second substances,
Figure BDA0002976187070000027
and (3) manually marking a model of the error probability method, wherein gamma is a hyper-parameter for controlling the sample weight.
As a preferred solution, the loss of diversity is specifically represented by the following formula:
Figure BDA0002976187070000028
Figure BDA0002976187070000029
wherein D isKLIs the KL divergence.
As a preferred technical scheme, the training process adopts a gradient descent method to carry out iterative optimization; the final loss function is constructed by weighting domain alignment loss, classification loss and diversity loss, and is specifically as follows:
L＝max(-αLd)+Lc-ηLdiv，
where α is the weight of the domain alignment penalty and η is the weight of the diversity penalty.
The invention also provides an intelligent colon lesion identification system based on unsupervised migration picture classification, which is applied to the intelligent colon lesion identification method based on unsupervised migration picture classification and comprises a preprocessing module, a model construction module, a model training module and a model prediction module;
the preprocessing module is used for defining the category of the colon microscopic image in the target field; collecting and processing a colon digital slice image in a source field to enable the label of the colon digital slice image to be consistent with the category of a colon microscopic image in a target field;
the model construction module constructs an intelligent colon lesion recognition model, and the model construction module comprises: the system comprises two sub-network modules with the same structure, a difficulty quantification module, a domain alignment module, a noise adaptability module and a diversity module;
the model training module utilizes the processed source field colon digital slice image as a sample to train the intelligent colon lesion recognition model, and specifically comprises the following steps:
inputting the samples into the two sub-network modules to obtain classification prediction results and feature vectors of the samples;
inputting the classification prediction result of the sample into a difficulty quantization module to obtain a difficulty coefficient of the sample;
the domain alignment module, the noise adaptability module and the diversity module are used for constructing a final loss function of the intelligent colon lesion recognition model; the domain alignment module constructs domain alignment loss by using the feature vector and the difficulty coefficient of the sample; the noise adaptability module processes the prediction result by adopting a modeling artificial labeling error probability method and constructs classification loss; the diversity module adopts KL divergence to measure the similarity between the two sub-network modules and construct diversity loss; the final loss function is used for iteratively optimizing the intelligent colon lesion recognition model;
and the model prediction module deploys a model and predicts, inputs the target field colon microscopic image into the trained colon lesion intelligent recognition model, and predicts whether a lesion occurs according to a model output result.
The invention also provides a storage medium which stores a program, and when the program is executed by a processor, the intelligent colon lesion identification method based on unsupervised migration image classification is realized. Compared with the prior art, the invention has the following advantages and beneficial effects:
(1) the intelligent colon lesion identification method provided by the invention does not need a colon microscopic image sample with labels, has high robustness to wrong labels, uses an easily obtained colon digital slice image training model with labels, uses the trained model for colon microscopic image prediction, and overcomes the defects that the prior intelligent colon lesion identification technology depends on the quantity and the quality of the labels of the difficult-to-obtain colon microscopic images, has very high cost, greatly reduces the performance when the labels have errors and the like.
(2) The intelligent colon lesion identification method provided by the invention is based on unsupervised transfer learning, and has the advantages of low cost, strong robustness and high flexibility and activity.
Drawings
FIG. 1 is a schematic overall flow chart of a colon lesion intelligent identification method based on unsupervised migration picture classification according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a training process of an intelligent colon lesion recognition model according to an embodiment of the present invention;
FIG. 3 is a diagram illustrating overcoming the effect of false labeling according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a colon microscopic image lesion prediction according to an embodiment of the present invention;
FIG. 5 is a schematic structural diagram of a colon lesion intelligent recognition system based on unsupervised migration picture classification according to an embodiment of the present invention;
fig. 6 is a schematic structural diagram of a storage medium according to an embodiment of the present invention.
Detailed Description
In order to make the technical solutions better understood by those skilled in the art, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. It is to be understood that the embodiments described are only a few embodiments of the present application and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
