Early auxiliary diagnosis and prediction method and system for neurodegenerative diseases for clinical magnetic resonance imaging

Abstract

The present application relates to the technical field of artificial intelligence assisted diagnosis, and discloses a method and system for early auxiliary diagnosis and prediction of neurodegenerative diseases in clinical magnetic resonance imaging, which is used to solve the problem that it is difficult to realize precise and automatic identification and progress prediction of neurodegenerative diseases in the early stage through clinical conventional magnetic resonance imaging examination. The method comprises the following steps: collecting multi-modal magnetic resonance imaging as training samples, extracting first brain structure morphological quantitative values by dimension reduction and simplification, labeling diagnosis category labels and training a first classification model, and extracting a feature coefficient matrix; based on the feature weight coefficients contained in the feature coefficient matrix, screening out core driven brain areas corresponding to different diagnosis categories, and constructing a second classification model; obtaining clinical magnetic resonance images of early or precursor expectant examinees, extracting second brain structure morphological quantitative values, calling the second classification model to calculate probability scores of different diagnosis categories, and outputting auxiliary diagnosis or progress prediction results based on the probability scores.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence-assisted diagnostic technology, and in particular to a method, system, electronic device, computer storage medium, and computer program product for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging. Background Technology

[0002] Neurodegenerative diseases are a group of illnesses caused by the progressive loss of neurons or myelin sheaths, characterized primarily by cognitive or physical dysfunction. Currently, treatment for these diseases mainly focuses on multi-target interventions to slow disease progression, and early diagnosis is crucial for controlling disease development, formulating appropriate treatment plans, and improving patient prognosis.

[0003] Parkinson's disease (PD) and atypical Parkinson's syndromes (including multiple system atrophy, MSA) are common types of Parkinson's syndromes in clinical practice. Their clinical manifestations highly overlap in the early stages (especially in the first 3-5 years of the disease course), often presenting with similar symptoms such as bradykinesia, rigidity, and tremor. However, their pathological nature, treatment response, and prognosis are drastically different. Therefore, accurate differentiation between PD and MSA in the early stages is directly related to the selection of treatment plans, prognosis assessment, and long-term patient management.

[0004] In recent years, some studies have found that high-resolution T1-weighted structural MRI (Nuclear Magnetic Resonance Imaging), diffusion tensor imaging (DTI), and resting-state functional MRI (rs-fMRI) have demonstrated excellent diagnostic efficacy in differentiating PD from other Parkinson's syndromes. However, the clinical application of these techniques is limited. They not only rely on high-field MRI scanners, but also require staff with specialized skills in sequence optimization, data acquisition, and analysis. Furthermore, they suffer from drawbacks such as long scan times (generally 40-60 minutes) and high examination costs. Additionally, they require high patient cooperation, demanding that patients tolerate prolonged examinations and remain immobile. All these factors make it difficult to apply these studies to routine clinical practice, and currently, there is no brain MRI-assisted diagnostic model that can be directly applied to clinical scenarios.

[0005] Therefore, even though most patients undergo at least one brain MRI scan at initial diagnosis, the low-resolution two-dimensional MRI images commonly used in clinical practice have limited ability to display changes in brain microstructure, resulting in low accuracy in assessment. This is especially true when patients are in the prodromal stage of disease, making it difficult to accurately predict the risk of differentiation and identify its direction. Therefore, there is an urgent clinical need for a universal method that can be directly applied to routine clinical MRI images to achieve accurate and automated identification of neurodegenerative diseases in their early stages without altering existing scanning protocols. Summary of the Invention

[0006] The main objective of this invention is to solve the technical problem in the prior art that it is difficult to achieve accurate and automated identification and progression prediction of neurodegenerative diseases in the early stages through routine clinical magnetic resonance imaging examinations.

