Medical image analysis method and system, electronic equipment and readable storage medium

Abstract

The invention discloses a medical image analysis method and system, electronic equipment and a readable storage medium, and belongs to the technical field of deep learning. The method comprises the following steps: S100, acquiring a lung CT image with a real label for pneumonia prediction; s200, constructing a medical image analysis network model, wherein the model comprises an image encoder, a feature disturbance module, a feature correction module, a cyclic correction module and a dual adaptive network; and S300, classifying the input lung CT images through the medical image analysis network model. According to the method, the robustness and generalization ability of the model in an image classification task can be improved, and higher adaptive capacity and higher diagnosis precision can be shown in the face of data distribution offset and diversified attack means.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and more specifically to a medical image analysis method, system, electronic device, and readable storage medium. Background Art

[0002] Disease diagnosis and anomaly detection are key links in medical image analysis (medical image analysis technology refers to the use of computer algorithms and artificial intelligence technology to process and analyze medical images (such as X-rays, CT, MRI, etc.) to assist doctors in disease diagnosis, anomaly detection, tissue classification, treatment planning, etc.). Accurate image analysis can help doctors detect lesions earlier, develop personalized treatment plans, and monitor the progression of the disease, improving diagnostic accuracy and efficiency. The application of deep learning models has broad application prospects in the field of medical image analysis. However, the performance of the model is highly dependent on the distribution of data. When faced with a complex clinical environment, distribution shift will cause the model to perform poorly in the new environment, increasing the misdiagnosis rate and missed diagnosis rate; at the same time, due to privacy protection requirements and the suddenness of medical and health events, the high sensitivity and lack of medical data have brought great challenges to model training. Therefore, how to improve model performance with limited data while protecting privacy has become an urgent problem to be solved.

[0003] Therefore, the Test-time Adaptation (TTA) mechanism was proposed, which assumes that the source data cannot be accessed, allowing the model to self-optimize according to the data in the actual application scenario after deployment, so that the model can quickly adapt to the new environment, thereby improving its generalization and robustness. Traditional transfer learning or domain adaptation methods usually require a large amount of labeled data, which may not be very practical. In contrast, TTA can not only use a small number of unlabeled samples for fine-tuning without violating privacy regulations, but also effectively deal with the differences in data distribution caused by different times, equipment, shooting conditions and imaging protocols. This makes the TTA mechanism particularly suitable for disease diagnosis and anomaly detection tasks in medical image analysis.

[0004] Traditionally, medical image analysis relies on a single network to obtain image features. However, a single path may not be able to fully capture the multi-scale information of the image, especially when faced with complex structures such as lung CT scans. In order to capture information comprehensively and multi-dimensionally, some studies have introduced dual structures. For example, Li et al. (Li, Y., Chen, X., & Hu, J. (2018). A multi-task learning framework for high-resolution and low-resolution image feature extraction. Journal of Medical Imaging, 5(3), 034501.) proposed a multi-task learning framework that combines two convolutional neural networks (CNNs) of different depths to extract features from high-resolution and low-resolution images respectively. This approach enhances the model's ability to capture information at different scales, but also brings about the problem of increased model complexity. However, these methods usually require more computing resources, and how to effectively utilize and integrate dual models or even multiple models remains a challenge. Contrastive Test-Time Adaptation proposed by Chen et al. fine-tunes the model during the test phase through contrastive learning, which not only maintains the original knowledge of the model but also enhances its adaptability to new fields, providing a new idea and direction for the application of multi-models in TTA. (Dian Chen, Dequan Wang, Trevor Darrell, and Sayna Ebrahimi. Contrastive test-time adaptation. In CVPR, 2022.) To address the domain difference problem, TTA methods mainly focus on adjusting the batch normalization (BN) layer, improving output predictions, or domain space alignment. Methods for adjusting the BN layer, such as the AdaBN (Adaptive Batch Normalization) method that dynamically updates the statistics and affine parameters of the BN layer according to the statistical information of the target domain to adapt, are widely used, and its performance is equivalent to updating the entire BN layer. However, it is easy to fail to accurately estimate the test data distribution due to insufficient sample size in the target domain, making the BN layer update invalid. In contrast, AdaIN (Adaptive Instance Normalization) is more suitable for single or few-sample learning environments. Because it allows personalized adjustments for each input by applying normalization processing to each instance separately. At the same time, AdaIN has similarities with the MixStyle method, both of which involve operations on feature map styles (i.e., channel-level mean and standard deviation) to generate diverse samples. Direct feature-level domain synthesis methods effectively generate different potential target domains by randomly perturbing feature channel statistics, but if the perturbation intensity is too large, it may cause highly activated channels to be overly disturbed, resulting in a decrease in model performance. In addition, too drastic changes may also cause the model to lose its original knowledge. Therefore, it is necessary to develop a more sophisticated control disturbance mechanism.

