Alzheimer's disease sMRI image classification method, system and equipment

Abstract

The invention provides an Alzheimer's disease sMRI image classification method, and the method comprises the steps: carrying out the detection of a collected to-be-detected image through an M3D classification model which is constructed based on an attention mechanism and ShuffleNetV2 and is trained through a training set, and forming detection feature data, and determining the probability of classifying the to-be-detected intermediate image as the Alzheimer's disease sMRI image according to the detection feature data, and obtaining a classification result of the to-be-detected image based on the probability. The Alzheimer's disease sMRI image classification method provided by the invention can quickly and accurately classify the Alzheimer's disease sMRI images, and has the characteristics of light and simple model construction, small operation pressure and high efficiency. The technical problems that in the prior art, when sMRI images are classified through an existing deep learning algorithm, calculation is complex, and the speed is low are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of Alzheimer's disease image classification, and specifically relates to an Alzheimer's disease sMRI image classification method, system and equipment. Background Art

[0002] Alzheimer's disease (AD) is a progressive neurodegenerative disorder characterized by the gradual loss of neurons and functional decline in the brain, often leading to a decline in cognitive abilities and impacting daily life and social function. Since AD ​​progresses rapidly and irreversibly once it reaches the moderate or severe stages, and AD is best treated in its early stages, predicting whether a patient has AD before the onset of moderate or severe symptoms is crucial.

[0003] In the actual predictive diagnosis of Alzheimer's disease, brain imaging technologies such as magnetic resonance imaging (MRI), structural magnetic resonance imaging (sMRI), and positron emission computed tomography (PET) are commonly used to generate imaging data. sMRI is a non-invasive method that can be used to measure brain morphology and exhibits excellent soft tissue contrast and spatial resolution, allowing us to capture subtle changes in brain tissue. Due to its relatively low cost and good imaging quality, it has been used in the diagnosis, treatment, and research of Alzheimer's disease.

[0004] In the diagnosis, treatment and research of Alzheimer's disease, after collecting sMRI images, they need to be judged and classified. In the existing technology, sMRI images are often classified through deep learning algorithms. However, the existing methods of classifying sMRI images through deep learning algorithms generally use a mixture of multiple feature extraction methods, which increases the complexity of the model and has the problems of high computational complexity and slow speed. Summary of the Invention

[0005] In order to solve the technical problems of complex calculation and slow speed when using existing deep learning algorithms to classify sMRI images in the background technology, the present invention provides a method, system and device for classifying Alzheimer's disease sMRI images.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a method for classifying Alzheimer's disease sMRI images, characterized by comprising the following steps:

[0008] S1: Acquire an image to be measured, and pre-process the image to be measured to obtain an intermediate image to be measured;

[0009] S2: Inputting the intermediate image to be tested into a trained M3D classification model for detection to form detection feature data, wherein the trained M3D classification model is constructed based on the attention mechanism and ShuffleNetV2 and is trained with a training set; the training set includes Alzheimer's disease sMRI images and normal sMRI images;

[0010] S3: Determine the probability of classifying the intermediate image to be tested as an Alzheimer's disease sMRI image according to the detection feature data, and obtain a classification result of the image to be tested based on the probability.

[0011] Optionally, the step of forming the trained M3D classification model includes:

[0012] Establish an sMRI imaging database;

[0013] Based on the sMRI image database, generating a training set;

[0014] Build an untrained M3D classification model based on the attention mechanism and ShuffleNetV2;

[0015] The untrained M3D classification model is iteratively trained using the training set to obtain the trained M3D classification model.

[0016] Optionally, the step of forming the trained M3D classification model further includes:

[0017] Based on the sMRI image database, a validation set is also generated;

[0018] During the process of training the untrained M3D classification model, verification is also performed using the verification set; when the verification result tends to converge, the training is stopped to obtain the trained M3D classification model.

[0019] Optionally, in step S2, inputting the intermediate image to be tested into the trained M3D classification model for detection specifically includes:

[0020] Performing dimensionality reduction processing on the intermediate image to be measured and extracting features to obtain first feature data;

[0021] Performing multiple feature extractions and enhanced feature expressions on the first feature data to obtain second feature data;

[0022] Extract key features of the second feature data to form detection feature data.

