Alzheimer's disease diagnosis prediction method and system based on multi-source knowledge guidance

Abstract

The invention discloses an Alzheimer's disease diagnosis and prediction method based on multi-source knowledge guidance, and belongs to the technical field of medical artificial intelligence assistance. Comprising the following steps: acquiring a nuclear magnetic resonance image, brain statistical data and clinical data of a participant; processing the nuclear magnetic resonance image by using a dual-path feature extraction strategy, obtaining global brain features and local brain features, carrying out collaborative learning, and generating fused visual features; the brain statistical data and the fused visual features are fused and then input into a trained diagnosis module for processing, and a diagnosis result at the current moment is obtained; and taking the clinical data as guidance, combining the diagnosis result at the current moment, and utilizing the trained prediction module to obtain the disease progress at the future moment. The multi-modal data can be integrated, the accuracy and comprehensiveness of Alzheimer's disease diagnosis and prediction are improved, and the problems that the accuracy of existing Alzheimer's disease diagnosis is not high, and clinical requirements are difficult to meet are solved.

Description

Technical Field

[0001] The present invention relates to the field of medical artificial intelligence assisted technology, and in particular to a method and system for diagnosing and predicting Alzheimer's disease based on multi-source knowledge guidance. Background Art

[0002] The statements in this section merely mention background art related to the present invention and do not necessarily constitute prior art.

[0003] Alzheimer's disease (AD) is an irreversible degenerative disease of the central nervous system characterized by progressive memory loss and cognitive impairment over time. This process is often accompanied by brain tissue atrophy, pathological β-amyloid plaque formation, and abnormal brain metabolism. As current drug treatments are unable to halt the progression of AD, attention has turned to mild cognitive impairment (MCI), a critical stage that precedes the onset of AD. MCI is considered a key transition stage from normal aging to AD, highlighting the importance of its identification and monitoring for early AD detection. However, diagnosis and intervention of AD and MCI face significant challenges due to subjective differences among experts.

[0004] Fortunately, multimodal neuroimaging techniques such as magnetic resonance imaging (MRI) and positron emission tomography combined with fluorescent deoxyglucose PET (FDG-PET) can facilitate diagnosis. In addition, various clinical indicators including age, gender, blood pressure, and objective cognitive function assessments obtained through scales such as the Clinical Dementia Rating Scale (CDRSB) and ADAS-Cog13 can improve the accuracy of AD prediction. For example, apolipoprotein E4 (apoE4), a key clinical biomarker found in more than half of AD patients, indicates disease risk. Therefore, objective and effective AD progression prediction methods based on multi-source knowledge are urgently needed.

[0005] In current research, artificial intelligence diagnosis and prognosis methods for Alzheimer's disease mainly focus on computer vision data-driven technologies, and most systems rely on single-modality data, such as MRI or PET images. These methods often ignore the potential interactions and complementary information between different modal data, thus limiting the comprehensive understanding of the disease. Although a few studies have begun to explore the interaction of multimodal data, the methods for feature fusion are relatively simple and fail to deeply explore the complex relationship between modalities, affecting the accuracy and depth of diagnosis. In addition, most methods aim at diagnosis and lack the ability to provide comprehensive and effective information on prognosis in real-world applications.

[0006] After in-depth research, the inventors found that there are three main problems with the current artificial intelligence methods used for Alzheimer's disease diagnosis and prognosis:

[0007] First, most systems rely on single-modality patient data and fail to fully explore and utilize the complementary information between different modalities of data, resulting in insufficient reliability of system output results, thus limiting their effective application in clinical practice.

[0008] Secondly, existing methods are relatively limited in the extraction of visual features and lack the effective integration of global and local features, making it impossible for the model to fully and deeply understand the patient's condition.

[0009] Finally, the existing systems are still insufficient in the application of prognostic analysis, which is crucial for clinical decision-making. These limitations together lead to the low accuracy of Alzheimer's disease diagnosis and are difficult to meet clinical needs. Summary of the invention

[0010] In order to address the deficiencies of the prior art, the present invention provides a multi-source knowledge-guided Alzheimer's disease diagnosis and prediction method, system, electronic device, computer-readable storage medium and computer program product, which provides a comprehensive assessment of the progression of Alzheimer's disease by integrating multimodal data (magnetic resonance imaging, brain statistics and clinical data).