[0005] In summary, how to provide a medical image analysis method, system, electronic device, and readable storage medium is a problem that technical personnel in this field urgently need to solve. Summary of the invention

[0006] In view of this, the present invention provides a medical image analysis method, system, electronic device, and readable storage medium, which aim to improve the robustness and generalization ability of the model in image classification tasks, thereby improving the accuracy and reliability of disease diagnosis and abnormality detection, and is particularly suitable for cross-institutional data sharing and rapid response to new diseases.

[0007] In order to achieve the above object, the present invention provides the following technical solutions: A medical image analysis method, comprising: S100: Obtain lung CT images with true labels for pneumonia prediction; S200: constructing a medical image analysis network model, the model comprising an image encoder, a feature perturbation module, a feature correction module, a cycle correction module and a dual adaptive network; S300: classifying the input lung CT image through the medical image analysis network model; Among them, the image encoder is used to extract features of the input lung CT image to obtain a feature map; the feature perturbation module is used to apply local or global perturbations to the feature map in the shallow layer of the medical image analysis network model to obtain the perturbed features; the feature correction module is used to correct the perturbed features to obtain the corrected features; the cyclic correction module is used to iteratively apply the correction network to perform secondary correction on the corrected features to obtain a statistical representation after cyclic correction; the dual adaptive network includes two sub-networks built on the same ResNet architecture but with independently adjusted parameters, which are respectively used to process feature maps, generate and output their own image classification results, and finally integrate them.

[0008] Furthermore, the image encoder is used to extract features from the input lung CT image to obtain a dimension of Feature map ;in, represents the channel dimension of the feature map, and Represent the height and width of the feature map respectively.

[0009] Furthermore, the feature perturbation module includes: a local perturbation module, a global perturbation module and a global statistics update mechanism; the processing steps of the feature perturbation module include: calculating feature channel statistics based on feature maps; randomly introducing local perturbations and global perturbations at specific network layers according to conditional probability; for local perturbations, first create an all-zero matrix and an all-one matrix with the same dimension as the mean of the feature map, and then randomly generate noise coefficients from a normal distribution with a mean of 0 and a standard deviation of 0.75. These noise coefficients are used to control the injection process of Gaussian noise, replacing the feature channel statistics (mean and standard deviation) of the original feature map with the perturbed channel statistics. Then, through adaptive instance normalization, the perturbation parameters (i.e., the mean and standard deviation after perturbation) are applied to the affine transformation of the features, thereby adjusting the feature map; if global perturbation is applied, the data set statistics of the current layer are determined, and the mean and standard deviation after perturbation are calculated using the global perturbation factor, similar to local perturbation, to generate a perturbed feature map.

[0010] Furthermore, the feature correction module creates a corresponding sequence model by several perturbation layers, including: three convolutional layers Conv2d, two custom normalization layers LayerNorm2d, and two nonlinear activation functions ReLU; wherein, the custom normalization layer: normalizes the feature map of each channel and calculates the mean and standard deviation of each channel.

[0011] Further, the cycle correction module includes: The random probability of local disturbance and global disturbance determines whether to make corrections, and when the conditions are met, the corresponding disturbance parameters are used for processing; The mean and standard deviation of the correction feature are adjusted according to the perturbation parameter; Perform standardization and rescaling translation operations to generate new feature distributions; The processed features are passed to the domain adapter of the correction network and the separated tensor method is used to avoid gradient calculation and ensure the gradient flow remains unchanged; Calculate and update the mean and standard deviation of the feature map for subsequent loss calculation.

[0012] Furthermore, the dual adaptive network includes two encoders with the same structure but different values, and respective corresponding decoders for mapping features to final classification results; Among them, two isomorphic but heterogeneous encoders share the same network configuration, but the initialization values, learning rates or optimizers are randomly different; The lung CT images are input into two dual adaptive encoders to extract multi-level feature representations, and then the integrated prediction output is output through two identical decoders.

[0013] Furthermore, the medical image analysis network is also trained using a cross entropy loss function, a similarity loss function, and an alignment cycle loss function until the loss function converges, thereby obtaining an optimal medical image analysis model for classifying and outputting input lung images.

