AD early diagnosis method and system based on deep learning, medium and equipment

Abstract

The invention provides an AD early diagnosis method and system based on deep learning, a medium and equipment. The method comprises the following steps: acquiring T1-MRI image data of AD and a corresponding classification label; constructing a discrete wavelet transform module, decomposing the MRI image into low-frequency sub-bands and high-frequency sub-bands with different scales, and performing Concat operation on all sub-band images to obtain a feature map; constructing a feature extraction module to extract multi-scale features from the feature map; designing a multi-class classifier, and classifying the feature maps; building a classification network for processing a classification task of the T1-MRI image; and performing classification network training on the test set, and outputting a diagnosis result. According to the method, the T1-MRI image is used for feature extraction, through the design of the multi-scale feature extraction module and the multi-class classifier, the tiny change of the brain of the MCI patient can be accurately captured, and the accuracy of MCI diagnosis is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing, and in particular, to an AD early diagnosis method, system, medium and device based on deep learning. Background Art

[0002] Alzheimer's disease (AD) is a common neurodegenerative disease that seriously affects patients' cognitive function and quality of life. Early screening of AD can greatly delay disease progression, improve patients' quality of life and reduce social burden. Mild cognitive impairment (MCI) is the prodromal stage of AD. Accurate diagnosis and timely intervention of patients during this period are expected to prevent or delay their transformation to AD. Therefore, the diagnosis of MCI occupies a core position in the early screening process of AD. Among the many technologies used for early screening of AD, T1-weighted magnetic resonance imaging (T1-MRI) technology shows unique advantages. T1-MRI can clearly present the anatomical structure of the brain, providing a rich and accurate data basis for subsequent image analysis and feature extraction, helping to improve the diagnostic accuracy of MCI and providing strong support for early screening of AD. Currently, the existing early screening methods for AD are limited by the network structure and cannot effectively extract key features when processing neuroimaging data, resulting in insufficient ability to capture subtle changes in the brains of MCI patients. This problem has caused the MCI diagnostic accuracy to remain at a low level for a long time, seriously restricting its application in clinical practice. Therefore, there is an urgent need for an early diagnosis method for AD based on T1-MRI to achieve accurate screening of MCI, thereby assisting doctors in early diagnosis and clinical intervention of AD.

[0003] Patent application document CN113208620A discloses a method and system for screening Alzheimer's disease based on sleep staging, which includes the following steps: the BCI device collects EEG signals during the user's sleep and transmits them to the sleep staging algorithm system; the sleep staging algorithm system extracts and analyzes the user's EEG signals to identify the sleep state and perform staging; the early screening system extracts and analyzes the sleep staging results to identify whether the user is at a high risk stage for the early onset of Alzheimer's disease. However, this patent cannot completely solve the existing technical problems, nor can it meet the needs of the present invention. Summary of the invention

[0004] In view of the defects in the prior art, the object of the present invention is to provide an AD early diagnosis method, system, medium and equipment based on deep learning.

[0005] The deep learning-based AD early diagnosis method provided by the present invention includes: Step S1: Obtain T1-weighted magnetic resonance imaging (T1-MRI) image data of Alzheimer's disease AD and the corresponding classification labels, respectively denoted as and ,Will Divide into training sets according to preset ratios With the test set , respectively used for model training and model performance evaluation and verification; Step S2: Construct a discrete wavelet transform module MWT to decompose the MRI image into low-frequency sub-bands and high-frequency sub-bands of different scales, and obtain the feature map by concat operation on all sub-band images. ; Step S3: Construct a feature extraction module to extract The feature extraction module includes Mamba module MB and Dense module DM. MB module extracts global information by global modeling, and DM module adopts densely connected convolution structure to enhance the local feature representation ability. Step S4: Design a multi-category classifier MCC to classify the feature map; Step S5: Based on the constructed MWT, MB, DM and MCC modules, a classification network ED-Mamba is designed and built to handle the classification task of T1-MRI images; Step S6: ED-Mamba Train on the test set until the loss function stabilizes, save the final model parameters, output the diagnosis results, and finally ED-Mamba was tested on .

