CN114549452A - New coronary pneumonia CT image analysis method based on semi-supervised deep learning - Google Patents

New coronary pneumonia CT image analysis method based on semi-supervised deep learning Download PDF

Info

Publication number
CN114549452A
CN114549452A CN202210151409.8A CN202210151409A CN114549452A CN 114549452 A CN114549452 A CN 114549452A CN 202210151409 A CN202210151409 A CN 202210151409A CN 114549452 A CN114549452 A CN 114549452A
Authority
CN
China
Prior art keywords
training sample
classification model
image
label
deep learning
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
CN202210151409.8A
Other languages
Chinese (zh)
Inventor
张永
苏立
朱文鹏
程骋
刘振兴
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.)
Wuhan University of Science and Engineering WUSE
Original Assignee
Wuhan University of Science and Engineering WUSE
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 Wuhan University of Science and Engineering WUSE filed Critical Wuhan University of Science and Engineering WUSE
Priority to CN202210151409.8A priority Critical patent/CN114549452A/en
Publication of CN114549452A publication Critical patent/CN114549452A/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
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10081—Computed x-ray tomography [CT]
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30061—Lung

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

The invention provides a new coronary pneumonia CT image analysis method based on semi-supervised deep learning, which comprises the following steps: establishing a classification model based on a residual error neural network and adding an attention module; carrying out different data enhancement twice on each label-free training sample to obtain two new images; classifying the two images and carrying out minimum entropy processing, wherein the processed result is regarded as a pseudo label of the image; carrying out Mixup on the unlabeled training sample after the data enhancement of the pseudo label and the labeled training sample after the data enhancement to obtain a new training sample; substituting the new training sample and the corresponding label into the classification model for training and gradually removing the labeled training sample; weighting and summing the weights of the fully connected layer and the characteristic diagram in the classification model to generate an attention diagram; and diagnosing the sample images in the test set by using the trained classification model, and generating a visual map for the diagnosis result. This has practical value in practical medical applications.