[0014] A medical image analysis system, comprising: Data acquisition module: obtain lung CT images with real labels for pneumonia prediction; Model building module: building a medical image analysis network model, the model including an image encoder, a feature perturbation module, a feature correction module, a cycle correction module and a dual adaptive network; Identification and detection module: classifying the input lung CT image through the medical image analysis network model; Among them, the image encoder is used to extract features of the input lung CT image to obtain a feature map; the feature perturbation module is used to apply local or global perturbations to the feature map in the shallow layer of the medical image analysis network model to obtain the perturbed features; the feature correction module is used to correct the perturbed features to obtain the corrected features; the cyclic correction module is used to iteratively apply the correction network to perform secondary correction on the corrected features to obtain a statistical representation after cyclic correction; the dual adaptive network includes two sub-networks built on the same ResNet architecture but with independently adjusted parameters, which are respectively used to process feature maps, generate and output their own image classification results, and finally integrate them.

[0015] An electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute a medical image analysis method, and the processor is configured to execute the program stored in the memory.

[0016] A computer-readable storage medium stores a computer program, which executes the steps of a medical image analysis method when executed by a processor.

[0017] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a medical image analysis method, system, electronic device, and readable storage medium, which have the following beneficial effects: 1. The present invention uses isomorphic but different-value feature extractors (such as initialization value, learning rate, optimizer, etc.) to form a dual-branch structure to capture diverse feature representations. Each branch generates two paths: one retains the original features, and the other is the enhanced features after perturbation and correction. The prediction results of the two decoders are fused through the idea of ​​ensemble learning, which significantly enhances the expressiveness and robustness of the model, especially in the face of complex data distribution changes, it can better capture diverse features, thereby improving the stability and accuracy of the final decision.

[0018] 2. The present invention randomly introduces local and global disturbances in the shallow layer of the network, and adjusts the disturbed features through the designed correction network, so that the model can learn more robust feature representations and improve the perception of abnormal areas. And a cyclic correction mechanism is used to further optimize the feature representation, so that the model shows stronger adaptability and higher diagnostic accuracy when dealing with diverse attack methods. This method enhances the model's adaptability to different environmental conditions and avoids overfitting specific data patterns. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0020] Figure 1 It is a schematic diagram of the method flow of the present invention; Figure 2 It is a schematic diagram of the medical image analysis network model structure of the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] The purpose of the present invention is to provide a medical image analysis method, system, electronic device, and readable storage medium, including: 1) using feature extractors with isomorphic but different initial settings (for example, ResNet, DenseNet, etc. based on convolutional neural networks CNN) to obtain feature representations of input images; 2) randomly introducing local perturbations and global perturbations in the shallow layers of the network (such as the 1st, 2nd, and 3rd layers) according to preset probabilities; 3) each feature extractor generates two branches: one retains the original features, and the other perturbs the features; 4) the perturbed features are adjusted through a correction network, and a cyclic correction mechanism is used to further optimize the feature representation; 5) using the idea of ​​ensemble learning to fuse the prediction results of the two decoders; 6) using a combination of cross entropy loss, similarity loss, and alignment cycle loss and other loss functions to guide model optimization; 7) in the adaptive stage during testing, when encountering new target domain samples, the system only updates some network parameters (such as the correction network parameters in the feature extractor) for fine-tuning through operations similar to the training stage. The present invention can improve the robustness and generalization ability of the model in image classification tasks, and can show stronger adaptability and higher diagnostic accuracy when facing data distribution deviation and diversified attack methods.

[0023] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] See also Figure 1-2 As shown, an embodiment of the present invention discloses a medical image analysis method, comprising: S100: Obtain lung CT images with true labels for pneumonia prediction; S200: constructing a medical image analysis network model, the model comprising an image encoder, a feature perturbation module, a feature correction module, a cycle correction module and a dual adaptive network; S300: classifying the input lung CT image through the medical image analysis network model; Among them, the image encoder is used to extract features of the input lung CT image to obtain a feature map; the feature perturbation module is used to apply local or global perturbations to the feature map in the shallow layer of the medical image analysis network model to obtain the perturbed features; the feature correction module is used to correct the perturbed features to obtain the corrected features; the cyclic correction module is used to iteratively apply the correction network to perform secondary correction on the corrected features to obtain a statistical representation after cyclic correction; the dual adaptive network includes two sub-networks built on the same ResNet architecture but with independently adjusted parameters, which are respectively used to process feature maps, generate respective image classification results and output them, and finally integrate the outputs.