Example one
As shown in fig. 1, the present embodiment provides an intelligent colon lesion identification method based on unsupervised migration image classification, which includes the following steps:
s1, defining the category of the colon microscopic image in the target field; collecting and processing a source field colon digital slice image to make the label of the source field colon digital slice image consistent with the category of a target field colon microscopic image;
more specifically, in step S1, the categories defining the target area colon microscopic image include "normal", "adenoma", "adenocarcinoma", and "mucinous adenocarcinoma".
S2, constructing an intelligent colon lesion recognition model, comprising the following steps: the system comprises two sub-network modules with the same structure, a difficulty quantification module, a domain alignment module, a noise adaptability module and a diversity module;
more specifically, in step S2, the sub-network module includes a feature extractor and a classifier; the difficulty quantization module is used for obtaining a difficulty coefficient; the domain alignment module, the noise adaptability module and the diversity module are used for constructing a final loss function of the model; the domain alignment module is equivalent to a discriminator, the two sub-network modules are equivalent to generators, and a generation countermeasure network is formed between the two parts.
S3, training an intelligent colon lesion recognition model by using the processed source field colon digital slice image as a sample, as shown in FIG. 2;
more specifically, in step S3, let the ith training sample be xi；
S3.1, passing the sample through feature extractors of two sub-network modules to obtain a feature vector Pτ(xi) Where τ ═ {1,2} represents two subnetworks; the feature vector Pτ(xi) Obtaining classification prediction results through the classifiers of the two sub-network modules
Figure BDA00029761870700000510
S3.2, inputting the classification prediction result of the sample into a difficulty quantization module, and obtaining a training sample x by adopting the quantization formula provided by the inventioniIs a difficulty coefficient λ (x)i) Specifically, the following formula:
Figure BDA0002976187070000051
wherein the content of the first and second substances,
Figure BDA0002976187070000052
predicting results for the ith classification of two sub-network modules;
s3.3, inputting the characteristic vector and the difficulty coefficient of the sample into a generation countermeasure network of a field alignment module, aligning loss by adopting the re-weighting method provided by the invention, constructing field alignment loss, and aligning a field characteristic space, wherein the following formula is specifically adopted:
Figure BDA0002976187070000053
wherein d isτ(. for a domain alignment module, the probability prediction of a sample from a source domain or a target domain, S is a source domain dataset, T is a target domain dataset, nsIs the number of source domain samples, ntFor the number of samples in the target domain
S3.4, the noise adaptability module processes the prediction result by adopting the modeling artificial labeling error probability method provided by the invention, as shown in FIG. 3, and constructs classification loss;
the method for processing the prediction result by adopting the modeling artificial labeling error probability method can reduce the damage of the artificial labeling error, and is characterized in that: when the model of the training stage predicts correctly and labels incorrectly, the converted prediction result is consistent with the label, and the prediction stage uses the non-converted prediction result, wherein the model of the artificial label error probability method has the following formula:
Figure BDA0002976187070000054
Figure BDA0002976187070000055
wherein, { wkm，bkmF is a prediction result of the model on the sample;
the classification loss is specifically as follows:
Figure BDA0002976187070000056
wherein the content of the first and second substances,
Figure BDA0002976187070000057
and (3) manually marking a model of the error probability method, wherein gamma is a hyper-parameter for controlling the weight of the model.
S3.5, the diversity module measures the similarity between the two sub-network modules by adopting KL divergence to ensure the integration effect of the two sub-networks and construct diversity loss, which is specifically as follows:
Figure BDA0002976187070000058
Figure BDA0002976187070000059
wherein D isKLIs the KL divergence.
S3.6, iteratively optimizing the intelligent colon lesion recognition model by utilizing a final loss function, wherein the training process adopts a gradient descent method for iterative optimization, and the final loss function is constructed by field alignment loss, classification loss and diversity loss in a weighting mode and is specifically as follows:
L＝max(-αLd)+Lc-ηLdiv，
where α is the weight of the domain alignment penalty and η is the weight of the diversity penalty.
S4, model deployment and prediction, as shown in figure 4, inputting the target field colon microscopic image into the trained colon lesion intelligent recognition model for prediction, and predicting whether lesion occurs according to the model output result.