[0007] The first aspect of this invention provides a method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging, comprising:

[0008] Multimodal magnetic resonance imaging (MRI) images were collected as training samples, and the quantitative values ​​of the first brain structure morphology of each training sample were extracted by dimensionality reduction and simplification; wherein, the training samples were labeled with diagnostic category tags for different neurodegenerative diseases;

[0009] A first classification model is trained based on the quantitative values ​​of the first brain structure morphology of each training sample and the diagnostic category label, and the feature coefficient matrix corresponding to the trained first classification model is extracted.

[0010] Based on the feature weight coefficients of different diagnostic categories contained in the feature coefficient matrix, the core driving brain regions corresponding to different diagnostic categories are selected.

[0011] A second classification model is constructed based on the core driving brain regions and their corresponding feature weight coefficients.

[0012] Acquire clinical magnetic resonance images of subjects in the early or prodromal stage, and extract quantitative values ​​of the second brain structure morphology from the clinical magnetic resonance images;

[0013] The second classification model is invoked to calculate the probability score of different diagnostic categories based on the quantitative value of the second brain structure morphology, and the auxiliary diagnosis or progression prediction result is output based on the probability score.

[0014] Optionally, in a first implementation of the first aspect of the present invention, the diagnostic categories include Parkinson's disease and multiple system atrophy; the second classification model consists of a simplified Parkinson's disease classifier and a simplified multiple system atrophy classifier.

[0015] The step of calling the second classification model to calculate probability scores for different diagnostic categories based on the quantitative values ​​of the second brain structure morphology, and outputting auxiliary diagnostic or progression prediction results based on the probability scores includes:

[0016] Based on the second quantitative value of brain structure morphology, the probability scores of Parkinson's disease and multiple system atrophy are calculated by calling the simplified Parkinson's disease classifier and the simplified multiple system atrophy classifier, respectively.

[0017] The diagnostic category corresponding to the highest probability score is output as the auxiliary diagnostic result.

[0018] Optionally, in a second implementation of the first aspect of the present invention, the first classification model consists of a full-feature Parkinson's disease classifier and a full-feature multisystem atrophy classifier;

[0019] The process of selecting the core driving brain regions corresponding to different diagnostic categories based on the feature weight coefficients of different diagnostic categories contained in the feature coefficient matrix includes:

[0020] Extract the Parkinson's disease feature coefficient matrix of the full-feature Parkinson's disease classifier under optimal hyperparameters and the multi-system atrophy feature coefficient matrix of the full-feature multi-system atrophy classifier under optimal hyperparameters respectively;

[0021] After removing the intercept terms and zero coefficients from the Parkinson's disease characteristic coefficient matrix and the multisystem atrophy characteristic coefficient matrix, multiple non-zero characteristic coefficients of Parkinson's disease and multiple non-zero characteristic coefficients of multisystem atrophy are obtained.

[0022] The absolute values ​​of the non-zero characteristic coefficients of Parkinson's disease and the non-zero characteristic coefficients of multiple system atrophy are sorted in descending order, and the corresponding brain regions within the preset ranking range are extracted to obtain the core driving brain regions of Parkinson's disease and multiple system atrophy.

[0023] Optionally, in a third implementation of the first aspect of the present invention, the dimensionality reduction and simplification extraction of the first brain structure morphological quantitative values ​​of each training sample includes:

[0024] Based on the spatial resolution characteristics of each training sample, the first extraction branch and the second extraction branch in the dual-stream structured data algorithm are used to reconstruct the cortex of training samples with different resolutions. The quantitative values ​​of the first brain structure morphology of each training sample are obtained through dimensionality reduction simplification, measurement and calculation.

[0025] The second quantitative brain structural morphology values ​​extracted from the clinical magnetic resonance images include:

[0026] Based on the second processing branch of the dual-stream structured data algorithm, cortical reconstruction, topological correction, and thickness estimation are performed on the clinical magnetic resonance images to extract quantitative morphological values ​​of the second brain structure of the core driving brain regions corresponding to different diagnostic categories in the clinical magnetic resonance images.