[0023] Optionally, performing multiple feature extractions and enhanced feature expressions on the first feature data specifically includes:

[0024] Performing feature extraction on the first feature data to obtain first sub-feature data;

[0025] Performing enhanced feature expression on the first sub-feature data to obtain second sub-feature data;

[0026] Performing feature extraction on the second sub-feature data to obtain third sub-feature data;

[0027] Performing enhanced feature expression on the third sub-feature data to obtain fourth sub-feature data;

[0028] Performing feature extraction on the fourth sub-feature data to obtain fifth sub-feature data;

[0029] The feature expression of the fifth sub-feature data is enhanced to obtain the second feature data.

[0030] Optionally, the process of obtaining a classification result based on the detection feature data specifically includes:

[0031] The detected feature data is input into the fully connected neural network layer and the discard layer for classification processing to obtain the probability of classifying the intermediate image to be tested as an Alzheimer's disease sMRI image;

[0032] If the probability of classifying the intermediate image to be tested as an Alzheimer's disease sMRI image is greater than 50%, the classification result is output as: the image to be tested is an Alzheimer's disease sMRI image;

[0033] If the probability of classifying the intermediate image to be tested as an Alzheimer's disease sMRI image is less than or equal to 50%, the classification result is output as: the image to be tested is a normal sMRI image.

[0034] Optionally, inputting the intermediate image to be tested into the trained M3D classification model for detection and classification further includes the following steps:

[0035] Identify and obtain the focus area of ​​the attention mechanism in the process of enhancing feature expression of the first feature data, and visualize the focus area to obtain a visualization view of the focus area.

[0036] In a second aspect, the present invention further provides an Alzheimer's disease sMRI image classification system, which is used in any of the above-mentioned Alzheimer's disease sMRI image classification methods, comprising:

[0037] The imaging module is used to collect the image to be tested, pre-process the image to be tested, and obtain the intermediate image to be tested;

[0038] A processing module is used to input the intermediate image to be tested into the trained M3D classification model for detection to form detection feature data;

[0039] The output module is configured to receive the detection feature data, determine a probability of classifying the intermediate image to be tested as an Alzheimer's disease sMRI image based on the detection feature data, and obtain a classification result of the image to be tested based on the probability.

[0040] Optionally, the trained M3D classification model includes:

[0041] A first feature processing layer is used to receive the intermediate image to be measured, perform dimensionality reduction processing on the intermediate image to be measured, and extract features to obtain first feature data;

[0042] an intermediate processing layer, comprising a plurality of processing layers and a plurality of attention mechanisms, configured to receive the first feature data, perform multiple feature extractions through the plurality of processing layers, and perform multiple enhanced feature expressions through the plurality of attention mechanisms to obtain the second feature data;

[0043] The second feature processing layer is used to receive the second feature data, extract key features from the second feature data, and obtain the detection feature data.

[0044] In a third aspect, the present invention provides an Alzheimer's disease sMRI image classification device, comprising: a storage component and a processing component;

[0045] The storage component stores instructions, wherein the instructions are generated based on any of the above-mentioned Alzheimer's disease sMRI image classification methods;

[0046] The processing component calls and executes the instructions.

[0047] The beneficial effects of the present invention are:

[0048] (1) The present invention provides an Alzheimer's disease sMRI image classification method that can quickly and accurately classify Alzheimer's disease sMRI images. Compared with the commonly used VGG16 model and GoogleLeNet model in the prior art, the present invention has the characteristics of lightweight and simple model construction, low computational pressure, and high computational efficiency.

[0049] (2) At the same time, the Alzheimer's disease sMRI image classification method provided by the present invention outputs a visualized image, thereby further classifying and judging the Alzheimer's disease sMRI image based on the visualized image. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 11 is a flow chart of the Alzheimer's disease sMRI image classification method of the present invention;

[0051] Figure 2 This is a schematic diagram of the process of preprocessing the image to be tested in the present invention;

[0052] Figure 3 It is a schematic diagram of the process of forming the M3D classification model in the present invention;

[0053] Figure 4 It is a data flow diagram of the M3D classification model in the present invention;

[0054] Figure 5 This is an example of outputting a heat map after performing visualization operations in the present invention;

[0055] Figure 6 It is another schematic diagram of the architecture of the M3D classification model in the present invention. DETAILED DESCRIPTION

[0056] The present invention provides a method, system and device for classifying Alzheimer's disease sMRI images, which are used to classify Alzheimer's disease sMRI images.