[0011] In a first aspect, the present invention provides a method for diagnosing and predicting Alzheimer's disease based on multi-source knowledge guidance;

[0012] A multi-source knowledge-guided Alzheimer's disease diagnosis and prediction method, comprising:

[0013] Obtain participants’ MRI images, brain demographics, and clinical data;

[0014] The magnetic resonance imaging image is processed by using a dual-path feature extraction strategy to obtain global brain features and local brain features and perform collaborative learning to generate fused visual features;

[0015] The brain statistical data is integrated with the fused visual features and then input into a trained diagnosis module for processing to obtain a diagnosis result at the current moment;

[0016] Guided by the clinical data and combined with the diagnosis results at the current moment, the disease progression at the future moment is obtained using the trained prediction module.

[0017] In some embodiments, the step of processing the magnetic resonance image using a dual-path feature extraction strategy to obtain global brain features and local brain features includes:

[0018] Processing the magnetic resonance imaging images using an improved ShuffleNetv2 model to obtain global brain features;

[0019] Processing the magnetic resonance imaging using a 3D CNN network to obtain local brain features;

[0020] Among them, the improved ShuffleNetv2 model and the 3D CNN network are set up in parallel.

[0021] In some embodiments, the global brain features and the local brain features are collaboratively learned to generate fused visual features, specifically by fusing the global brain features and the local brain features through an interactive attention mechanism to obtain fused visual features.

[0022] In some embodiments, fusing the brain statistical data with the fused visual features specifically includes: concatenating the encoded brain statistical data with the fused visual features, adding tokens and position codes, and constructing a feature vector guided by brain statistical knowledge.

[0023] In some embodiments, guided by the clinical data and combined with the diagnosis results at the current moment, a trained prediction module is used to obtain the disease progression at a future moment. Specifically, the diagnosis results at the current moment are fused with the encoded clinical data, and the fusion results are processed through a multi-head attention mechanism to predict the disease progression at a future moment.

[0024] In some embodiments, a comprehensive loss function is constructed by combining the diagnosis stage loss, the prediction stage loss and the consistency loss to optimize the diagnosis module and the prediction module.

[0025] In a second aspect, the present invention provides an Alzheimer's disease diagnosis and prediction system based on multi-source knowledge guidance;

[0026] A multi-source knowledge-guided Alzheimer's disease diagnosis and prediction system, comprising:

[0027] an acquisition module, configured to: acquire MRI images, brain statistics, and clinical data of the participant;

[0028] The diagnosis module is configured to: process the magnetic resonance image using a dual-path feature extraction strategy, obtain global brain features and local brain features, perform collaborative learning, and generate fused visual features; fuse the brain statistical data with the fused visual features and input them into the trained diagnosis module for processing to obtain the diagnosis result at the current moment;

[0029] The prediction module is configured to: use the clinical data as a guide, combine the diagnosis results at the current moment, and use the trained prediction module to obtain the disease progression at a future moment.

[0030] In a third aspect, the present invention provides an electronic device;

[0031] An electronic device includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-mentioned Alzheimer's disease diagnosis and prediction method based on multi-source knowledge guidance.

[0032] In a fourth aspect, the present invention provides a computer-readable storage medium;

[0033] A computer-readable storage medium stores a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned Alzheimer's disease diagnosis and prediction method based on multi-source knowledge guidance.

[0034] In a fifth aspect, the present invention provides a computer program product;

[0035] A computer program product includes a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned Alzheimer's disease diagnosis and prediction method based on multi-source knowledge guidance.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1. The technical solution provided by the present invention follows the operating steps of clinical practice, comprehensively utilizes multimodal data, including nuclear magnetic resonance imaging (sMRI), clinical data statistics and brain statistical knowledge, extracts dual-path visual features, combines global brain area and key local brain area features, and uses an interactive attention mechanism to achieve deep learning and fusion between features; guides image feature extraction through brain statistical knowledge, and guides prognosis prediction using clinical knowledge, which greatly improves the accuracy of diagnosis and prediction and significantly reduces the misdiagnosis rate. Improves the accuracy of diagnosis and prediction and reduces misdiagnosis.