[0006] Preferably, step S1 includes: performing data screening on the acquired T1-MRI images, excluding preset low-quality images and data desensitization, and ensuring that all images meet the predetermined quality standards; resampling the screened T1-MRI images using image processing technology, adjusting the size of each image to 224×224×16; dividing into training sets at a ratio of 8:2 With the test set .

[0007] Preferably, the step S2 comprises: Step S2.1: Establishing the scaling function and wavelet function , as follows:

[0008]

[0009] Among them, j is the scale parameter and k is the translation parameter; Step S2.2: For each sample ,in are length, width and number of channels respectively; along the W direction Perform one-dimensional wavelet transform operation, low-frequency coefficients in the W direction and high frequency coefficients Calculated by the following formula:

[0010]

[0011] Step S2.3: Perform one-dimensional wavelet transform in the H direction to obtain low-frequency coefficients and high-frequency coefficients. Get 4 sub-bands , ; Each subband passes 1 1. Convolution is used for filtering, and the wavelet features that fuse multi-scale global and local information are obtained through the Concat operation. , obtained by the following formula:

[0012] in, It is a convolutional layer.

[0013] Preferably, step S3 comprises: Step S3.1: Construct the MB module, which includes a dual-branch VSS layer and a patch merging layer; the dual-branch VSS layer contains a convolution branch and a spatial perception module SSM branch to extract global and local features; the patch merging layer reduces the size of the feature map by half and doubles the number of channels; define the The input of each MB module is , in the dual branch VSS layer, Perform channel division operation to obtain and ; Then, and Perform a series of operations to obtain the output of the dual-branch VSS layer , It can be obtained as follows:

[0014] in, Represents the channel aggregation operation that restores the number of channels of the feature map to the original number of channels. The channel discrete operation makes the feature map randomly scattered at the channel level and preserves the information in the channel. It is the feature map obtained by the convolution branch, which is obtained by the following formula:

[0015] in, and Represents the ReLU activation function and batch normalization operation; It is a channel aggregation operation, which is used to restore the number of channels of the feature map to the original number of channels; It is the feature map obtained by SSM branch, which is obtained by the following formula:

[0016] represents the addition operation of corresponding elements of the matrix, Represents a linear transformation operation, feature map and It is obtained from the following formula:

[0017]

[0018] in, and SiLU activation function and layer normalization operation, SSM operation makes the feature map contain the information of the global receptive field, It is a depth-wise separable convolution; Step S3.2: Construct DM modules. Each DM module contains 6 layers of dense convolution. Each layer of dense convolution contains The convolution operation, batch normalization operation and ReLU activation function define the input of the DM module as , the output of the DM module Obtained by the following formula:

[0019] in, and Represents the ReLU activation function and batch normalization operation.

[0020] Preferably, step S4 comprises: Construct a multi-category classifier. When there are k categories, construct k binary classifiers , Used to learn the decision boundary of the kth category and other categories; construct a dynamic threshold adjustment strategy to determine the threshold of each category according to the difficulty of distinguishing each category. The difficulty of distinguishing each category is determined by calculating the predicted value of the classifier for the sample. The difficulty of distinguishing each category is defined as , Calculated by the following formula:

[0021] in, is the total number of samples, Represents the sample under the tth training round number The predicted value of The larger the The easier it is to distinguish the samples of the class; then Normalized to [0,1], and use the normalized To scale the predefined threshold , the formula is as follows:

[0022] in, It is The dynamic threshold of the class in the tth training round, is all categories under the tth training round The maximum value of the difficulty level.

[0023] Preferably, the step S5 comprises: Step S5.1: Build the modules proposed in steps S2 to S4 to obtain ED-Mamba, which includes 1 MWT module, 4 MB modules, 4 DM modules and 1 MCC module; Send it to the MWT module to get the output feature map ; Step S5.2: Send it to 1 MB module and 1 DM module to get the output feature map ; Step S5.3: Repeat step S5.2 three times to get the feature map ; Step S5.4: Send it to the MCC module to get the final output of the model , for Category.

[0024] Preferably, step S6 comprises: Step S6.1: Place ED-Mamba in For training, the Adam optimizer is used and the learning rate is After 400 iterations, the network becomes stable and the model parameters are saved; Step S6.2: Load the test set into the trained model parameters to predict the classification results and calculate the accuracy; Step S6.3: If the accuracy is lower than 0.9, repeat steps S6.1 and S6.2.