Description

New coronary pneumonia CT image analysis method based on semi-supervised deep learning
Technical Field
The invention relates to a new coronary pneumonia CT image analysis method, in particular to a new coronary pneumonia CT image analysis method based on semi-supervised deep learning.
Background
One of the most critical steps in preventing and combating new coronary pneumonia is the effective screening of patients suspected of infection. In the early stages of the epidemic, reverse transcription polymerase chain reaction (RT-PCR) is commonly used to determine whether a patient is suffering from new coronary pneumonia. However, due to the rapid outbreak of an epidemic, many countries lack sufficient kits to detect suspected patients. Moreover, the RT-PCR detection needs several days to obtain a result, and the overlong detection time can cause delay of epidemic situation control and treatment. In addition, RT-PCR detection sensitivity is low, and accurate judgment can not be made by one-time detection, so that the final judgment can be made by multiple tests. In clinical practice, researchers have found that imaging features such as frosty glass shadows, multifocal lamellar lesions, etc., appear in Computed Tomography (CT) images of the breast of COVID-19 patients. But also the doctor can get the CT scan of the chest and the corresponding diagnosis more quickly compared with the RT-PCR detection. Moreover, the equipment of CT scanning is very popular in modern health care systems, so CT has become another effective method for early screening and diagnosing new coronary pneumonia.
In recent years, with the breakthrough progress of deep learning in the aspect of computer vision, the deep learning is widely applied to the fields of image classification, image positioning and detection, medical image segmentation and the like, and the burden of doctors caused by massive medical image data is greatly relieved. Most of the currently used medical image diagnosis methods are based on supervised learning, and a large amount of labeled data is needed. In many practical works, however, there may be few marked samples available because the marking of data is costly. CT acquisition and labeling of new coronary pneumonia requires a significant amount of time and effort by a professional physician, which is exacerbated during epidemic situations. Training deep learning models requires large amounts of labeled data to achieve clinically standard performance. Insufficient data volume can result in overfitting of the model, resulting in poor performance of the model. Secondly, many CT image datasets are not disclosed, since medical image data involves privacy issues for the patient. Models trained with these non-public data sets cannot be used in other hospitals.
Disclosure of Invention
The invention aims to provide a new coronary pneumonia CT image analysis method based on semi-supervised learning and attention mechanism.
The invention adopts the following technical scheme:
a new coronary pneumonia CT image analysis method based on semi-supervised deep learning is characterized by comprising the following steps:
(1) establishing a relation between the image and the class label, namely a classification model, wherein the classification model is based on a residual error neural network in deep learning and is added into an attention module;
(2) carrying out different data enhancement twice on each label-free training sample to obtain two new images;
(3) classifying the images obtained after data enhancement through the classification model trained in the step (1) to obtain classification results;
(4) carrying out minimum entropy processing on the classification result of the image after the label-free sample enhancement, and taking the processed result as a pseudo label;
(5) performing data enhancement on each labeled training sample once to obtain an enhanced image;
(6) carrying out Mixup on the unlabeled training sample after the data enhancement of the pseudo label and the labeled training sample after the data enhancement to obtain a new training sample;
(7) substituting the new training sample and the corresponding label into the classification model for training, and updating the network parameter information;
(8) in the training process, with the increase of unlabeled training samples, the labeled training samples are gradually removed, and the training signals with the supervision data are gradually released;
(9) weighting and summing the weights of the fully connected layer and the characteristic diagram in the classification model to generate an attention diagram, and highlighting an important area closely related to a prediction result;
(10) and analyzing the sample images in the test set by using the trained classification model, and generating a visual map for the analysis result.
Further, the new coronary pneumonia CT image analysis method based on semi-supervised deep learning of the invention also has the characteristics that the specific process of the step (1) is as follows: the classification model adopts a residual error neural network model in a classic image classification model, an attention mechanism module is added into the model, firstly, a feature graph extracted by the residual error neural network is respectively subjected to a global maximum pool and a global average pool based on width and height, and two feature graphs subjected to convolutional layers are obtained. Then, the output characteristics of the convolutional layer are added. The final attention map a is generated by the sigmoid function:
a ═ Sigmoid (Conv (AvgPool (F)) + Conv (MaxPool (F))), where F is the feature extracted by the residual neural network, AvgPool is the average pooling function, MaxPool is the maximum pooling function, and Conv is the convolution function.
Further, the new coronary pneumonia CT image analysis method based on semi-supervised deep learning of the invention also has the characteristics that the specific process of the step (2) is as follows: carrying out data enhancement on the label-free image twice, wherein the enhancement method comprises the following steps: normalization, geometric transformation, random adjustment of brightness, and random adjustment of contrast.
Further, the new coronary pneumonia CT image analysis method based on semi-supervised deep learning of the invention also has the characteristics that the specific process of the step (4) is as follows: and (3) carrying out minimum entropy processing on the classification result, forcing a classifier to make low entropy prediction on the unlabeled training sample, and using a sharpening function to minimize the entropy of the unlabeled data, wherein the form is as follows:
Figure BDA0003510583130000041
where p is the probability class and T is the temperatureAnd the parameter is used for adjusting the classification entropy. i is the number of samples, j represents the number from 1 to the number of categories, and L is the total number of categories.
Further, the new coronary pneumonia CT image analysis method based on semi-supervised deep learning of the invention also has the characteristics that the specific process of the step (6) is as follows: and carrying out Mixup on the unlabeled training sample after the data enhancement of the pseudo label and the labeled training sample after the data enhancement to obtain a new training sample and enhance the robustness of the classification model. Wherein the formula of mixup is as follows:
x′＝μ′x1+(1-μ′)x2
p′＝μ′p1+(1-μ′)p2
μ～Beta(α，α)
μ′＝max(μ，1-μ)
wherein x1，p1Is an image of a labeled training sample and corresponding label, x2，p2Are images of unlabeled training samples and corresponding labels. α represents the distribution parameter of Beta and μ represents the sample mixing weight.
Further, the new coronary pneumonia CT image analysis method based on semi-supervised deep learning of the invention also has the characteristics that the specific process of the step (8) is as follows: setting a threshold eta t at the t moment of training, wherein eta is more than or equal to 1/K t1, where K is the number of classes. When the probability of the correct class P of a tag instance is above a threshold ηtWhen the model deletes this instance from the penalty function, only the instances of other markers under this minipatch are trained.
The invention also provides a new coronary pneumonia CT image analysis system based on semi-supervised deep learning, which is characterized by comprising the following steps:
the deep learning module is used for establishing the relation between the image and the class label to form a classification model module, and the classification model module is based on a residual error neural network in the deep learning and is added into the attention module;
the deep learning module performs two different data enhancements on each label-free training sample to obtain two new images;
the classification model module classifies the images obtained after data enhancement to obtain classification results;
the classification model module carries out minimum entropy processing on the classification result of the image after the label-free sample enhancement, and the processed result is taken as a pseudo label;
the data enhancement module is used for carrying out data enhancement on each labeled training sample once to obtain an enhanced image;
the data mixing module is used for mixing the unlabeled training sample subjected to data enhancement and the labeled training sample subjected to data enhancement to obtain a new training sample;
the classification model module trains the new training sample and the corresponding label and updates the network parameter information; in the training process, with the increase of unlabeled training samples, the labeled training samples are gradually removed, and the training signals with the supervision data are gradually released;
and the attention map generation module is used for carrying out weighted summation on the weights of the fully connected layer and the characteristic map in the classification model module to generate an attention map and highlight the important area closely related to the prediction result.
The invention has the beneficial effects that: the invention provides a general semi-supervised deep learning method, which can promote a classifier to move towards a correct decision direction by introducing label-free samples and utilizing hidden distribution information learned from a model, thereby obtaining higher generalization and accuracy. In the field of natural image recognition, semi-supervised learning can alleviate the problem of data shortage using a small amount of labeled data and a large amount of unlabeled data.