[0025] In a specific embodiment, first obtain A CT image of the lungs , for detecting pneumonia; using a general image encoder Each performs feature extraction; the dimension is Feature map ;in, represents the channel dimension of the feature map, and Represent the height and width of the feature map respectively.

[0026] Then, a feature perturbation module is constructed, including: local perturbation module, global perturbation module and global statistics update mechanism. At a specific network layer, the feature map is randomly perturbed according to the preset probability. Apply local or global perturbations to simulate image changes under different conditions. Local perturbations simulate small changes in the lesion area of ​​the image, and global perturbations simulate the influence of factors such as overall image quality or illumination. The specific steps are as follows: Step 2.1: Given a feature map , calculate the feature channel statistics, that is, the average value along each channel dimension and standard deviation ; Step 2.2: In the 1st, 2nd, and 3rd layers of the network, the channel dimensions are [64, 128, 256], and local and global perturbations are randomly introduced according to conditional probability. If , then a local perturbation is applied; if , then global perturbation is applied; only when , do not meet the conditional probability of disturbance, the feature is not disturbed, then the feature of the disturbance ; Step 2.3: Local perturbation: Create two tensors of the same dimension as the feature map mean, namely, an all-zero matrix and an all-one matrix of [1, C, 1, 1]. Randomly generate noise coefficients from a normal distribution with a mean of 0 and a standard deviation of 0.75. and , Controlled feature map The difference between its mean and This controls the contribution of the mean to the feature map.

[0027] Step 2.4: Using two perturbation factors and To control and Gaussian noise injection process: In formula (1), represents the mean of the disturbance; represents the variance of the disturbance.

[0028] Step 2.5: Map the original features The characteristic channel statistics of Replaced by the perturbed channel statistic , using the idea of ​​adaptive instance normalization (AdaIN), the perturbed statistics are applied to the original feature map to obtain the perturbed feature map , which not only retains the structure of the original feature map, but also introduces the changes brought by the disturbance. The calculation formula is as follows: In formula (2), represents the perturbation feature map.

[0029] Step 2.6: If global perturbation is applied, the data set statistics datum_center and datum_var of the current layer need to be determined. Similar to the local perturbation module, the perturbation mean and standard deviation are calculated using the perturbation factor to replace the original statistics, and the perturbation feature map is obtained using formula (2).

[0030] In a specific embodiment, a feature correction module is constructed to correct the deviation caused by the perturbation and ensure the consistency and stability of the feature representation. The purpose is to correct different target domain features to align with the source domain, including creating a corresponding sequence model according to the perturbation layer output channel dimension [128, 256, 512]. Specifically, for each perturbation layer, the created sequence model includes the following components, including: three convolutional layers Conv2d, two custom normalization layers LayerNorm2d, and two nonlinear activation functions ReLU; among them, the custom normalization layer: normalizes the feature map of each channel and calculates the mean of each channel and standard deviation , the specific process is: Step 3.1: The root layer normalizes the feature map according to the normalization formula, and uses learnable parameters: weights (initialized to all 1) and biases (initialized to all 0) to scale and translate the standardized feature map. Learnable parameters allow the model to adaptively adjust the distribution of feature maps to retain important feature information.

[0031] Step 3.2: The feature correction module uses the perturbed features And its statistics get the preliminary correction characteristics and its correction factor , .

[0032] In formula (3), express The mean of express The standard deviation of is a constant to prevent division by zero errors, usually .

[0033] Step 3.3: The feature correction module uses formula (4) to obtain a new mean and standard deviation based on the correction factor and replaces the original statistics: In formula (4), represents the mean of the calibration characteristics; Represents the standard deviation of the calibration characteristic.

[0034] Step 3.4: The feature correction module uses formula (5) to Normalized to zero mean and unit variance. The feature correction module uses equation (6) to rescale and translate the standardized feature map using the updated mean and standard deviation to achieve the final correction: (6) In a specific embodiment, a cyclic correction module is constructed to iteratively apply the perturbation mechanism and the correction network to the feature after it has been corrected once. Perform secondary corrections to further optimize feature representation and improve the model's ability to learn complex patterns. The specific process is as follows: Step 4.1: Determine whether to perform cyclic correction on the feature map based on the random probability of local disturbance and global disturbance. , that is, to meet the probability condition of local perturbation, the local perturbation parameter is used and Perform a cyclic correction; if but , that is, only the probability condition of global perturbation is satisfied, and the global perturbation parameter is used and Perform cyclic correction; if both conditions are not met, no correction is performed. and are all less than 0.5, will trigger first This means that in this case the local perturbation parameter ( and ) for cyclic correction.