[0025] The AD early diagnosis system based on deep learning provided by the present invention includes: A data acquisition module, used to acquire T1-weighted magnetic resonance imaging (T1-MRI) image data and corresponding classification labels of the patient; A preprocessing module, used for performing data screening on the T1-MRI image, resampling the image size to 224×224×16, and dividing the image into a training set and a test set in proportion; A discrete wavelet transform module MWT is used to decompose the T1-MRI image into multi-scale low-frequency sub-bands and high-frequency sub-bands, and generate a multi-scale feature map through convolution fusion; A feature extraction module, comprising a Mamba module MB and a dense convolution module DM, for extracting global and local features from the multi-scale feature map; The multi-category classifier MCC adopts a dynamic threshold adjustment strategy to adaptively adjust the decision boundary based on the classification difficulty; The classification network ED-Mamba integrates the MWT module, MB module, DM module and MCC module to output the final diagnosis result.

[0026] According to the computer-readable storage medium storing a computer program provided by the present invention, when the computer program is executed by a processor, the steps of the deep learning-based AD early diagnosis method are implemented.

[0027] The electronic device provided according to the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the steps of the deep learning-based AD early diagnosis method are implemented.

[0028] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention uses T1-MRI images for feature extraction. Through the design of a multi-scale feature extraction module and a multi-category classifier, it can accurately capture subtle changes in the brain of MCI patients, significantly improving the accuracy of MCI diagnosis, with an accuracy rate of more than 0.9; (2) Through the MWT module and MB module, multi-scale features in T1-MRI images are effectively extracted and fused in the convolutional neural network, enhancing the model's ability to capture local and global information of the image; (3) The present invention adopts a dynamic threshold adjustment strategy to set different thresholds for each category, which optimizes the performance of the classifier, so that the classifier can adaptively adjust the decision boundary according to the difficulty of each category, further improving the classification accuracy; (4) The ED-Mamba network of the present invention combines the MWT, MB, DM and MCC modules to form an efficient and robust classification network. This network structure can not only effectively extract features from T1-MRI images, but also perform accurate classification, providing strong support for the early diagnosis of AD. (5) This method is based on T1-MRI image data for diagnosis and can adapt to the different conditions of different patients. It has strong versatility and clinical application value. Through efficient training and prediction, it can provide doctors with accurate diagnostic information, assist doctors in early intervention, and delay the progression of AD. (6) The present invention adopts the Adam optimizer and the method of dynamically adjusting the learning rate to ensure the efficiency of the training process and the stability of the model, so that a higher accuracy can be achieved within a smaller number of training rounds, thereby reducing the training cost and time. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings: Figure 1 Flowchart of the early diagnosis method of Alzheimer's disease based on deep learning. DETAILED DESCRIPTION

[0030] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0031] Example like Figure 1 The present invention provides an AD early diagnosis method based on deep learning, comprising: Step S1: Obtain T1-weighted magnetic resonance imaging (T1-MRI) image data of Alzheimer's disease AD and the corresponding classification labels, respectively denoted as and ,Will Divide into training sets in a ratio of 8:2 With the test set ,in For model training, Used for performance evaluation and verification of models; Step S2: Design and construct the discrete wavelet transform module MWT to decompose the MRI image into low-frequency sub-bands and high-frequency sub-bands of different scales. Then all sub-band images are concat-operated to obtain feature maps. ; Step S3: Construct a feature extraction module to extract The feature extraction module includes Mamba module MB and Dense module DM. MB module extracts global information by global modeling, and DM module adopts densely connected convolution structure to enhance the local feature representation ability. Step S4: To improve the classification accuracy, a multi-category classifier MCC is designed to classify the feature map; Step S5: Based on the MWT, MB, DM and MCC modules constructed above, a classification network ED-Mamba is designed and constructed to process the classification task of T1-MRI images; Step S6: ED-Mamba Train on until the loss function tends to stabilize and save the final model parameters. The ED-Mamba was tested on the above data, and the accuracy of the deep learning-based AD early diagnosis method proposed in the present invention was no less than 0.9.