Drawings
Fig. 1 is a flowchart of a new coronary pneumonia CT image diagnosis method based on semi-supervised learning and attention mechanism provided by the present invention.
Fig. 2 is an image example of an open source new coronary pneumonia CT image dataset.
Fig. 3 is a structural diagram of a classification model proposed by the present invention.
Fig. 4(a) is a classification index histogram of ablation learning.
Fig. 4(b) is a classification index confusion matrix for ResNet50 ablation learning.
Fig. 4(c) is a categorical index confusion matrix for ResNet50+ Attention + SSL ablation learning.
Fig. 4(d) is a classification index ROC curve of ablation learning.
Fig. 5 is a diagram of the visualization of the lesion area for better understanding the model decision.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and examples in order to make the objects, aspects and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other. The process of the invention is illustrated in fig. 1 and 3, and comprises the following steps:
(1) and establishing a relation between the image and the class label, namely a classification model, wherein the classification model is based on a residual error neural network in deep learning, and adding an attention module on the basis of the classification model.
The classification model adopts a residual error neural network model in a classical image classification model, and an attention mechanism module is added in the model, so that the capability of the model for extracting features is enhanced. Firstly, the feature maps extracted by the residual neural network are respectively subjected to global maximum pooling and global average pooling based on width and height to obtain two feature maps subjected to convolutional layers. Then, the output characteristics of the convolutional layer are added. Finally, the attention map a is generated by the sigmoid function: a ═ Sigmoid (Conv (AvgPool (F)) + Conv (MaxPool (F))), where F is the feature extracted by the residual neural network, AvgPool is the average pooling function, MaxPool is the maximum pooling function, and Conv is the convolution function.
(2) Two different data enhancements are performed on each unlabeled training sample to obtain two new images. Data enhancement methods commonly used in images generally include: normalization, geometric transformation (translation, flipping, rotation), random adjustment of brightness, random adjustment of contrast, etc.
(3) And classifying the images obtained after the data enhancement through the classification model trained in the last stage to obtain the classification results.
(4) And performing minimum entropy processing on the classification result of the image after the label-free sample enhancement, and regarding the processed result as a pseudo label. And carrying out minimum entropy processing on the classification result, and forcing a classifier to make low entropy prediction on the unlabeled training sample, wherein the entropy of the unlabeled data is minimized by using a sharpening function in the invention in the following form:
Figure BDA0003510583130000071
where p is the probability class and T is the temperature parameter, which is used to adjust the classification entropy. i is the number of samples, j represents the number from 1 to the number of categories, and L is the total number of categories.
(5) And performing data enhancement on each labeled training sample once to obtain an enhanced image.
(6) And carrying out Mixup on the unlabeled training sample after the data enhancement of the pseudo label and the labeled training sample after the data enhancement to obtain a new training sample. And the robustness of the classification model is enhanced. Wherein the formula of mixup is as follows:
x′＝μ′x1+(1-μ′)x2
p′＝μ′p1+(1-μ′)p2
μ～Beta(α，α)
μ′＝max(μ，1-μ)
wherein x1，p1Is an image of a labeled training sample and corresponding label, x2，p2Are images of unlabeled training samples and corresponding labels. α represents the distribution parameter of Beta and μ represents the sample mixing weight.
(7) And substituting the new training sample and the corresponding label into the classification model for training, and updating the network parameter information.
(8) In the training process, along with the increase of unlabeled training samples, the labeled training samples are gradually removed, and the training signals of the supervised data are gradually released. Setting a threshold eta t at the t moment of training, wherein eta is more than or equal to 1/KtLess than or equal to 1, wherein,k is the number of classes. When the probability of a label instance being correct for the class P is above a threshold ηtWhen the model deletes this instance from the loss function, only the instances of other markers under this minimatch are trained. Threshold ηtTo prevent over-fitting of the model to the tag data. With ηtApproaching 1, the model can only be slowly supervised from the labeled instances, greatly alleviating the overfitting problem.
(9) And carrying out weighted summation on the weights of the fully connected layer and the characteristic diagram in the classification model to generate an attention diagram, and highlighting an important area closely related to a prediction result.