[0057] The following embodiments of the present invention are further described in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0058] Example 1

[0059] First, see Figure 1 , shows a schematic diagram of a method for classifying Alzheimer's disease sMRI images described in this application, which method should at least include the following steps:

[0060] S1: Acquire the image to be measured, pre-process the image to be measured, and obtain the intermediate image to be measured.

[0061] Optionally, refer to Figure 2 In this embodiment, the pre-processing of the image to be tested may include the following steps:

[0062] Perform head motion correction on the image to be measured to eliminate image deviation caused by the subject's head movement during the acquisition of the image to be measured, ensure the stability of the image, and obtain the first intermediate image;

[0063] Performing intensity standardization on the first intermediate image to uniformly adjust the signal intensity of the image and eliminate the problem of inconsistent signal intensity caused by different scanning devices or protocols, thereby obtaining a second intermediate image;

[0064] Furthermore, non-intensity-related correction can be performed to further ensure the consistency and reliability of the image data;

[0065] The second intermediate image is processed to remove the influence of the skull and non-brain tissue in the image, focusing on the representation of the brain structure, to obtain the third intermediate image;

[0066] Performing linear registration on the third intermediate image to align the image to a common space, such as MNI152_T1_2mm space, to facilitate comparison and analysis of the images in subsequent models, and obtaining the fourth intermediate image;

[0067] The information-free area of ​​the fourth intermediate image is cropped to obtain the intermediate image to be tested. The information-free area in the image is cropped to reduce the overall size of the image, reduce unnecessary burdens in the subsequent model calculation process, and improve classification efficiency.

[0068] The above preprocessing steps can be performed by FreeSurfer, FSL or any other neuroimaging processing software known to those skilled in the art.

[0069] S2: The intermediate image to be tested is input into the trained M3D classification model for detection to form detection feature data, wherein the trained M3D classification model is constructed based on the attention mechanism and ShuffleNetV2, and is trained with a training set; the training set includes Alzheimer's disease sMRI images and normal sMRI images.

[0070] Optionally, refer to Figure 3 In this embodiment, the steps of forming a trained M3D classification model may include:

[0071] Establish an sMRI imaging database;

[0072] Generate a training set based on the sMRI image database;

[0073] Build an untrained M3D classification model based on the attention mechanism and ShuffleNetV2;

[0074] The untrained M3D classification model is iteratively trained using the training set to obtain a trained M3D classification model.

[0075] Among them, the sMRI image database can be obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) project website (http: / / adni.loni.usc.edu), thereby establishing an sMRI image database. The established sMRI image database should at least include: sMRI images of patients with Alzheimer's disease and sMRI images of normal people; and generate a training set for training the model from the sMRI image database. The training set should include sMRI images of patients with Alzheimer's disease and sMRI images of normal people in equal proportions.

[0076] Optionally, the step of forming the trained M3D classification model may further include:

[0077] Based on the sMRI image database, a training set and a validation set are generated at the same time;

[0078] During the process of training the untrained M3D classification model, verification is also performed using a verification set; when the verification result tends to converge, the training is stopped to obtain the trained M3D classification model.

[0079] Specifically, based on the sMRI image database, the ratio of sMRI image data included in the generated training set and validation set can be selected as 4:1.

[0080] Furthermore, the steps of constructing the M3D classification model may be:

[0081] (1) Construct the first feature processing layer, the second feature processing layer and the result classification layer;

[0082] (2) Based on ShuffleNetV2, we obtain the two basic processing layers it contains: the Channel Split layer and the Channel Shuffle layer;

[0083] (3) Perform three-dimensional transformation on the two basic processing layers of ShuffleNetV2 to obtain the first sub-processing layer and the second sub-processing layer respectively;

[0084] (4) constructing a generative intermediate processing layer through the first sub-processing layer and the second sub-processing layer and the first attention mechanism, the second attention mechanism, and the third attention mechanism;

[0085] (5) An untrained M3D classification model is constructed in the order of the first feature processing layer, the intermediate processing layer, the second feature processing layer, and the result classification layer.