[0038] 2. The technical solution provided by the present invention designs a comprehensive loss function for multiple periods, balances the relationship between diagnosis and prediction tasks, ensures that the model can maintain consistency in predictions at multiple time points, and effectively finds the optimal convergence solution of the model.

[0039] 3. The technical solution provided by the present invention has strong interpretability. By quantitatively analyzing the attention mechanism of the model, it can accurately locate the patient's key brain areas and clinical indicators, making the diagnosis and prediction results closer to clinical reality, and providing new ideas for disease prognosis and prevention. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0041] Figure 1A schematic diagram of a process flow of a multi-source knowledge-guided Alzheimer's disease diagnosis and prediction method provided by an embodiment of the present invention;

[0042] Figure 2 A structural block diagram of an Alzheimer's disease diagnosis and prediction system based on multi-source knowledge guidance provided by an embodiment of the present invention;

[0043] Figure 3 A schematic diagram of the structure of a feature matrix fusion module based on a clinical knowledge prompting mechanism provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0044] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0045] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "have" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0046] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0047] Embodiment 1

[0048] Existing artificial intelligence diagnosis and prediction methods for Alzheimer's disease fail to fully explore the complementarity and correlation between different modal data, and there are limitations in feature extraction, which restricts the accuracy and comprehensiveness of diagnosis and prediction; therefore, the present invention provides an Alzheimer's disease diagnosis and prediction method based on multi-source knowledge guidance, which comprehensively utilizes multimodal data.

[0049] Next, combine Figure 1 , Figure 3 A method for diagnosing and predicting Alzheimer's disease based on multi-source knowledge guidance disclosed in this embodiment is described in detail. The method for diagnosing and predicting Alzheimer's disease based on multi-source knowledge guidance includes the following steps:

[0050] S1. Obtain the participants' MRI images, brain statistics, and clinical data and map them into visual feature matrix, clinical data feature matrix, and brain statistics feature matrix, respectively.

[0051] In this embodiment, the brain statistical data includes the gray matter density statistics of each brain region of the participant, and the clinical data includes the comprehensive scale, education level and basic information of the participant.

[0052] Here, the visual feature matrix is ​​a structured feature matrix, and the clinical data feature matrix and the brain statistical feature matrix are unstructured feature matrices.

[0053] S2. Use the dual-path feature extraction strategy to process the MRI images, obtain global brain features and local brain features, and perform collaborative learning to generate fused visual features. Specifically include:

[0054] S201. Use the improved ShuffleNetv2 model and 3D CNN network to process the visual feature matrix and obtain global brain features respectively. and local brain features C1 is the dimension of global brain features, and C2 is the dimension of local brain features.

[0055] In this embodiment, the improved ShuffleNetv2 model includes a channel splitting (ChannelSplit) layer, a 1×1 three-dimensional convolution layer, a 3×3 depth-separable three-dimensional convolution layer, a 1×1 three-dimensional convolution layer, a splicing layer and a channel mixing (Channel shuffle) layer connected in sequence, and the output of the channel splitting layer is directly input into the splicing layer.

[0056] In order to adapt to three-dimensional magnetic resonance images, here, by adjusting the network structure and introducing channel configuration adapted to 3D data, the number of channels in the channel shuffle layer is expanded from 1024 to 2048 to adapt to the higher capacity of 3D data; both the depthwise separable convolution and convolution layers use three-dimensional convolution layers, grouped convolutions and point-by-point convolutions, which improves the flexibility and generalization ability of the model and makes the improved ShuffleNetv2 model architecture suitable for more complex 3D data scenarios.

[0057] Before application, a training set was constructed using the collected MRI images, and the improved ShuffleNetv2 model and 3D CNN network were trained respectively.

[0058] Based on this, changes in brain structure can be captured more comprehensively, improving the accuracy of feature expression and the sensitivity of diagnosis; global feature extraction identifies the overall pattern of the disease through the macroscopic image structure of the brain, while local feature extraction focuses on fine-grained brain areas to capture local abnormalities.