(10) And analyzing the sample images in the test set by using the trained classification model, and generating a visual map for the analysis result.
The embodiment adopts the method for the CT diagnosis of the new coronary pneumonia based on the semi-supervised learning and the attention mechanism, and the verification is carried out through the CT image public data set of the new coronary pneumonia.
The use of a labeled CT dataset and an unlabeled CT dataset to evaluate the proposed method for new coronary pneumonia CT image diagnosis. The labeled CT dataset is a new coronary pneumonia public dataset containing 349 positive and 397 negative CT scans. Fig. 3 shows an example of a CT image of new coronary pneumonia. The data set partitioning is shown in table 1.
Table 1: division of new coronary pneumonia CT image data set
Class Training set Verification set Test set
New coronary pneumonia 191 60 98
Is normal 234 58 105
Total of 425 118 203
The positive samples were 760 COVID-19 preprints from medRxiv and bioRxiv, and the negative samples were CT scans of normal persons or other types of disease. The unlabeled sample dataset is from the LUNA dataset, which consists of low-dose lung CT images, designed specifically for patient detection and segmentation of lung nodules. And selecting 500 samples as unlabeled samples to be added into a training set, and performing semi-supervised learning on the images as the unlabeled images.
To fully understand the effect of each part of the method provided by the present invention, the following ablation studies were conducted, divided into four cases: (1) using a residual neural network alone; (2) using a residual neural network plus an attention mechanism; (3) using a residual neural network plus semi-supervised learning; (4) a residual neural network plus attention mechanism plus semi-supervised learning is used.
First, it can be seen from fig. 4(a) to 4(d) that the model with the attention module performs better than the model without the attention module. This illustrates that the proposed attention module can ensure that the decision of the model is mainly dependent on the infected area and suppress the contribution of irrelevant parts in the image, thereby improving the performance of the model. Second, it is apparent from this table that the performance of the model is improved when semi-supervised learning is used. This indicates that semi-supervised learning can improve the generalization ability of the model by expanding the data set, reducing the risk of over-fitting the model on smaller data sets. Both models achieve the best performance when using both attention module and semi-supervised learning. Third, when the semi-supervised learning and attention module is used alone, the specificity of the model is reduced, which suggests that some of the advantages of the model may be sacrificed in order to meet specific needs.
Fig. 5 shows a visualization of the classification results for the baseline and model. The first column represents the original new coronary pneumonia CT image. The second three columns of the figure show the results of using the residual neural network alone. The colors from deep red to deep blue correspond to values of the pixel's class saliency from large to small. The fourth fifth column shows the results of the proposed model. By comparing the baseline results with the results of the present invention, we observed that the proposed model can capture almost all salient regions of the prediction.
The present embodiment also provides a new coronary pneumonia CT image analysis system based on semi-supervised deep learning, including:
the deep learning module is used for establishing the relation between the image and the class label to form a classification model module, and the classification model module is based on a residual error neural network in the deep learning and is added into the attention module;
the deep learning module performs two different data enhancements on each label-free training sample to obtain two new images;
the classification model module classifies the images obtained after data enhancement to obtain classification results;
the classification model module carries out minimum entropy processing on the classification result of the image after the label-free sample enhancement, and the processed result is taken as a pseudo label;
the data enhancement module is used for carrying out data enhancement on each labeled training sample once to obtain an enhanced image;
the data mixing module is used for mixing the unlabeled training sample subjected to data enhancement and the labeled training sample subjected to data enhancement to obtain a new training sample;
the classification model module trains the new training sample and the corresponding label and updates the network parameter information; in the training process, with the increase of unlabeled training samples, the labeled training samples are gradually removed, and the training signals with the supervision data are gradually released;
and the attention map generation module is used for carrying out weighted summation on the weights of the fully connected layer and the characteristic map in the classification model module to generate an attention map and highlight the important area closely related to the prediction result.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (7)