[0027] Optionally, in a fourth implementation of the first aspect of the present invention, the early auxiliary diagnosis further includes predicting the differentiation risk of the examinee in the prodromal stage of neurodegenerative disease.

[0028] When the subject is in the prodromal stage of a neurodegenerative disease, after acquiring the subject's clinical magnetic resonance imaging (MRI) images and extracting the quantitative values ​​of the second brain structure morphology from the MRI images, the procedure further includes:

[0029] The second classification model is invoked to calculate the total feature score corresponding to different diagnostic categories based on the quantitative value of the second brain structure morphology, and the differentiation risk prediction result is generated.

[0030] Among them, the examinee in the prodromal stage of the neurodegenerative disease is the examinee who has symptoms of idiopathic rapid eye movement sleep behavior disorder.

[0031] Optionally, in a fifth implementation of the first aspect of the present invention, the total feature scores corresponding to the different diagnostic categories include the total feature score for Parkinson's disease and the total feature score for multiple system atrophy.

[0032] The step of calling the second classification model to calculate the feature total score corresponding to different disease categories based on the second brain structure morphological quantitative value, and generating differentiation risk prediction results includes:

[0033] The second classification model was invoked to calculate the total score of Parkinson's disease characteristics and the total score of multiple system atrophy characteristics based on the second quantitative value of brain structure morphology.

[0034] Obtain a preset differentiation prediction threshold, and divide the differentiation trajectory based on the total score of Parkinson's disease features and the total score of multiple system atrophy features;

[0035] If the total score of Parkinson's disease features is higher than the differentiation prediction threshold and the total score of multiple system atrophy features is lower than the differentiation prediction threshold, it indicates that there is a typical risk of Parkinson's disease.

[0036] If the total score of the multi-system atrophy feature is higher than the differentiation prediction threshold, it indicates that there is a risk of multi-system atrophy.

[0037] If both the total score for Parkinson's disease characteristics and the total score for multiple system atrophy characteristics are lower than the differentiation prediction threshold, it indicates that the current macrostructure has not undergone specific irreversible changes, and the risk of differentiation is uncertain.

[0038] Optionally, in a sixth implementation of the first aspect of the present invention, the step of calling the second classification model to calculate the total score of Parkinson's disease characteristics and the total score of multiple system atrophy characteristics based on the second quantitative value of brain structure morphology includes:

[0039] Obtain the weight coefficients of the frozen core driving brain regions related to Parkinson's disease in the second classification model, and obtain the total score of Parkinson's disease characteristics by weighted summation based on the second brain structure morphological quantitative value and the weight coefficients.

[0040] Obtain the weight coefficients of the core driving brain regions frozen and associated with multiple system atrophy in the second classification model, call the second classification model to perform a weighted summation based on the second brain structure morphological quantitative value and the weight coefficients, and obtain the total score of multiple system atrophy features.

[0041] Optionally, in a seventh implementation of the first aspect of the present invention, the method further includes:

[0042] We obtain the structural connectivity features derived from diffusion tensor imaging and the functional connectivity features derived from resting-state functional magnetic resonance imaging of the training samples, and construct a multimodal high-dimensional feature matrix by combining the high-resolution T1-weighted structural features of the training samples.

[0043] A multimodal benchmark classifier is established by combining a mixed penalty term and a 10-fold hierarchical cross-validation strategy, and the classification performance of the multimodal benchmark classifier is evaluated based on a pre-set test set.

[0044] After constructing the second classification model based on the core driving brain regions and their corresponding feature weight coefficients, the method further includes:

[0045] The prediction accuracy of the second classification model is evaluated based on a pre-set test set;

[0046] The feasibility assessment range is calculated based on the prediction accuracy of the multimodal baseline classifier. When the prediction accuracy of the second classification model is within the feasibility assessment range, the parameters of the second classification model are frozen.