[0086] Specifically, in this embodiment, based on ShuffleNetV2, a three-dimensional first sub-processing layer and a second sub-processing layer are obtained. Based on the first sub-processing layer, the second sub-processing layer, the first attention mechanism, the second attention mechanism and the third attention mechanism, an intermediate processing layer is constructed, so that the M3D classification model provided in the present invention can not only be used for the detection and classification of planar sMRI image slices, but also for the analysis and classification of three-dimensional images, without the need for slicing the three-dimensional images, thereby avoiding information fragmentation or omission of clues.

[0087] Therefore, refer to Figure 4 and Figure 6 , the M3D classification model constructed in this embodiment should include:

[0088] a first feature processing layer for receiving an intermediate image to be measured, performing dimensionality reduction processing on the intermediate image to be measured, and extracting features to obtain first feature data; wherein the intermediate image to be measured may be a three-dimensional image;

[0089] The intermediate processing layer includes multiple processing layers and an attention mechanism, and is used to receive the first feature data, perform multiple feature extractions and enhance feature expressions on the first feature data, and obtain the second feature data;

[0090] The second feature processing layer is used to receive the second feature data, extract key features from the second feature data, and obtain detection feature data.

[0091] Furthermore, a result classification layer may be included, which is used to receive the detection feature data and perform classification processing on the detection feature data to obtain the probability of classifying the intermediate image to be tested as an Alzheimer's disease sMRI image.

[0092] Furthermore, the first feature processing layer in this embodiment may specifically include a convolution layer, a pooling layer, a batch normalization layer, and a linear correction layer. The specific processing process includes: after receiving the intermediate image to be measured, performing convolution and pooling operations on the intermediate image to reduce its dimension and refine its core information; performing batch normalization and linear correction activation operations to obtain the first feature data.

[0093] Specifically, during feature extraction, the convolutional kernels in the convolutional layer scan the input image, capturing local features such as edges and textures. Next, activation functions (such as ReLU) are used to enhance nonlinear representations and highlight key features. Finally, the pooling layer downsamples the data to reduce dimensionality and improve the model's robustness to minor deformations. Multiple layers of convolution and pooling are performed alternately to gradually extract higher-level feature representations.

[0094] Optionally, in this embodiment, performing multiple feature extractions and enhanced feature expressions on the first feature data specifically includes:

[0095] Performing feature extraction on the first feature data to obtain first sub-feature data;

[0096] Performing enhanced feature expression on the first sub-feature data to obtain second sub-feature data;

[0097] Performing feature extraction on the second sub-feature data to obtain third sub-feature data;

[0098] Performing enhanced feature expression on the third sub-feature data to obtain fourth sub-feature data;

[0099] Performing feature extraction on the fourth sub-feature data to obtain fifth sub-feature data;

[0100] The feature expression of the fifth sub-feature data is enhanced to obtain the second feature data.

[0101] Therefore, the intermediate processing layer in this embodiment may specifically include:

[0102] The first processing layer is used to receive the first feature data and perform feature extraction on the first feature data to obtain first sub-feature data;

[0103] The first attention mechanism is used to receive the first sub-feature data and enhance the feature expression of the first sub-feature data to obtain the second sub-feature data;

[0104] The second processing layer is used to receive the second sub-feature data and perform feature extraction on the second sub-feature data to obtain third sub-feature data;

[0105] The second attention mechanism is used to receive the third sub-feature data and enhance the feature expression of the third sub-feature data to obtain the fourth sub-feature data;

[0106] a third processing layer, configured to receive the fourth sub-feature data and perform feature extraction on the fourth sub-feature data to obtain fifth sub-feature data;

[0107] The third attention mechanism is used to receive the fifth sub-feature data and enhance the feature expression of the fifth sub-feature data to obtain the second feature data.

[0108] Furthermore, the first attention mechanism and the second attention mechanism in this embodiment can be selected as the selective kernel network (SKNet) to achieve dynamic evaluation and strengthen the key feature expression; the third attention mechanism can be selected as the squeeze and excitation network (SENet) to achieve recalibration of the key feature expression at the channel level.

[0109] Furthermore, the first processing layer in this embodiment is composed of two first sub-processing layers and one second sub-processing layer arranged in sequence, thereby deepening the level of feature learning; the second processing layer is composed of three first sub-processing layers and one second sub-processing layer arranged in sequence, and feature extraction is refined through a more complex second processing layer; the third processing layer is composed of two first sub-processing layers and one second sub-processing layer arranged in sequence, and further feature extraction is performed through the third processing layer with a structure similar to the first processing layer, while ensuring the consistency and depth of feature processing.