1. A new coronary pneumonia CT image analysis method based on semi-supervised deep learning is characterized by comprising the following steps:
(1) establishing a relation between the image and the class label, namely a classification model, wherein the classification model is based on a residual error neural network in deep learning and is added into an attention module;
(2) carrying out different data enhancement twice on each label-free training sample to obtain two new images;
(3) classifying the images obtained after data enhancement through the classification model trained in the step (1) to obtain classification results;
(4) carrying out minimum entropy processing on the classification result of the image after the label-free sample enhancement, and taking the processed result as a pseudo label;
(5) performing data enhancement on each labeled training sample once to obtain an enhanced image;
(6) carrying out Mixup on the unlabeled training sample after the data enhancement of the pseudo label and the labeled training sample after the data enhancement to obtain a new training sample;
(7) substituting the new training sample and the corresponding label into the classification model for training, and updating the network parameter information;
(8) in the training process, with the increase of unlabeled training samples, the labeled training samples are gradually removed, and the training signals with the supervision data are gradually released;
(9) weighting and summing the weights of the fully connected layer and the characteristic diagram in the classification model to generate an attention diagram, and highlighting an important area closely related to a prediction result;
(10) and analyzing the sample images in the test set by using the trained classification model, and generating a visual map for the analysis result.
2. The method for analyzing the new coronary pneumonia CT image based on semi-supervised deep learning as claimed in claim 1, wherein the specific process of the step (1) is as follows: the classification model adopts a residual error neural network model in a classic image classification model, an attention mechanism module is added into the model, firstly, a feature graph extracted by the residual error neural network is respectively subjected to a global maximum pool and a global average pool based on width and height to obtain two feature graphs subjected to convolution layers, and then, the output features of the convolution layers are added. The final attention map a is generated by the sigmoid function: a ═ Sigmoid (Conv (AvgPool (F)) + Conv (MaxPool (F))), where F is the feature extracted by the residual neural network, AvgPool is the average pooling function, MaxPool is the maximum pooling function, and Conv is the convolution function.
3. The method for analyzing the new coronary pneumonia CT image based on semi-supervised deep learning as claimed in claim 1, wherein the specific process of the step (2) is as follows: carrying out data enhancement on the unlabeled image twice, wherein the enhancement method comprises the following steps: normalization, geometric transformation, random adjustment of brightness, and random adjustment of contrast.
4. The method for analyzing the new coronary pneumonia CT image based on semi-supervised deep learning as claimed in claim 1, wherein the specific process of the step (4) is as follows: and (3) carrying out minimum entropy processing on the classification result, forcing a classifier to make low entropy prediction on the unlabeled training sample, and using a sharpening function to minimize the entropy of the unlabeled data, wherein the form is as follows:
Figure FDA0003510583120000021
where p is the probability class and T is the temperature parameter, which is used to adjust the classification entropy. i is the number of samples, j represents the number from 1 to the number of categories, and L is the total number of categories.
5. The method for analyzing the new coronary pneumonia CT image based on semi-supervised deep learning as claimed in claim 1, wherein the specific process of the step (6) is as follows: and carrying out Mixup on the unlabeled training sample after the data enhancement of the pseudo label and the labeled training sample after the data enhancement to obtain a new training sample and enhance the robustness of the classification model. Wherein the formula of mixup is as follows:
x′＝μ′x1+(1-μ′)x2
p′＝μ′p1+(1-μ′)p2
μ～Beta(α，α)
μ′＝max(μ，1-μ)
wherein x1，p1Is an image of a labeled training sample and corresponding label, x2，p2The image of the unlabeled training sample and the corresponding label are shown, alpha represents the distribution parameter of Beta, and mu represents the sample mixing weight.
6. The method for analyzing the new coronary pneumonia CT image based on semi-supervised deep learning as claimed in claim 1, wherein the specific process of the step (8) is as follows: at time t of training, a threshold eta is settAnd 1/K is not more than etat1, where K is the number of classes. When the probability of the correct class P of a tag instance is above a threshold ηtWhen the model deletes this instance from the loss function, only the instances of other markers under this minimatch are trained.
7. A new coronary pneumonia CT image analysis system based on semi-supervised deep learning is characterized by comprising:
the deep learning module is used for establishing the relation between the image and the class label to form a classification model module, and the classification model module is based on a residual error neural network in the deep learning and is added into the attention module;
the deep learning module performs two different data enhancements on each label-free training sample to obtain two new images;
the classification model module classifies the images obtained after the data enhancement to obtain classification results;
the classification model module carries out minimum entropy processing on the classification result of the image after the label-free sample enhancement, and the processed result is taken as a pseudo label;
the data enhancement module is used for carrying out data enhancement on each labeled training sample once to obtain an enhanced image;
the data mixing module is used for mixing the unlabeled training sample subjected to data enhancement and the labeled training sample subjected to data enhancement to obtain a new training sample;
the classification model module trains the new training sample and the corresponding label and updates the network parameter information; in the training process, with the increase of unlabeled training samples, the labeled training samples are gradually removed, and the training signals with the supervision data are gradually released;
and the attention map generation module is used for carrying out weighted summation on the weights of the fully connected layer and the characteristic map in the classification model module to generate an attention map and highlight the important area closely related to the prediction result.
CN202210151409.8A 2022-02-18 2022-02-18 New coronary pneumonia CT image analysis method based on semi-supervised deep learning Pending CN114549452A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210151409.8A CN114549452A (en) 2022-02-18 2022-02-18 New coronary pneumonia CT image analysis method based on semi-supervised deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210151409.8A CN114549452A (en) 2022-02-18 2022-02-18 New coronary pneumonia CT image analysis method based on semi-supervised deep learning