[0035] Step 4.2: Process the mean and standard deviation of the correction feature according to the disturbance parameters.

[0036] Step 4.3, Normalize and rescale and translate: Normalize to zero mean and unit variance, and rescale and translate using the updated mean and standard deviation to generate a new distribution .

[0037] Step 4.4: Pass it to the domain adapter of the corresponding correction network, and avoid gradient calculation by separating the tensor method (detach()) to ensure that the gradient flow of the original feature map is not changed, and generate the feature map after secondary processing 。

[0038] The detach() method returns a new tensor that has the same values ​​as the original tensor, but is no longer associated with other operations in the computation graph and therefore will not be used to calculate gradients.

[0039] Step 4.5: Calculate feature map The mean and standard deviation of Step 4.6: Update the final mean and standard deviation based on right Adjust to get the final mean ;based on right Adjust to get the final standard deviation , which is used in subsequent loss calculations.

[0040] In formula (10), represents the characteristic mean after cycle correction; represents the characteristic standard deviation after cycle correction.

[0041] Step 5: Build a dual adaptive network, including: dual adaptive encoder (ResNet18), decoder including simple linear layer; Step 5.1: Build two parallel and Image encoder, the two models share the same network configuration ResNet18, but the initialization values, learning rates or optimizers are randomly different, thus introducing diverse and robust feature representations.

[0042] Step 5.2: Input the image X into two encoders respectively, extract multi-level feature representations, and perform feature and Processing and , and , and then pass two identical decoders to output the integrated prediction output ,

[0043] In formula (11), express and The mean of the predicted labels output by the encoder about the original features, express and The mean of the predicted labels output by the encoder with respect to the rectified features.

[0044] Step 6: The medical image analysis network is composed of the image encoder, the feature perturbation module, the feature correction module, the cyclic correction module and the dual adaptive ResNet model, and the cross entropy loss function of the medical image analysis network is constructed using equations (12), (13), (14) and (15) respectively. , minimize the difference between the predicted output and the actual label; similarity loss function ; Measure the similarity and difference of samples in the feature space, reduce the difference in feature distribution between the source domain and the target domain; Alignment cycle loss and To constrain the adapter, the purpose is to enable the adapter to correct the perturbed features back to the source domain space, improve the adaptability and stability of the model in a changing environment, and ensure that the learned features are both discriminative and internally consistent: In formula (13), Represents the first encoder The original features in Represents the second encoder The enhanced features after perturbation correction in Represents the second encoder The original features in Represents the first encoder Enhanced features after disturbance correction in ; (14) In formula (14), C represents the number of perturbed network layers, and N represents the number of encoders.

[0045] In formula (15), It is a feature after two corrections.

[0046] Step 7: Train the medical image analysis network using the Adam optimizer, and calculate the total loss function using formula (16): To update the network parameters until the loss function Until convergence, the optimal medical image analysis model is obtained, which is used to classify the input lung images and output the corresponding disease predictions.

[0047] In formula (16), express The weight of express and The weight of .

[0048] Step 8: During the test adaptation phase, the network is still in training mode and obtains an unlabeled target domain image. , continue training according to the above steps, fine-tune some network parameters, and only allow the network to be corrected and its related parameters participate in the gradient descent update, and the rest of the network parameters remain unchanged. Meanwhile, global statistics are continuously updated.

[0049] In formula (17), Represents the learning rate of the network during the adaptation phase at test time.

[0050] On the other hand, an embodiment of the present invention further discloses an electronic device, including a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the above method, and the processor is configured to execute the program stored in the memory.

[0051] On the other hand, an embodiment of the present invention further discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above method are executed.

[0052] As for the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0053] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A medical image analysis method, characterized in that: include: S100: Obtain lung CT images with true labels for pneumonia prediction; S200: constructing a medical image analysis network model, wherein the medical image analysis network model includes an image encoder, a feature perturbation module, a feature correction module, a cycle correction module and a dual adaptive network; S300: classifying the input lung CT image through the medical image analysis network model; Among them, the image encoder is used to extract features of the input lung CT image to obtain a feature map; the feature perturbation module is used to apply local or global perturbations to the feature map in the shallow layer of the medical image analysis network model to obtain the perturbed features; the feature correction module is used to correct the perturbed features to obtain the corrected features; the cyclic correction module is used to iteratively apply the correction network to perform secondary correction on the corrected features to obtain a statistical representation after cyclic correction; the dual adaptive network includes two sub-networks built on the same ResNet architecture but with independently adjusted parameters, which are respectively used to process feature maps, generate and output their own image classification results, and finally integrate them.