[0110] Furthermore, in this embodiment, the data processing flow of the first sub-processing layer includes:

[0111] (1) obtaining first original data, and evenly splitting the first original data into first sub-data and second sub-data;

[0112] (2) performing three convolution operations on the second sub-data to enhance feature expression and obtain the third sub-data;

[0113] (3) performing a channel shuffling operation on the third sub-data to obtain fourth sub-data;

[0114] (4) merging the first sub-data with the fourth sub-data to obtain first output data;

[0115] In the data processing of the first sub-processing layer, the feature map size and the number of channels of the data can be kept unchanged, which not only achieves effective reorganization of features but also keeps the basic structure of the data.

[0116] The data processing flow of the second sub-processing layer includes:

[0117] (1) obtaining the second original data, and copying the second original data into two copies, namely the fifth sub-data and the sixth sub-data;

[0118] (2) performing separable convolution processing on the fifth sub-data and then performing normalized convolution processing to obtain the seventh intermediate data;

[0119] (3) performing normalized convolution processing on the sixth sub-data and then performing separable convolution processing to obtain eighth intermediate data;

[0120] (4) merging the seventh intermediate data and the eighth intermediate data, and performing a channel shuffling operation on the data to obtain second output data;

[0121] In the data processing of the second sub-processing layer, the feature map of the input raw data is reduced to half of its original size, cleverly compressing the spatial dimension. The number of feature channels is significantly increased through internal mechanisms, thereby reducing the computational burden while enhancing the model's ability to express complex features, achieving efficient space-channel conversion and feature enhancement.

[0122] S3: Determine the probability of classifying the intermediate image to be tested as an Alzheimer's disease sMRI image according to the detection feature data, and obtain a classification result of the image to be tested based on the probability.

[0123] Optionally, the process of obtaining a classification result based on the detection feature data specifically includes:

[0124] The detected feature data is input into the fully connected neural network layer and the discard layer for classification processing to obtain the probability of classifying the intermediate image to be tested as an Alzheimer's disease sMRI image;

[0125] If the probability of classifying the intermediate image to be tested as an Alzheimer's disease sMRI image is greater than 50%, the classification result is output as: the image to be tested is an Alzheimer's disease sMRI image;

[0126] If the probability of classifying the intermediate image to be tested as an Alzheimer's disease sMRI image is less than or equal to 50%, the classification result is output as: the image to be tested is a normal sMRI image.

[0127] Specifically, the detection feature data obtained through the aforementioned operations is input into a fully connected neural network layer, where it is classified using a three-layer fully connected neural network. Dropout layers are set between each layer to reduce overfitting during model training and use. Specifically, the dropout layer can be selected with a probability of 0.5.

[0128] It should be noted that those skilled in the art may select specific probability judgment criteria for the output results according to actual applications, and this embodiment only provides one implementation scheme.

[0129] Optionally, the present invention further includes the steps of inputting the intermediate image to be tested into the trained M3D classification model for detection and classification:

[0130] Identify and obtain the focus area of ​​the attention mechanism in the process of enhancing the feature expression of the first feature data, and visualize the focus area to obtain a visualization view of the focus area.

[0131] The Alzheimer's disease sMRI image classification method in this embodiment can be configured to set a visualization processing layer, based on the enhanced gradient weighted class mapping technology (Grad-CAM++), identify the gradient of the convolution layer output in the feature, and generate the following Figure 5 The heat map shown intuitively shows the area that the model pays most attention to during classification, that is, the area that contributes the most to the model classification process, and outputs the heat map, so that medical staff can further classify and judge Alzheimer's disease sMRI images based on the heat map.

[0132] The Alzheimer's disease sMRI image classification method provided by the present invention detects and classifies the collected images to be tested through a trained M3D classification model constructed based on the attention mechanism and ShuffleNetV2 and trained on a training set, thereby obtaining a classification result of the images to be tested. Referring to Table 1, the data of the M3D classification model provided by the present invention and two models that have been maturely used in the prior art, VGG16 and GoogleLeNet, when tested on the same test set are shown. According to Table 1, it can be seen that the M3D classification model provided by the present invention has advantages in terms of accuracy (ACC), sensitivity (SEN), precision (PRE), training time (time) and giga-multiply-accumulate operations (GMACs).