[0047] A second aspect of the present invention provides an early auxiliary diagnostic and prediction system for neurodegenerative diseases using clinical magnetic resonance imaging, comprising:

[0048] The data acquisition module is used to acquire multimodal magnetic resonance images as training samples and to simplify the extraction of quantitative values ​​of the first brain structure morphology of each training sample by dimensionality reduction; wherein, the training samples are labeled with diagnostic category tags for different neurodegenerative diseases;

[0049] The region extraction module is used to train a first classification model based on the quantitative values ​​of the first brain structure morphology of each training sample and the disease category label, and to extract the feature coefficient matrix corresponding to the trained first classification model; based on the feature weight coefficients of different diagnostic categories contained in the feature coefficient matrix, the core driving brain regions corresponding to different diagnostic categories are selected.

[0050] The model building module is used to build a second classification model based on the core driving brain regions and their corresponding feature weight coefficients;

[0051] The auxiliary diagnostic module is used to extract the second brain structure morphological quantitative value from the clinical magnetic resonance image of the patient to be examined; it calls the second classification model to calculate the probability score of different diagnostic categories based on the second brain structure morphological quantitative value, and outputs the auxiliary diagnosis or progression prediction result based on the probability score.

[0052] A third aspect of the present invention provides an early auxiliary diagnosis and prediction device for neurodegenerative diseases using clinical magnetic resonance imaging, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the early auxiliary diagnosis and prediction device for neurodegenerative diseases using clinical magnetic resonance imaging to perform the steps of the above-described method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging.

[0053] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the steps of the above-described method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging.

[0054] A fifth aspect of the present invention provides a computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the above-described method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging.

[0055] The technical solution provided by this invention involves acquiring multimodal magnetic resonance imaging (MRI) images as training samples and simplifying the extraction of quantitative values ​​of the first brain structure morphology for each training sample through dimensionality reduction. The training samples are labeled with diagnostic category tags for different neurodegenerative diseases. A first classification model is trained based on the quantitative values ​​of the first brain structure morphology and the diagnostic category tags of each training sample, and the feature coefficient matrix corresponding to the trained first classification model is extracted. Based on the feature weight coefficients of different diagnostic categories contained in the feature coefficient matrix, core driving brain regions corresponding to different diagnostic categories are selected. A second classification model is constructed based on the core driving brain regions and their corresponding feature weight coefficients. Clinical MRI images of subjects in the early or prodromal stages are acquired, and the quantitative values ​​of the second brain structure morphology of the clinical MRI images are extracted. The second classification model is called to calculate probability scores for different diagnostic categories based on the quantitative values ​​of the second brain structure morphology, and the auxiliary diagnostic or progression prediction results are output based on the probability scores. This method can use routine clinical MRI images to achieve accurate and automated identification of neurodegenerative diseases and prediction of progression risk in the early stages of neurodegenerative diseases, improving the generalization ability of diagnostic assistance technologies. The system, electronic device, computer-readable storage medium, and computer program product provided by this invention also solve the corresponding technical problems. Attached Figure Description

[0056] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0057] Figure 1 This is a flowchart illustrating the first embodiment of a method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging, as described in this invention.

[0058] Figure 2 This is a flowchart illustrating the second embodiment of the method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging in this invention.

[0059] Figure 3 This is a specific example diagram illustrating the model construction of a second embodiment of a method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging, as described in this invention.

[0060] Figure 4 This is a feasibility effect diagram of each model under a specific validation set in the second embodiment of the method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging in the present invention.

[0061] Figure 5This is a schematic diagram of selected features in a specific example of a second embodiment of a method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging in this invention.

[0062] Figure 6 This is a schematic diagram of a specific iRBD stratified prediction result in the second embodiment of the method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging in this invention.

[0063] Figure 7 This is a schematic diagram of an embodiment of the early auxiliary diagnosis and prediction system for neurodegenerative diseases using clinical magnetic resonance imaging in this invention.