As shown in fig. 5, the present embodiment provides an intelligent colon lesion identification system based on unsupervised migration image classification, which includes a preprocessing module, a model building module, a model training module, and a model prediction module;
the preprocessing module is used for defining the category of the target field colon microscopic image; collecting and processing a source field colon digital slice image to enable the label of the source field colon digital slice image to be consistent with the category of a target field colon microscopic image;
the model construction module constructs an intelligent colon lesion recognition model, and the method comprises the following steps: the system comprises two sub-network modules with the same structure, a difficulty quantization module, a domain alignment module, a noise adaptability module and a diversity module;
the model training module trains a colon lesion intelligent identification model by using the processed source field colon digital slice image as a sample, and specifically comprises the following steps:
inputting the samples into the two sub-network modules to obtain classification prediction results and feature vectors of the samples;
inputting the classification prediction result of the sample into a difficulty quantization module to obtain a difficulty coefficient of the sample;
the domain alignment module, the noise adaptability module and the diversity module are used for constructing a final loss function of the intelligent colon lesion recognition model; the domain alignment module constructs domain alignment loss by using the feature vector and the difficulty coefficient of the sample; the noise adaptability module processes the prediction result by adopting a modeling artificial labeling error probability method and constructs classification loss; the diversity module adopts KL divergence to measure the similarity between the two sub-network modules and construct diversity loss; the final loss function is used for iteratively optimizing the intelligent colon lesion recognition model;
the model prediction module is used for deploying models and predicting, inputting the target field colon microscopic images into the trained colon lesion intelligent recognition model, and predicting whether lesions occur according to the model output result.
It should be noted that, the system provided in this embodiment is only exemplified by the division of the functional modules, and in practical applications, the function allocation may be completed by different functional modules according to needs, that is, the internal structure is divided into different functional modules to complete all or part of the functions described above.
As shown in fig. 6, the present embodiment further provides a storage medium storing a program, and when the program is executed by a processor, the method for intelligently identifying a colon lesion based on unsupervised migration image classification is implemented, specifically:
s1, defining the category of the colon microscopic image in the target field; collecting and processing a source field colon digital slice image to make the label of the source field colon digital slice image consistent with the category of a target field colon microscopic image;
s2, constructing an intelligent colon lesion recognition model, comprising the following steps: the system comprises two sub-network modules with the same structure, a difficulty quantification module, a domain alignment module, a noise adaptability module and a diversity module;
s3, training a colon lesion intelligent recognition model by using the processed source field colon digital slice image as a sample, specifically:
s3.1, inputting the samples into the two sub-network modules to obtain classification prediction results and feature vectors of the samples;
s3.2, inputting the classification prediction result of the sample into a difficulty quantization module to obtain a difficulty coefficient of the sample;
s3.3, the domain alignment module, the noise adaptability module and the diversity module are used for constructing a final loss function of the intelligent colon lesion identification model; the domain alignment module constructs domain alignment loss by using the feature vector and the difficulty coefficient of the sample; the noise adaptability module processes the prediction result by adopting a modeling artificial labeling error probability method and constructs classification loss; the diversity module adopts KL divergence to measure the similarity between the two sub-network modules and construct diversity loss; the final loss function is used for iteratively optimizing a colon lesion intelligent identification model;
and S4, model deployment and prediction, wherein the target field colon microscopic image is input into the trained colon lesion intelligent recognition model, and whether lesion occurs is predicted according to the model output result.
It should be understood that portions of the present application may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or combination of the following techniques, which are known in the art, may be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application specific integrated circuit having an appropriate combinational logic gate circuit, a Programmable Gate Array (PGA), a Field Programmable Gate Array (FPGA), or the like.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.