[0059] S202. Fusing the global brain features and the local brain features through an interactive attention mechanism to obtain fused visual features.

[0060] The interactive attention mechanism allows features from different sources to learn and calibrate each other, ultimately obtaining richer and more relevant comprehensive features. In this process, the interactive attention mechanism weighs the importance of features in different feature dimensions to ensure that the final fused features can better represent the functional abnormalities and structural changes of the brain to enhance the relevance in each feature.

[0061] Exemplarily, the calculation process of the interactive attention mechanism is as follows:

[0062] Add(cross-modal attention(G→L), cross-modal attention(L→G));

[0063]

[0064] G=A G V G , L=A L V L ;

[0065] F = G + L;

[0066] Among them, G represents the weighted global brain features, L represents the weighted local brain features, d represents the scaling factor, and Q G The query matrix Q represents the global brain features. L The query matrix K represents the local brain features. G The key matrix K represents the global brain features. L represents the key matrix of global brain features, F represents the fused visual features, V G represents the global brain features, V L Represents local brain features, A G Represents the calculated global attention, i.e. Attention G→L ; A L Represents the calculated local attention, i.e. Attention L→G .

[0067] In the above way, global and local features are weightedly learned through the interactive attention mechanism, and finally feature fusion is achieved by addition, so as to better capture global and local information.

[0068] S3. The brain statistical data is integrated with the fused visual features and then input into the trained diagnosis module for processing to obtain the diagnosis result at the current moment.

[0069] Specifically, first, the brain statistical feature matrix and the fused visual features are spliced ​​and tokens and position codes are added to construct a feature vector guided by brain statistical knowledge; then, the feature vector guided by brain statistical knowledge is input into the trained diagnostic module to obtain the diagnosis result of Alzheimer's disease of the participant at the current moment.

[0070] Here, the brain statistical feature matrix and the fused visual features are concatenated to obtain the fused comprehensive vector, which is implemented by expanding the predefined classification token (cls_token) to the input batch size and concatenating it with the comprehensive vector along the channel dimension. At the same time, in order to add the position encoding, the dimension of the position encoding matches the length of the fused comprehensive vector sequence, and the position encoding and the comprehensive vector are directly added. By adding classification tokens and position encoding, the global representation ability of sequence features is enhanced, and position and category information is provided for subsequent processing.

[0071] In this embodiment, the diagnosis module is a Transformer encoder.

[0072] The Transformer model can effectively identify complex brain functional network connection patterns by using its ability to capture long-range dependencies, thereby generating diagnostic results at the current moment. Through the multi-head self-attention mechanism, the Transformer can analyze the relationships between different brain regions and help identify abnormal connections related to neurodegenerative diseases.

[0073] S4. Guided by clinical data, combined with the current diagnosis results, and using the trained prediction module, obtain the disease progression at future times.

[0074] Specifically, the diagnosis result E at the current moment is fused with the encoded clinical data feature matrix G, and the fusion result is processed through the multi-head attention mechanism to predict the disease progression at future moments.

[0075] The clinical data feature matrix is ​​expressed as:

[0076] G=f clinical (H);

[0077] The fusion result is expressed as:

[0078] F = concat(E, G);

[0079] The disease progression at future time, that is, the prediction result at future time t+k, is expressed as:

[0080] prediction t+k =f pred (F);

[0081] Where H is the clinical data and k is the predicted time step.

[0082] Based on this, it not only helps predict the possible course of the disease, but also assists doctors in formulating long-term treatment plans, timely intervening and tracking changes in the disease.

[0083] S5. Perform explanatory analysis on the diagnostic and predictive results.

[0084] Furthermore, the Lasso regression method was used to verify the validity of the adopted brain statistical information. Then, the importance of clinical features was ranked through the self-attention mechanism in Transformer, and their importance in different fields was ranked and visualized, and the effect was verified in the real data set, as shown in Table 1.

[0085] Table 1. Validation table of the knowledge-guided multimodal auxiliary diagnosis and prediction system for Alzheimer's disease on real datasets

[0086]

[0087] In this embodiment, based on the guidance of multi-source information, a deep learning model integrating clinical knowledge and brain statistical knowledge is designed for the diagnosis and prediction of Alzheimer's disease. Through the fusion of multiple information, brain statistical knowledge is used to guide the extraction of dual-path visual features in the diagnosis stage, and clinical knowledge is applied to further guide the diagnostic output features in the prediction stage, thereby improving the prediction accuracy of future time nodes.