[0032] The step S1 specifically includes: The acquired T1-MRI images were screened to exclude low-quality images (such as those with excessive noise, blur, incomplete images, etc.) and data desensitization to ensure that all images met the predetermined quality standards; the screened T1-MRI images were resampled using image processing technology to adjust the size of each image to 224×224×16; and divided into training sets at a ratio of 8:2 With the test set ; The step S2 specifically includes: Step S2.1: Construct the MWT module, first establish the scaling function and wavelet function , as follows:

[0033]

[0034] Among them, j is the scale parameter and k is the translation parameter; Step S2.2: For each sample ,in are length, width and number of channels respectively; first, Perform one-dimensional wavelet transform operation, low-frequency coefficients in the W direction and high frequency coefficients It can be calculated by the following formula:

[0035]

[0036] Step S2.3: Similarly, Performing one-dimensional wavelet transform in the H direction also obtains low-frequency coefficients and high-frequency coefficients. 4 sub-bands available , ; Then, each subband passes 1 1. Convolution is used for filtering, and the wavelet features that integrate multi-scale global and local information are obtained through the Concat operation. , can be obtained by the following formula:

[0037] in, is the convolutional layer; Step S3 specifically includes: Step S3.1: Construct the MB module, which includes a dual-branch VSS layer and a patch merging layer; the dual-branch VSS layer contains a convolution branch and a spatial perception module SSM branch to extract global and local features; the patch merging layer reduces the size of the feature map by half and doubles the number of channels; define the The input of each MB module is , in the dual branch VSS layer, Perform channel division operation to obtain and ; Then, and Perform a series of operations to obtain the output of the dual-branch VSS layer , It can be obtained by the following formula:

[0038] in, Represents the channel aggregation operation that restores the number of channels of the feature map to the original number of channels. The channel discrete operation makes the feature map randomly scattered at the channel level and preserves the information in the channel. It is the feature map obtained by the convolution branch, which can be obtained by the following formula:

[0039] in, and Represents the ReLU activation function and batch normalization operation; It is a channel aggregation operation, which is used to restore the number of channels of the feature map to the original number of channels; It is the feature map obtained by SSM branch, which can be obtained by the following formula:

[0040] represents the addition operation of corresponding elements of the matrix, Represents a linear transformation operation, feature map and It can be obtained by the following formula:

[0041]

[0042] in, and SiLU activation function and layer normalization operation, SSM operation enables the feature map to contain information of the global receptive field. It is a depth-wise separable convolution.

[0043] Step S3.2: Construct DM modules. Each DM module contains 6 layers of dense convolution. Each layer of dense convolution contains The convolution operation, batch normalization operation and ReLU activation function define the input of the DM module as , the output of the DM module It can be obtained by the following formula:

[0044] in, and Represents the ReLU activation function and batch normalization operation.

[0045] Step S4 specifically includes: Construct a multi-category classifier. When there are k categories, construct k binary classifiers , It is used to learn the decision boundary of the kth category and other categories; construct a dynamic threshold adjustment strategy so that the threshold of each category can be determined according to the difficulty of distinguishing each category. The difficulty of distinguishing each category can be determined by calculating the predicted value of the classifier for the sample. The difficulty of distinguishing each category is defined as , It can be calculated by the following formula:

[0046] in, is the total number of samples, Represents the sample under the tth training round number The predicted value of The larger the The easier it is to distinguish the samples of the class; then Normalized to [0,1], and use the normalized To scale the predefined threshold , the formula is as follows:

[0047] in, It is The dynamic threshold of the class in the tth training round, is all categories under the tth training round The maximum value of the difficulty level.

[0048] Step S5 specifically includes: Step S5.1: Build the modules proposed in Steps S2 to S4 to obtain ED-Mamba, which includes 1 MWT module, 4 MB modules, 4 DM modules and 1 MCC module. First, the image Send it to the MWT module to get the output feature map ; Step S5.2: Send it to 1 MB module and 1 DM module to get the output feature map ; Step S5.3: Repeat step S5.2 three times to get the feature map ; Step S5.4: Send it to the MCC module to get the final output of the model , for Category.

[0049] Step S6 specifically includes: Step S6.1: Place ED-Mamba in For training, the Adam optimizer is used and the learning rate is After 400 iterations, the network becomes stable and the model parameters are saved; Step S6.2: Load the test set into the trained model parameters to predict the classification results and calculate the accuracy; Step S6.3: If the accuracy is lower than 0.9, repeat steps S6.1 and S6.2.