Publications (1)

Publication Number Publication Date
CN114549452A true CN114549452A (en) 2022-05-27

Family

ID=81674670

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210151409.8A Pending CN114549452A (en) 2022-02-18 2022-02-18 New coronary pneumonia CT image analysis method based on semi-supervised deep learning

Country Status (1)

Country Link
CN (1) CN114549452A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116523840A (en) * 2023-03-30 2023-08-01 苏州大学 Lung CT image detection system and method based on deep learning
CN117095241A (en) * 2023-10-17 2023-11-21 四川大学 Screening method, system, equipment and medium for drug-resistant phthisis class
CN118015358A (en) * 2024-02-04 2024-05-10 南京大学 Unsupervised domain adaptation method for chest X-ray image pulmonary tuberculosis diagnosis

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116523840A (en) * 2023-03-30 2023-08-01 苏州大学 Lung CT image detection system and method based on deep learning
CN116523840B (en) * 2023-03-30 2024-01-16 苏州大学 Lung CT image detection system and method based on deep learning
CN117095241A (en) * 2023-10-17 2023-11-21 四川大学 Screening method, system, equipment and medium for drug-resistant phthisis class
CN117095241B (en) * 2023-10-17 2024-01-12 四川大学 Screening method, system, equipment and medium for drug-resistant phthisis class
CN118015358A (en) * 2024-02-04 2024-05-10 南京大学 Unsupervised domain adaptation method for chest X-ray image pulmonary tuberculosis diagnosis

Similar Documents

Publication Publication Date Title
Jiang et al. An automatic detection system of lung nodule based on multigroup patch-based deep learning network
Varela-Santos et al. A new modular neural network approach with fuzzy response integration for lung disease classification based on multiple objective feature optimization in chest X-ray images
CN114549452A (en) New coronary pneumonia CT image analysis method based on semi-supervised deep learning
Rajee et al. Gender classification on digital dental x-ray images using deep convolutional neural network
CN113420826B (en) Liver focus image processing system and image processing method
Chen et al. Pulmonary nodule detection on chest radiographs using balanced convolutional neural network and classic candidate detection
Shahangian et al. Automatic brain hemorrhage segmentation and classification in CT scan images
CN111738997A (en) Method for calculating new coronary pneumonia lesion area ratio based on deep learning
Hasanzadeh et al. Segmentation of covid-19 infections on ct: Comparison of four unet-based networks
Akkar et al. Diagnosis of lung cancer disease based on back-propagation artificial neural network algorithm
Tian et al. Radiomics and its clinical application: artificial intelligence and medical big data
Tang et al. NSCGCN: A novel deep GCN model to diagnosis COVID-19
Hasan et al. Lung Segmentation from Chest X-Ray Images Using Deeplabv3plus-Based CNN Model
Malathi et al. Active contour based segmentation and classification for pleura diseases based on Otsu’s thresholding and support vector machine (SVM)
Liu et al. Lung CT Image Segmentation via Dilated U-Net Model and Multi-scale Gray Correlation-Based Approach
Costa et al. A deep learning-based radiomics approach for COVID-19 detection from CXR images using ensemble learning model
CN108230341A (en) A kind of eye fundus image blood vessel segmentation method that nomography is scratched based on layering
CN112508943A (en) Breast tumor identification method based on ultrasonic image
CN116778250A (en) Coronary artery lesion classification method based on transfer learning and CBAM
CN117746042A (en) Liver tumor CT image segmentation method based on APA-UNet
Jose et al. Liver Tumor Classification using Optimal Opposition-Based Grey Wolf Optimization
CN116843984A (en) GLTransNet: global feature fusion molybdenum target image classification detection method
Singh et al. A novel neural network based automated system for diagnosis of breast cancer from real time biopsy slides
CN116310525A (en) Pathological image classification method based on contrast representation distillation and output distillation
Koppisetti et al. A study on machine learning algorithms for lung cancer classification and prediction

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