[0064] Figure 8 This is a schematic diagram of an embodiment of a device for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging, as described in this invention.

[0065] Figure 9 This is a schematic diagram illustrating the principle of a computer-readable medium according to an embodiment of the present invention. Detailed Implementation

[0066] Exemplary embodiments of the invention will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limiting the invention to the embodiments set forth herein. Rather, these exemplary embodiments are provided to make the invention more comprehensive and complete, and to facilitate a full communication of the inventive concept to those skilled in the art. The same reference numerals in the drawings denote the same or similar elements, components, or parts, and therefore repeated descriptions of them will be omitted.

[0067] Subject to the technical concept of this invention, the features, structures, characteristics or other details described in a particular embodiment may be combined in one or more other embodiments in a suitable manner.

[0068] In the description of specific embodiments, the features, structures, characteristics, or other details described in this invention are intended to enable those skilled in the art to fully understand the embodiments. However, it is not excluded that those skilled in the art can practice the technical solutions of this invention without one or more of the specific features, structures, characteristics, or other details.

[0069] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0070] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0071] The terms “and / or” or “and / or” include all combinations of any one or more of the listed items.

[0072] See Figure 1 The first embodiment of the method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging in this invention includes:

[0073] It is understood that the executing entity of this invention can be a device or system for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging, or it can be a terminal or a server; the specific implementation is not limited here. This embodiment of the invention will be described using a server as an example.

[0074] Before performing specific predictions, it is necessary to collect multimodal magnetic resonance imaging (MRI) data and related information to construct training samples. Based on these training samples, a first classification model is constructed and trained to predict the diagnostic category label corresponding to the sample based on the quantitative values ​​of brain structural morphology from multimodal MRI. In this embodiment, after obtaining the first classification model, the feature weight coefficients after training are optimized based on the first classification model to determine the core driving brain regions for different diagnostic categories. Next, based on the core driving brain regions and their feature weight coefficients, a second classification model is constructed to predict the diagnostic category based on the quantitative values ​​of brain structural morphology of specific core driving brain regions in clinical MRI images. Furthermore, this embodiment also includes constructing a multimodal model based on multimodal information from multimodal MRI to provide an accurate anchoring benchmark, and evaluating the classification accuracy of the second classification model based on the prediction accuracy of the multimodal model on the test set. When the classification accuracy of the second classification model meets the preset requirements, auxiliary diagnostic results can be output based on the second classification model, or the subsequent differentiation risk of patients in the prodromal stage can be predicted. Moreover, the diagnostic and prediction method described in this embodiment is an information processing method in which all steps are implemented by a computer or other device. See [link to relevant documentation]. Figure 1 The following steps will be explained in detail.

[0075] S101. Collect multimodal magnetic resonance imaging as training samples, and simplify the extraction of the first brain structure morphological quantitative value of each training sample by dimensionality reduction, thereby reducing and optimizing the multimodal magnetic resonance imaging to a single-modal brain morphological structure image.

[0076] This step begins with the data acquisition process. During this process, multiple modal magnetic resonance imaging (MRI) images at different resolutions are acquired to construct training and testing samples. The acquired MRI images are primarily research-grade images, including high-resolution multimodal 3D T1-weighted imaging, diffusion tensor imaging (DTI), and resting-state functional magnetic resonance imaging (rs-fMRI).

[0077] In one specific implementation, the collected samples also include standard 2D thick-slice T1-weighted imaging for clinical use in constructing a clinical validation set, which can be used to verify the generalization ability of the method in this embodiment.

[0078] Simultaneously, based on the specific circumstances of the collected samples, diagnostic category labels for neurodegenerative diseases are assigned to the training samples. In one specific implementation, the diagnostic category labels include healthy controls (HC), patients with idiopathic rapid eye movement sleep behavior disorder (iRBD) and their conversion outcomes, patients with Parkinson's disease (PD), and patients with multiple system atrophy (MSA). For simplicity, patients with idiopathic rapid eye movement sleep behavior disorder are referred to as iRBD patients, patients with Parkinson's disease as PD patients, and patients with multiple system atrophy as MSA patients.