Claims (10)

1. The intelligent colon lesion identification method based on unsupervised migration picture classification is characterized by comprising the following steps of:
defining the category of the colon microscopic image of the target field; collecting and processing a source field colon digital slice image to enable the label of the source field colon digital slice image to be consistent with the category of a target field colon microscopic image;
constructing an intelligent colon lesion recognition model, which comprises the following steps: the system comprises two sub-network modules with the same structure, a difficulty quantification module, a domain alignment module, a noise adaptability module and a diversity module;
the method for training the intelligent colon lesion recognition model by using the processed source field colon digital slice image as a sample specifically comprises the following steps:
inputting the samples into the two sub-network modules to obtain classification prediction results and feature vectors of the samples;
inputting the classification prediction result of the sample into a difficulty quantization module to obtain a difficulty coefficient of the sample;
the domain alignment module, the noise adaptability module and the diversity module are used for constructing a final loss function of the intelligent colon lesion recognition model; the domain alignment module constructs domain alignment loss by using the feature vector and the difficulty coefficient of the sample; the noise adaptability module processes the prediction result by adopting a modeling artificial labeling error probability method and constructs classification loss; the diversity module adopts KL divergence to measure the similarity between the two sub-network modules and construct diversity loss; the final loss function is used for iteratively optimizing a colon lesion intelligent identification model;
and (3) model deployment and prediction, namely inputting the target field colon microscopic image into the trained colon lesion intelligent recognition model, and predicting whether lesion occurs according to a model output result.
2. The method for intelligently identifying colon lesions based on unsupervised migration picture classification as claimed in claim 1, wherein the defining the categories of the target domain colon microscopic images comprises: normal, adenoma, adenocarcinoma, and mucinous adenocarcinoma.
3. The intelligent colon lesion recognition method based on unsupervised migration picture classification as claimed in claim 1, wherein during the training process, the ith training sample is set as xi；
The sample passes through feature extractors of two sub-network modules to obtain a feature vector Pτ(xi) Where τ ═ {1,2} represents two subnetworks; the feature vector Pτ(xi) Obtaining classification prediction results through the classifiers of the two sub-network modules
Figure FDA0002976187060000011
4. The intelligent colon lesion recognition method based on unsupervised migration picture classification as claimed in claim 3, wherein the difficulty quantification module adopts a quantification formula to obtain the training sample xiIs a difficulty coefficient λ (x)i) Specifically, the following formula:
Figure FDA0002976187060000012
wherein the content of the first and second substances,
Figure FDA0002976187060000013
the result is predicted for the ith class of both sub-network modules.
5. The method for intelligently identifying colon lesions based on unsupervised migration picture classification as claimed in claim 4, wherein said domain alignment module aligns the loss by using a re-weighting method to obtain the domain alignment loss, specifically as follows:
Figure FDA0002976187060000014
wherein d isτ(. for a domain alignment module, the probability prediction of a sample from a source domain or a target domain, S is a source domain dataset, T is a target domain dataset, nsIs the number of source domain samples, ntThe target domain sample number.
6. The method for intelligently identifying colon lesions based on unsupervised migration picture classification as claimed in claim 5, wherein said employing modeling artificial labeling error probability method to process the prediction result specifically comprises: when the model of the training stage predicts correctly and labels wrongly, the converted prediction result is consistent with the label, and the prediction stage uses the non-converted prediction result, wherein the model of the artificial labeling error probability method has the following formula:
Figure FDA0002976187060000021
Figure FDA0002976187060000022