[0088] In order to balance the relationship between diagnosis and prediction tasks, ensure that the model can maintain consistency in predictions at multiple time points, and effectively find the best convergence solution of the model, as an implementation method, before the prediction module and the diagnosis module, it also includes:

[0089] A comprehensive loss function is constructed by combining the diagnosis stage loss, prediction stage loss and consistency loss to optimize the diagnosis module and prediction module.

[0090] In the diagnosis stage, the model generates the current diagnosis result D through dual-path feature extraction, interactive attention mechanism and Transformer processing. In order to measure the difference between the diagnosis result and the true label, the classification loss (usually cross entropy loss) is used to optimize the classification ability of the model. This loss can be expressed as:

[0091]

[0092] Where N represents the number of samples, y i represents the true label of sample i, Represents the output diagnosis probability distribution.

[0093] In the prediction stage, the model generates prediction outputs for multiple future time nodes by integrating diagnostic features E and encoded clinical features G and inputting them into the prediction module. The prediction loss is used to measure the difference between the future time node results predicted by the model and the actual disease progression. The prediction loss is usually measured using mean square error (MSE) or time series-related regression loss. This part can be defined as:

[0094]

[0095] Among them, T represents the number of predicted time nodes, y t+k and They represent the true label and model prediction value at the future t+k time node respectively.

[0096] By minimizing the prediction loss, the model can improve the accuracy of predicting the patient's future disease progression.

[0097] After diagnosis and prediction, the consistency loss can be obtained according to the diagnosis and prediction results, and the formula is:

[0098]

[0099] Among them, f i,j Represents the output result of the diagnostic module, Represents the output of the prediction module at the current moment. The mean square error (MSE) is used as the loss function to evaluate the consistency between the two modules. This choice allows to penalize higher error significance in the case of model prediction deviation.

[0100] Based on this, a comprehensive loss function is constructed:

[0101]

[0102] Among them, λ is the regularization coefficient, a non-negative real number (λ∈R+), which is used to adjust the model's focus on consistency loss during training and optimization.

[0103] To verify the optimal value of the hyperparameter λ, this embodiment is tested on a real data set. The experiment shows that the best result is achieved when λ=0.4. The verification results are shown in Table 2.

[0104] Table 2. Verification table of the optimal α value in the objective function. All indicators in the table are average values.

[0105] Accuracy MAUC SEN SPE λ=0.2 0.785 0.911 0.820 0.901 λ=0.4 0.801 0.920 0.819 0.911 λ=0.6 0.790 0.904 0.806 0.897 λ=0.8 0.779 0.910 0.798 0.889 λ=1 0.770 0.904 0.786 0.870

[0106] Embodiment 2

[0107] Combination Figure 2Based on the Alzheimer's disease diagnosis and prediction method based on multi-source knowledge guidance described in Example 1, this embodiment discloses an Alzheimer's disease diagnosis and prediction system based on multi-source knowledge guidance, including:

[0108] an acquisition module, configured to: acquire MRI images, brain statistics, and clinical data of the participant;

[0109] The diagnosis module is configured to: process the magnetic resonance image using a dual-path feature extraction strategy, obtain global brain features and local brain features, perform collaborative learning, and generate fused visual features; fuse the brain statistical data with the fused visual features and input them into the trained diagnosis module for processing to obtain the diagnosis result at the current moment;

[0110] The prediction module is configured to: use the clinical data as a guide, combine the diagnosis results at the current moment, and use the trained prediction module to obtain the disease progression at a future moment.

[0111] It should be noted that the acquisition module, diagnosis module and prediction module described above correspond to the steps in Embodiment 1, and the examples and application scenarios implemented by the modules and the corresponding steps are the same, but are not limited to the contents disclosed in Embodiment 1. It should be noted that the modules described above as part of the system can be executed in a computer system such as a set of computer executable instructions.