[0079] After sample collection, this step also includes obtaining quantitative values ​​of brain structural morphology for each sample. In a specific embodiment, since the ultimate goal of this approach is to assist in the diagnosis and prediction of neurodegenerative diseases through clinical magnetic resonance imaging, this embodiment simplifies multimodal information to a single modality, using quantitative values ​​of brain structural morphology that can be obtained from various modalities of magnetic resonance imaging to perform subsequent training and prediction processes; for example, the thickness, surface area, and subcortical volume of the cerebral cortex are used as the quantitative values ​​of brain structural morphology after dimensionality reduction and simplification.

[0080] Specifically, when extracting quantitative values ​​of brain structural morphology, due to the differences in spatial resolution between different types of magnetic resonance imaging image samples, a dual-stream structural data algorithm is constructed in this embodiment to achieve better alignment. This dual-stream structural data algorithm includes two extraction branches with different information: a first extraction branch for extracting higher-resolution morphological information and a second extraction branch suitable for extracting lower-resolution information.

[0081] In one specific implementation, the first extraction branch can be implemented based on the recon-all workflow of FreeSurfer software. The recon-all workflow receives high-quality, high-resolution magnetic resonance imaging data, performs preprocessing operations such as skull removal and inhomogeneity correction on the image data, and extracts information about the brain regions. After information extraction, tissue segmentation and surface construction are performed, and cortical thickness and cortical surface area are calculated. Furthermore, it includes partition mapping based on existing professional brain functional atlases (such as the Schaefer atlas) to obtain information such as cortical thickness and cortical surface area for different brain regions. In addition, the first extraction branch also calculates the subcortical volume of each brain region based on the NextBrain probabilistic histological atlas.

[0082] In one specific implementation, to accurately obtain quantitative information on brain structural morphology based on clinically common heterogeneous (anisotropic, slice thickness 5-6 mm) clinical MRI data, the second extraction branch in this embodiment is based on a hybrid scheme combining deep learning and classical geometric processing. It uses a pre-trained convolutional neural network (CNN) to predict voxel grids from lower-quality clinical MRI data, thereby inferring more accurate cortical structural information. Based on the predicted voxel grids, classical geometric algorithms are used to determine the accurate location of the cortical surface, thus reconstructing the cortical surface. Calculations are then performed based on the reconstructed cortical surface to obtain the cortical thickness and surface area of ​​different brain regions, as well as the subcortical volume. For example, the second extraction branch can be implemented based on a recon-all-clinical workflow.

[0083] Based on the first and second extraction branches in the dual-stream structured data algorithm, reliable information can be extracted from massive heterogeneous clinical MRI data, achieving information alignment of data with different resolutions.

[0084] S102. Train the first classification model based on the quantitative values ​​of the first brain structure morphology and the diagnostic category labels of each training sample, and extract the feature coefficient matrix corresponding to the trained first classification model.

[0085] After obtaining each training sample and its related information, a training dataset is constructed based on this information, and a first classification model is trained based on the training dataset. Furthermore, when training the first classification model, the diagnostic category labels selected for the samples include healthy controls (HC), PD patients, and MSA patients.

[0086] In this embodiment, the first classification model consists of two independent classifiers. These classifiers can be binary classification models built upon Generalized Linear Models (GLMs), including a full-feature Parkinson's disease classifier for distinguishing between "PD patients" and "healthy controls + MSA patients," and a full-feature multisystem atrophy classifier for distinguishing between "MSA patients" and "healthy controls + PD patients." Both independent classifiers focus on distinguishing the specific radiological boundary between a single target disease and all other clinical states. Specifically, a nested cross-validation framework can be used for high-weight selection of one-to-many features and GLMNET hyperparameter tuning to complete the training of the first classification model.