wherein, { wkm,bkmF is a prediction result of the model on the sample;
the classification loss is specifically as follows:
Figure FDA0002976187060000023
wherein the content of the first and second substances,
Figure FDA0002976187060000024
model for manually labeling error probability methodAnd gamma is a hyper-parameter controlling the sample weight.
7. The method for intelligently identifying colon lesions based on unsupervised migration picture classification as claimed in claim 6, wherein said diversity loss is specifically represented by the following formula:
Figure FDA0002976187060000025
Figure FDA0002976187060000026
wherein D isKLIs the KL divergence.
8. The intelligent colon lesion recognition method based on unsupervised migration picture classification as claimed in claim 7, wherein the training process adopts a gradient descent method for iterative optimization; the final loss function is constructed by weighting domain alignment loss, classification loss and diversity loss, and is specifically as follows:
L＝max(-αLd)+Lc-ηLdiv，
where α is the weight of the domain alignment penalty and η is the weight of the diversity penalty.
9. The intelligent colon lesion identification system based on unsupervised migration picture classification is characterized by being applied to the intelligent colon lesion identification method based on unsupervised migration picture classification in any one of claims 1 to 8, and comprising a preprocessing module, a model construction module, a model training module and a model prediction module;
the preprocessing module is used for defining the category of the target field colon microscopic image; collecting and processing a source field colon digital slice image to enable the label of the source field colon digital slice image to be consistent with the category of a target field colon microscopic image;
the model construction module constructs an intelligent colon lesion recognition model, and the method comprises the following steps: the system comprises two sub-network modules with the same structure, a difficulty quantification module, a domain alignment module, a noise adaptability module and a diversity module;
the model training module trains a colon lesion intelligent recognition model by using the processed source field colon digital slice image as a sample, and specifically comprises the following steps:
inputting the samples into the two sub-network modules to obtain classification prediction results and feature vectors of the samples;
inputting the classification prediction result of the sample into a difficulty quantization module to obtain a difficulty coefficient of the sample;
the domain alignment module, the noise adaptability module and the diversity module are used for constructing a final loss function of the intelligent colon lesion recognition model; the domain alignment module constructs domain alignment loss by using the feature vector and the difficulty coefficient of the sample; the noise adaptability module processes the prediction result by adopting a modeling artificial labeling error probability method and constructs classification loss; the diversity module adopts KL divergence to measure the similarity between the two sub-network modules and construct diversity loss; the final loss function is used for iteratively optimizing a colon lesion intelligent identification model;
the model prediction module is used for deploying models and predicting, inputting the target field colon microscopic images into the trained colon lesion intelligent recognition model, and predicting whether lesions occur according to the model output result.
10. A storage medium storing a program, wherein the program, when executed by a processor, implements the intelligent colon lesion identification method based on unsupervised migration picture classification according to any one of claims 1 to 8.
CN202110274834.1A 2021-03-15 2021-03-15 Intelligent colon lesion identification method, system and medium based on unsupervised migration image classification Pending CN113160135A (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202110274834.1A CN113160135A (en) 2021-03-15 2021-03-15 Intelligent colon lesion identification method, system and medium based on unsupervised migration image classification
PCT/CN2021/123276 WO2022193628A1 (en) 2021-03-15 2021-10-12 Colon lesion intelligent recognition method and system based on unsupervised transfer picture classification, and medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110274834.1A CN113160135A (en) 2021-03-15 2021-03-15 Intelligent colon lesion identification method, system and medium based on unsupervised migration image classification