[0133] Therefore, the Alzheimer's disease sMRI image classification method provided by the present invention can quickly and accurately classify Alzheimer's disease sMRI images, and has the characteristics of lightweight and simple model construction, low computational pressure, and high efficiency.

[0134] Table 1

[0135]

[0136] It should be further explained that the Alzheimer's disease sMRI image classification method provided by the present invention can also detect and classify sMRI images of various other diseases based on different training sets provided when training the M3D classification model. For example, the M3D classification model can be trained by sMRI images of four different states including Alzheimer's disease (AD), early mild cognitive impairment (EMCI), late mild cognitive impairment (LMCI) and healthy controls (CN), so that the M3D classification model obtained after training can detect and classify sMRI images of these four different states. Therefore, the Alzheimer's disease sMRI image classification method provided by the present invention is also scalable, and those skilled in the art can modify and expand it according to actual needs.

[0137] Example 2

[0138] In a second aspect, the present invention further provides an Alzheimer's disease sMRI image classification system, which is used in the Alzheimer's disease sMRI image classification method in Example 1 and may at least include:

[0139] The imaging module is used to collect the image to be tested, pre-process the image to be tested, and obtain the intermediate image to be tested;

[0140] A processing module is used to input the intermediate image to be tested into the trained M3D classification model for detection to form detection feature data;

[0141] The output module is configured to receive the detection feature data, determine a probability of classifying the intermediate image to be tested as an Alzheimer's disease sMRI image based on the detection feature data, and obtain a classification result of the image to be tested based on the probability.

[0142] It should be noted that the Alzheimer's disease sMRI image classification system provided in this embodiment is completely corresponding to the virtual reality data synchronization method in Example 1. The specific implementation content can be found in Example 1 and will not be described in detail here.

[0143] In this embodiment, a sMRI image classification system for Alzheimer's disease is provided. The collected images to be tested are detected by an M3D classification model constructed based on the attention mechanism and ShuffleNetV2 and trained with a training set, and finally the classification results of the images to be tested are obtained. The system has the advantages of being lightweight, simple, low in computational pressure, highly efficient and scalable.

[0144] Optionally, the trained M3D classification model in this embodiment includes:

[0145] A first feature processing layer is used to receive the intermediate image to be measured, perform dimensionality reduction processing on the intermediate image to be measured, and extract features to obtain first feature data;

[0146] an intermediate processing layer, comprising a plurality of processing layers and a plurality of attention mechanisms, configured to receive the first feature data, perform multiple feature extractions through the plurality of processing layers, and perform multiple enhanced feature expressions through the plurality of attention mechanisms to obtain the second feature data;

[0147] The second feature processing layer is used to receive the second feature data, extract key features from the second feature data, and obtain the detection feature data.

[0148] It should be noted that the M3D classification model provided in this embodiment is completely corresponding to the M3D classification model provided in Example 1. The specific implementation content can be found in the content of Example 1 and will not be repeated here.

[0149] Example 3

[0150] In a third aspect, the present invention further provides an Alzheimer's disease sMRI image classification device, characterized by comprising: a storage component and a processing component;

[0151] The storage component stores instructions, which are generated based on the Alzheimer's disease sMRI image classification method described in Example 1;

[0152] The processing component calls and executes the instructions.

[0153] Specifically, the storage component described in the present invention may include a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, or a CD-ROM. It should be noted that those skilled in the art may select the form and type of storage component based on actual production and use requirements, and this is not further limited in this embodiment.

[0154] It should be noted that those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the system and device described above can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

Claims

1. A method for classifying sMRI images of Alzheimer's disease, characterized in that: The following steps are involved: S1: Acquire an image to be tested, and pre-process the image to be tested to obtain an intermediate image to be tested; S2: Inputting the intermediate image to be tested into the trained M3D classification model for detection to form detection feature data, wherein the trained M3D classification model is constructed based on the attention mechanism and ShuffleNetV2, and is formed by training with a training set; the training set includes Alzheimer's disease sMRI images and normal sMRI images; S3: Determine the probability of classifying the intermediate image to be tested as an Alzheimer's disease sMRI image according to the detection feature data, and obtain a classification result of the image to be tested based on the probability.