[0087] After training is complete, the feature coefficients matrix of the first classification model under the optimal hyperparameters is extracted.

[0088] Before training the first classification model, a multimodal model that provides an accurate anchoring benchmark can be pre-built, and the feasibility assessment range can be calculated based on the prediction accuracy of the multimodal model on the test set. Training ends when the prediction accuracy of the first classification model reaches the feasibility assessment range.

[0089] S103. Based on the feature weight coefficients of different diagnostic categories contained in the feature coefficient matrix, the core driving brain regions corresponding to different diagnostic categories are selected.

[0090] After training the two classifiers in the first classification model, the Parkinson's disease feature coefficient matrix of the full-feature Parkinson's disease classifier under the optimal hyperparameters and the multi-system atrophy feature coefficient matrix of the full-feature multi-system atrophy classifier under the optimal hyperparameters are extracted respectively. After removing the intercept terms and zero coefficient features from the Parkinson's disease feature coefficient matrix and the multi-system atrophy feature coefficient matrix, multiple non-zero feature coefficients of Parkinson's disease and multiple non-zero feature coefficients of multi-system atrophy are obtained.

[0091] The absolute values ​​of the non-zero characteristic coefficients for Parkinson's disease and multiple system atrophy are sorted in descending order to obtain the characteristic sequences for Parkinson's disease and multiple system atrophy. Brain regions and characteristic coefficients corresponding to a predetermined ranking range are then extracted from these two sequences to obtain the core driving brain regions for Parkinson's disease and multiple system atrophy. It is possible for overlapping regions to exist between the two sets of core driving brain regions.

[0092] In a preferred embodiment, the preset ranking range is Top 10 (i.e., the top 10), and the Top 10 core driving brain regions for Parkinson's disease and the Top 10 core driving brain regions for multiple system atrophy are obtained respectively, resulting in a total of Top 20 core driving brain regions (wherein, the Top 20 core driving brain regions may have overlapping areas because they correspond to different diagnostic types).

[0093] S104. Construct a second classification model based on the core driving brain regions and their corresponding feature weight coefficients;

[0094] After obtaining the core driving brain regions, no additional artificial weights are introduced. Instead, a second classification model is reconstructed based on the core driving brain regions and their corresponding feature weight coefficients. This second classification model also contains two classifiers. Each classifier focuses on one of the top 10 core driving brain regions for Parkinson's disease and the other on the other of the top 10 core driving brain regions for multiple system atrophy (MSA), rather than all brain regions. In this embodiment, these are referred to as the simplified Parkinson's disease classifier and the simplified MSA classifier, and they are used to predict Parkinson's disease and MSA, respectively. Furthermore, the input to the second classification model can be set to low-resolution clinical magnetic resonance images.

[0095] In a preferred embodiment, a preset validation set is obtained, and the classification accuracy of the second classification model is calculated based on the validation set; wherein the sample images included in the validation set are clinical magnetic resonance images.

[0096] Based on the prediction accuracy calculation of the multimodal model constructed in S102, which provides accurate anchoring benchmarks, the feasibility assessment range is determined. When the prediction accuracy of the second classification model falls within the feasibility assessment range, the hyperparameters of the second classification model are frozen, completing the construction of the second classification model. The frozen hyperparameters refer to the 10 feature weight coefficients of the Top 10 core driving brain regions for Parkinson's disease in the simplified Parkinson's disease classifier of the second classification model, and the 10 feature weight coefficients of the Top 10 core driving brain regions for multiple system atrophy in the simplified multisystem atrophy classifier.

[0097] S105. Obtain clinical magnetic resonance images of the subject in the early or prodromal stage, and extract quantitative values ​​of the second brain structure morphology from the clinical magnetic resonance images.

[0098] In this embodiment, the constructed second classification model can accurately determine whether a patient has Parkinson's disease and multiple system atrophy based on clinical magnetic resonance imaging in a clinical setting, and provide diagnostic opinions. In addition to determining whether individuals in the early or prodromal stages of Parkinson's disease and multiple system atrophy have these conditions, it can also be used for the early diagnosis and risk prediction of neurodegenerative diseases.