Publications (1)

Publication Number Publication Date
CN113160135A true CN113160135A (en) 2021-07-23

Family

ID=76887254

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110274834.1A Pending CN113160135A (en) 2021-03-15 2021-03-15 Intelligent colon lesion identification method, system and medium based on unsupervised migration image classification

Country Status (2)

Country Link
CN (1) CN113160135A (en)
WO (1) WO2022193628A1 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2022193628A1 (en) * 2021-03-15 2022-09-22 华南理工大学 Colon lesion intelligent recognition method and system based on unsupervised transfer picture classification, and medium

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108198179A (en) * 2018-01-03 2018-06-22 华南理工大学 A kind of CT medical image pulmonary nodule detection methods for generating confrontation network improvement
CN110717526A (en) * 2019-09-23 2020-01-21 华南理工大学 Unsupervised transfer learning method based on graph convolution network
AU2020103905A4 (en) * 2020-12-04 2021-02-11 Chongqing Normal University Unsupervised cross-domain self-adaptive medical image segmentation method based on deep adversarial learning

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110135579A (en) * 2019-04-08 2019-08-16 上海交通大学 Unsupervised field adaptive method, system and medium based on confrontation study
US20220309653A1 (en) * 2019-04-30 2022-09-29 The Trustees Of Dartmouth College System and method for attention-based classification of high-resolution microscopy images
CN111523621B (en) * 2020-07-03 2020-10-20 腾讯科技（深圳）有限公司 Image recognition method and device, computer equipment and storage medium
CN111860670B (en) * 2020-07-28 2022-05-17 平安科技（深圳）有限公司 Domain adaptive model training method, image detection method, device, equipment and medium
CN113160135A (en) * 2021-03-15 2021-07-23 华南理工大学 Intelligent colon lesion identification method, system and medium based on unsupervised migration image classification

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108198179A (en) * 2018-01-03 2018-06-22 华南理工大学 A kind of CT medical image pulmonary nodule detection methods for generating confrontation network improvement
CN110717526A (en) * 2019-09-23 2020-01-21 华南理工大学 Unsupervised transfer learning method based on graph convolution network
AU2020103905A4 (en) * 2020-12-04 2021-02-11 Chongqing Normal University Unsupervised cross-domain self-adaptive medical image segmentation method based on deep adversarial learning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
YIFAN YANG ET AL: ""Collaborative Unsupervised Domain Adaptation for Medical Image Diagnosis"", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2022193628A1 (en) * 2021-03-15 2022-09-22 华南理工大学 Colon lesion intelligent recognition method and system based on unsupervised transfer picture classification, and medium

Also Published As

Publication number Publication date
WO2022193628A1 (en) 2022-09-22

Similar Documents

Publication Publication Date Title
CN109949317B (en) Semi-supervised image example segmentation method based on gradual confrontation learning
CN113378632B (en) Pseudo-label optimization-based unsupervised domain adaptive pedestrian re-identification method
Li et al. Contour knowledge transfer for salient object detection
Kukleva et al. Unsupervised learning of action classes with continuous temporal embedding
US8379994B2 (en) Digital image analysis utilizing multiple human labels
Tsai et al. Image co-saliency detection and co-segmentation via progressive joint optimization
CN109711426B (en) Pathological image classification device and method based on GAN and transfer learning
CN111814854A (en) Target re-identification method adaptive to unsupervised domain
CN110717526A (en) Unsupervised transfer learning method based on graph convolution network
CN110378911B (en) Weak supervision image semantic segmentation method based on candidate region and neighborhood classifier
US20220156528A1 (en) Distance-based boundary aware semantic segmentation
WO2021034399A1 (en) Negative sampling algorithm for enhanced image classification
CN112836675B (en) Unsupervised pedestrian re-identification method and system for generating pseudo tags based on clusters
CN112990282B (en) Classification method and device for fine-granularity small sample images
CN111860823A (en) Neural network training method, neural network training device, neural network image processing method, neural network image processing device, neural network image processing equipment and storage medium
CN112488229A (en) Domain self-adaptive unsupervised target detection method based on feature separation and alignment
CN114882521A (en) Unsupervised pedestrian re-identification method and unsupervised pedestrian re-identification device based on multi-branch network
Jiang et al. A CNN model for semantic person part segmentation with capacity optimization
Patel et al. An optimized deep learning model for flower classification using nas-fpn and faster r-cnn
CN113052017A (en) Unsupervised pedestrian re-identification method based on multi-granularity feature representation and domain adaptive learning
Zhang et al. Fine-grained vehicle recognition using lightweight convolutional neural network with combined learning strategy
CN112183464A (en) Video pedestrian identification method based on deep neural network and graph convolution network
CN115841464A (en) Multi-modal brain tumor image segmentation method based on self-supervision learning
CN113160135A (en) Intelligent colon lesion identification method, system and medium based on unsupervised migration image classification
CN110647917B (en) Model multiplexing method and system

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
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20210723