2. A medical image analysis method according to claim 1, characterized in that: The image encoder is used to extract features from the input lung CT image to obtain a dimension of Feature map ;in, represents the channel dimension of the feature map, and Represent the height and width of the feature map respectively.

3. A medical image analysis method according to claim 1, characterized in that: The feature perturbation module includes: a local perturbation module, a global perturbation module and a global statistics update mechanism; the processing steps of the feature perturbation module include: calculating feature channel statistics based on feature maps; randomly introducing local perturbations and global perturbations at a specific network layer according to conditional probability; for local perturbations, firstly creating an all-zero matrix and an all-one matrix with the same dimension as the mean of the feature map, and then randomly generating a noise coefficient from a normal distribution with a mean of 0 and a standard deviation of 0.75; the noise coefficient is used to control the injection process of Gaussian noise, replacing the feature channel statistics of the original feature map with the perturbed channel statistics; through adaptive instance normalization, the perturbation parameters are applied to the affine transformation of the features, thereby achieving adjustment of the feature map; if global perturbation is applied, the data set statistics of the current layer are determined, and the mean and standard deviation after perturbation are calculated using the global perturbation factor, similar to the local perturbation, thereby generating a perturbed feature map.

4. A medical image analysis method according to claim 1, characterized in that: The feature correction module creates a corresponding sequence model by several perturbation layers, including: three convolutional layers Conv2d, two custom normalization layers LayerNorm2d, and two nonlinear activation functions ReLU; wherein, the custom normalization layer: normalizes the feature map of each channel and calculates the mean and standard deviation of each channel.

5. A medical image analysis method according to claim 1, characterized in that: The cycle correction module comprises: The random probability of local disturbance and global disturbance determines whether to make corrections, and when the conditions are met, the corresponding disturbance parameters are used for processing; The mean and standard deviation of the correction feature are adjusted according to the perturbation parameter; Perform normalization and rescaling translation operations to generate new feature distributions; The processed features are passed to the domain adapter of the correction network and the separated tensor method is used to avoid gradient calculation and ensure the gradient flow remains unchanged; Calculate and update the mean and standard deviation of the feature map for subsequent loss calculation.

6. A medical image analysis method according to claim 1, characterized in that: The dual adaptive network includes two encoders with the same structure but different values, and a corresponding decoder for mapping features to the final classification result; Among them, two isomorphic but heterogeneous encoders share the same network configuration, but the initialization values, learning rates or optimizers are randomly different; The lung CT images are input into two dual adaptive encoders to extract multi-level feature representations, and then the integrated prediction output is output through two identical decoders.

7. A medical image analysis method according to claim 1, characterized in that: The medical image analysis network is also trained using a cross entropy loss function, a similarity loss function, and an alignment cycle loss function until the loss function converges, thereby obtaining an optimal medical image analysis model for classifying and outputting input lung images.

8. A medical image analysis system using the medical image analysis method according to any one of claims 1 to 7, characterized in that: include: Data acquisition module: obtain lung CT images with real labels for pneumonia prediction; Model building module: building a medical image analysis network model, the model including an image encoder, a feature perturbation module, a feature correction module, a cycle correction module and a dual adaptive network; Identification and detection module: classifying the input lung CT image through the medical image analysis network model; Among them, the image encoder is used to extract features of the input lung CT image to obtain a feature map; the feature perturbation module is used to apply local or global perturbations to the feature map in the shallow layer of the medical image analysis network model to obtain the perturbed features; the feature correction module is used to correct the perturbed features to obtain the corrected features; the cyclic correction module is used to iteratively apply the correction network to perform secondary correction on the corrected features to obtain a statistical representation after cyclic correction; the dual adaptive network includes two sub-networks built on the same ResNet architecture but with independently adjusted parameters, which are respectively used to process feature maps, generate and output their own image classification results, and finally integrate them.

9. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the medical image analysis method described in any one of claims 1 to 7, and the processor is configured to execute the program stored in the memory.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the medical image analysis method according to any one of claims 1 to 7 are executed.

Images