[0050] The present invention also provides an AD early diagnosis system based on deep learning, comprising: A data acquisition module, used to acquire T1-weighted magnetic resonance imaging (T1-MRI) image data and corresponding classification labels of the patient; A preprocessing module, used for performing data screening on the T1-MRI image, resampling the image size to 224×224×16, and dividing the image into a training set and a test set in proportion; A discrete wavelet transform module MWT is used to decompose the T1-MRI image into multi-scale low-frequency sub-bands and high-frequency sub-bands, and generate a multi-scale feature map through convolution fusion; A feature extraction module, comprising a Mamba module MB and a dense convolution module DM, for extracting global and local features from the multi-scale feature map; The multi-category classifier MCC adopts a dynamic threshold adjustment strategy to adaptively adjust the decision boundary based on the classification difficulty; The classification network ED-Mamba integrates the MWT module, MB module, DM module and MCC module to output the final diagnosis result.

[0051] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the deep learning-based AD early diagnosis method.

[0052] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the deep learning-based AD early diagnosis method.

[0053] Those skilled in the art know that, in addition to implementing the system, device and its various modules provided by the present invention in a purely computer-readable program code, it is entirely possible to implement the same program in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers and embedded microcontrollers by logically programming the method steps. Therefore, the system, device and its various modules provided by the present invention can be considered as a hardware component, and the modules included therein for implementing various programs can also be considered as structures within the hardware component; the modules for implementing various functions can also be considered as both software programs for implementing the method and structures within the hardware component.

[0054] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A method for early diagnosis of AD based on deep learning, characterized in that: include: Step S1: Obtain T1-weighted magnetic resonance imaging (T1-MRI) image data of Alzheimer's disease AD and the corresponding classification labels, respectively denoted as and ,Will Divide into training sets according to preset ratios With the test set , respectively used for model training and model performance evaluation and verification; Step S2: Construct a discrete wavelet transform module MWT to decompose the MRI image into low-frequency sub-bands and high-frequency sub-bands of different scales, and obtain the feature map by concat operation on all sub-band images. ; Step S3: Construct a feature extraction module to extract The feature extraction module includes Mamba module MB and Dense module DM. MB module extracts global information by global modeling, and DM module adopts densely connected convolution structure to enhance the local feature representation ability. Step S4: Design a multi-category classifier MCC to classify the feature map; Step S5: Based on the constructed MWT, MB, DM and MCC modules, a classification network ED-Mamba is designed and built to handle the classification task of T1-MRI images; Step S6: ED-Mamba Train on the test set until the loss function stabilizes, save the final model parameters, output the diagnosis results, and finally ED-Mamba was tested on .

2. The deep learning-based AD early diagnosis method according to claim 1, characterized in that: The step S1 includes: performing data screening on the acquired T1-MRI images, excluding preset low-quality images and data desensitization, and ensuring that all images meet the predetermined quality standards; using image processing technology to resample the size of the screened T1-MRI images, adjusting the size of each image to 224×224×16; dividing into training sets at a ratio of 8:2 With the test set .

3. The deep learning-based AD early diagnosis method according to claim 1, characterized in that: The step S2 comprises: Step S2.1: Establishing the scaling function and wavelet function , as follows: Among them, j is the scale parameter and k is the translation parameter; Step S2.2: For each sample ,in are length, width and number of channels respectively; along the W direction Perform one-dimensional wavelet transform operation, low-frequency coefficients in the W direction and high frequency coefficients Calculated by the following formula: Step S2.3: Perform one-dimensional wavelet transform in the H direction to obtain low-frequency coefficients and high-frequency coefficients. Get 4 sub-bands , ; Each subband passes 1 1. Convolution is used for filtering, and the wavelet features that fuse multi-scale global and local information are obtained through the Concat operation. , obtained by the following formula: in, It is a convolutional layer.