[0112] Embodiment 3

[0113] Embodiment 3 of the present invention provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps of the above-mentioned Alzheimer's disease diagnosis and prediction method based on multi-source knowledge guidance are completed.

[0114] Embodiment 4

[0115] Embodiment 4 of the present invention provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps of the above-mentioned Alzheimer's disease diagnosis and prediction method based on multi-source knowledge guidance are completed.

[0116] Embodiment 5

[0117] Embodiment 5 of the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the above-mentioned Alzheimer's disease diagnosis and prediction method based on multi-source knowledge guidance.

[0118] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0119] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0120] These computer program instructions can also be loaded onto a computer or other programmable data processing device, and a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0121] The description of each embodiment in the above embodiments has different emphases. For parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0122] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-source knowledge-guided Alzheimer's disease diagnosis and prediction method, characterized in that: include: Obtain participants’ MRI images, brain demographics, and clinical data; The magnetic resonance imaging image is processed by using a dual-path feature extraction strategy to obtain global brain features and local brain features and perform collaborative learning to generate fused visual features; The brain statistical data is fused with the fused visual features and then input into a trained diagnosis module for processing to obtain a diagnosis result at the current moment; Guided by the clinical data and combined with the diagnosis results at the current moment, the disease progression at the future moment is obtained using the trained prediction module.

2. The Alzheimer's disease diagnosis and prediction method based on multi-source knowledge guidance as claimed in claim 1, characterized in that: The method of processing the magnetic resonance image using a dual-path feature extraction strategy to obtain global brain features and local brain features includes: Processing the magnetic resonance imaging images using an improved ShuffleNetv2 model to obtain global brain features; Processing the magnetic resonance imaging using a 3D CNN network to obtain local brain features; Among them, the improved ShuffleNetv2 model and the 3D CNN network are set up in parallel.

3. The Alzheimer's disease diagnosis and prediction method based on multi-source knowledge guidance as claimed in claim 1, characterized in that: The global brain features and local brain features are collaboratively learned to generate fused visual features. Specifically, the global brain features and local brain features are fused through an interactive attention mechanism to obtain fused visual features.

4. The Alzheimer's disease diagnosis and prediction method based on multi-source knowledge guidance as claimed in claim 1, characterized in that: The fusion of the brain statistical data with the fused visual features is specifically as follows: the encoded brain statistical data is spliced ​​with the fused visual features, and tokens and position codes are added to construct a feature vector guided by brain statistical knowledge.

5. The Alzheimer's disease diagnosis and prediction method based on multi-source knowledge guidance as claimed in claim 1, characterized in that: Guided by the clinical data, combined with the diagnosis results at the current moment, and using the trained prediction module, the disease progression at the future moment is obtained. Specifically, the diagnosis results at the current moment are fused with the encoded clinical data, and the fusion results are processed through a multi-head attention mechanism to predict the disease progression at the future moment.

6. The Alzheimer's disease diagnosis and prediction method based on multi-source knowledge guidance as claimed in claim 1, characterized in that: A comprehensive loss function is constructed by combining the diagnosis stage loss, the prediction stage loss and the consistency loss to optimize the diagnosis module and the prediction module.

7. Alzheimer's disease diagnosis and prediction system based on multi-source knowledge guidance, characterized by: include: an acquisition module, configured to: acquire MRI images, brain statistics, and clinical data of the participant; The diagnosis module is configured to: process the magnetic resonance image using a dual-path feature extraction strategy, obtain global brain features and local brain features, perform collaborative learning, and generate fused visual features; fuse the brain statistical data with the fused visual features and input them into the trained diagnosis module for processing to obtain the diagnosis result at the current moment; The prediction module is configured to: use the clinical data as a guide, combine the diagnosis results at the current moment, and use the trained prediction module to obtain the disease progression at a future moment.

8. An electronic device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the Alzheimer's disease diagnosis and prediction method based on multi-source knowledge guidance as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instruction is executed by a processor, the steps of the multi-source knowledge-guided Alzheimer's disease diagnosis and prediction method described in any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the steps of the multi-source knowledge-guided Alzheimer's disease diagnosis and prediction method described in any one of claims 1 to 6 are implemented.

Images