2. The Alzheimer's disease sMRI image classification method according to claim 1, characterized in that: The steps of forming the trained M3D classification model include: Establish a sMRI imaging database; Based on the sMRI image database, generating a training set; Construct an untrained M3D classification model based on the attention mechanism and ShuffleNetV2; The untrained M3D classification model is iteratively trained using the training set to obtain the trained M3D classification model.

3. The Alzheimer's disease sMRI image classification method according to claim 2, characterized in that: The step of forming the trained M3D classification model also includes: Based on the sMRI image database, a validation set is also generated; In the process of training the untrained M3D classification model, verification is also performed through the verification set; when the verification result tends to converge, the training is stopped to obtain the trained M3D classification model.

4. The Alzheimer's disease sMRI image classification method according to claim 1, characterized in that: In step S2, inputting the intermediate image to be tested into the trained M3D classification model for detection specifically includes: Performing dimensionality reduction processing on the intermediate image to be measured and extracting features to obtain first feature data; Performing multiple feature extractions and enhanced feature expressions on the first feature data to obtain second feature data; The key features of the second feature data are extracted to form detection feature data.

5. The Alzheimer's disease sMRI image classification method according to claim 4, characterized in that: The multiple feature extraction and enhanced feature expression of the first feature data specifically include: Performing feature extraction on the first feature data to obtain first sub-feature data; Performing enhanced feature expression on the first sub-feature data to obtain second sub-feature data; Performing feature extraction on the second sub-feature data to obtain third sub-feature data; Performing enhanced feature expression on the third sub-feature data to obtain fourth sub-feature data; Performing feature extraction on the fourth sub-feature data to obtain fifth sub-feature data; The feature expression of the fifth sub-feature data is enhanced to obtain the second feature data.

6. The Alzheimer's disease sMRI image classification method according to claim 4, characterized in that: The process of obtaining classification results based on detection feature data specifically includes: The detected feature data is input into the fully connected neural network layer and the discard layer for classification processing, and the probability of classifying the intermediate image to be tested as an Alzheimer's disease sMRI image is obtained; If the probability of classifying the intermediate image to be tested as an Alzheimer's disease sMRI image is higher than 50%, the classification result is output as follows: the image to be tested is an Alzheimer's disease sMRI image; If the probability of classifying the intermediate image to be tested as an Alzheimer's disease sMRI image is less than or equal to 50%, the classification result is output as: the image to be tested is a normal sMRI image.

7. The Alzheimer's disease sMRI image classification method according to claim 4, characterized in that: Inputting the intermediate image to be tested into the trained M3D classification model for detection and classification also includes the following steps: Identify and obtain a focus area of ​​the attention mechanism in the process of enhancing feature expression of the first feature data, and visualize the focus area to obtain a visualized view of the focus area.

8. An Alzheimer's disease sMRI image classification system, used in the Alzheimer's disease sMRI image classification method according to any one of claims 1 to 7, characterized in that: include: An image module is used to collect the image to be tested, pre-process the image to be tested, and obtain the intermediate image to be tested; A processing module, used for inputting the intermediate image to be tested into the trained M3D classification model for detection to form detection feature data; The output module is used to receive the detection feature data, determine the probability of classifying the intermediate image to be tested as an Alzheimer's disease sMRI image according to the detection feature data, and obtain the classification result of the image to be tested based on the probability.

9. The Alzheimer's disease sMRI image classification system according to claim 8, characterized in that: The trained M3D classification model includes: A first feature processing layer is used to receive the intermediate image to be measured, perform dimensionality reduction processing on the intermediate image to be measured, and extract features to obtain first feature data; An intermediate processing layer, comprising a plurality of processing layers and a plurality of attention mechanisms, for receiving the first feature data, performing multiple feature extractions through the plurality of processing layers, and performing multiple enhanced feature expressions through the plurality of attention mechanisms, to obtain the second feature data; The second feature processing layer is used to receive the second feature data, extract key features from the second feature data, and obtain the detection feature data.

10. An Alzheimer's disease sMRI image classification device, characterized in that: include: storage components and processing components; The storage component stores instructions, wherein the instructions are generated based on the Alzheimer's disease sMRI image classification method according to any one of claims 1 to 7; The processing component calls and executes the instructions.

Images