4. The deep learning-based AD early diagnosis method according to claim 3, characterized in that: The step S3 comprises: Step S3.1: Construct the MB module, which includes a dual-branch VSS layer and a patch merging layer; the dual-branch VSS layer contains a convolution branch and a spatial perception module SSM branch to extract global and local features; the patch merging layer reduces the size of the feature map by half and doubles the number of channels; define the The input of each MB module is , in the dual branch VSS layer, Perform channel division operation to obtain and ; Then, and Perform a series of operations to obtain the output of the dual-branch VSS layer , It can be obtained as follows: in, Represents the channel aggregation operation that restores the number of channels of the feature map to the original number of channels. The channel discrete operation makes the feature map randomly scattered at the channel level and preserves the information in the channel. It is the feature map obtained by the convolution branch, which is obtained by the following formula: in, and Represents the ReLU activation function and batch normalization operation; It is a channel aggregation operation, which is used to restore the number of channels of the feature map to the original number of channels; It is the feature map obtained by SSM branch, which is obtained by the following formula: represents the addition operation of corresponding elements of the matrix, Represents a linear transformation operation, feature map and It is obtained from the following formula: in, and SiLU activation function and layer normalization operation, SSM operation makes the feature map contain the information of the global receptive field, It is a depth-wise separable convolution; Step S3.2: Construct DM modules. Each DM module contains 6 layers of dense convolution. Each layer of dense convolution contains The convolution operation, batch normalization operation and ReLU activation function define the input of the DM module as , the output of the DM module Obtained by the following formula: in, and Represents the ReLU activation function and batch normalization operation.

5. The deep learning-based AD early diagnosis method according to claim 4, characterized in that: The step S4 comprises: Construct a multi-category classifier. When there are k categories, construct k binary classifiers , Used to learn the decision boundary of the kth category and other categories; construct a dynamic threshold adjustment strategy to determine the threshold of each category according to the difficulty of distinguishing each category. The difficulty of distinguishing each category is determined by calculating the predicted value of the classifier for the sample. The difficulty of distinguishing each category is defined as , Calculated by the following formula: in, is the total number of samples, Represents the sample under the tth training round number The predicted value of The larger the The easier it is to distinguish the samples of the class; then Normalized to [0,1], and use the normalized To scale the predefined threshold , the formula is as follows: in, It is The dynamic threshold of the class in the tth training round, is all categories under the tth training round The maximum value of the difficulty level.

6. The deep learning-based AD early diagnosis method according to claim 5, characterized in that: The step S5 comprises: Step S5.1: Build the modules proposed in steps S2 to S4 to obtain ED-Mamba, which includes 1 MWT module, 4 MB modules, 4 DM modules and 1 MCC module; Send it to the MWT module to get the output feature map ; Step S5.2: Send it to 1 MB module and 1 DM module to get the output feature map ; Step S5.3: Repeat step S5.2 three times to get the feature map ; Step S5.4: Send it to the MCC module to get the final output of the model , for Category.

7. The deep learning-based AD early diagnosis method according to claim 1, characterized in that: The step S6 comprises: Step S6.1: Place ED-Mamba in For training, the Adam optimizer is used and the learning rate is After 400 iterations, the network becomes stable and the model parameters are saved; Step S6.2: Load the test set into the trained model parameters to predict the classification results and calculate the accuracy; Step S6.3: If the accuracy is lower than 0.9, repeat steps S6.1 and S6.

2.

8. An AD early diagnosis system based on deep learning, characterized in that: The deep learning-based AD early diagnosis method according to claim 6 comprises: A data acquisition module, used to acquire T1-weighted magnetic resonance imaging (T1-MRI) image data and corresponding classification labels of the patient; A preprocessing module, used for performing data screening on the T1-MRI image, resampling the image size to 224×224×16, and dividing the image into a training set and a test set in proportion; A discrete wavelet transform module MWT is used to decompose the T1-MRI image into multi-scale low-frequency sub-bands and high-frequency sub-bands, and generate a multi-scale feature map through convolution fusion; A feature extraction module, comprising a Mamba module MB and a dense convolution module DM, for extracting global and local features from the multi-scale feature map; The multi-category classifier MCC adopts a dynamic threshold adjustment strategy to adaptively adjust the decision boundary based on the classification difficulty; The classification network ED-Mamba integrates the MWT module, MB module, DM module and MCC module to output the final diagnosis result.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the deep learning-based AD early diagnosis method described in any one of claims 1 to 7 are implemented.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the computer program is executed by a processor, the steps of the deep learning-based AD early diagnosis method described in any one of claims 1 to 7 are implemented.

Images