[0099] The second model requires clinical magnetic resonance imaging (MRI) images for both diagnosis and risk prediction. Subsequently, quantitative values ​​of the corresponding second brain structure morphology are extracted based on the MRI images. As described in step S101 of this embodiment, the second extraction branch based on the dual-stream structural data algorithm obtains cortical thickness, cortical surface area, and subcortical volume as quantitative values ​​of the second brain structure morphology.

[0100] S106. Call the second classification model to calculate the probability score of different diagnostic categories based on the quantitative value of the second brain structure morphology, and output the auxiliary diagnosis or progression prediction results based on the probability score.

[0101] In this embodiment, the second quantitative value of brain structure morphology is the feature vector of two Top 10 core driving brain regions. The second classification model includes a simplified Parkinson's disease classifier and a simplified multiple system atrophy (MSA) classifier. The simplified Parkinson's disease classifier calculates a probability score indicating the subject has Parkinson's disease and a linear score indicating the subject is healthy, based on the 10 frozen weights and the feature vectors of the Top 10 core driving brain regions for Parkinson's disease. The simplified MSA classifier calculates a probability score indicating the subject has multiple system atrophy and a linear score indicating the subject is healthy, based on the 10 frozen weights and the feature vectors of the Top 10 core driving brain regions for multiple system atrophy. After obtaining the two linear scores, the three scores are converted into probability scores for different diagnostic categories using the softmax function in the second classification model. The diagnostic category corresponding to the highest probability score is used as the auxiliary diagnostic result.

[0102] In a preferred embodiment, the early auxiliary diagnosis further includes predicting the differentiation risk of the examinee in the prodromal stage of neurodegenerative disease by calling a second classification model to calculate the feature total score corresponding to different disease categories based on the quantitative value of second brain structure morphology, and generating differentiation risk prediction results.

[0103] When the clinical MRI images are from patients in the prodromal phase of neurodegenerative diseases (e.g., patients with iRBD symptoms), the second classification model is invoked to calculate the total score for Parkinson's disease characteristics and the total score for multiple system atrophy characteristics based on the second quantitative brain structure morphology values. Specifically, when calculating the total score, the weight coefficients of the frozen core driving brain regions related to Parkinson's disease in the second classification model are obtained. After weighted summation based on the second quantitative brain structure morphology values ​​and the weight coefficients, the total score for Parkinson's disease characteristics is obtained. Similarly, the weight coefficients of the frozen core driving brain regions related to multiple system atrophy in the second classification model are obtained. After weighted summation based on the second classification model and the second quantitative brain structure morphology values, the total score for multiple system atrophy characteristics is obtained.

[0104] Obtain a preset differentiation prediction threshold, and divide the differentiation trajectory based on the total score of Parkinson's disease features and the total score of multiple system atrophy features. During the division:

[0105] (1) If the total score of Parkinson’s disease features is higher than the differentiation prediction threshold and the total score of multiple system atrophy features is lower than the differentiation prediction threshold, then it indicates that there is a typical risk of Parkinson’s disease.

[0106] (2) If the total score of multiple system atrophy features is higher than the differentiation prediction threshold, it indicates an atypical risk of Parkinson's disease, which suggests the presence of multiple system atrophy risk.

[0107] (3) If the total score of Parkinson’s disease characteristics and the total score of multiple system atrophy characteristics are both lower than the differentiation prediction threshold, it indicates that the current macrostructure has not undergone specific irreversible changes and the risk of differentiation is uncertain.

[0108] The solution provided in this embodiment of the invention can use routine clinical magnetic resonance imaging to achieve accurate and automated identification of neurodegenerative diseases in the early stages, improve the generalization ability of diagnostic aid technology, and ensure diagnostic efficiency while reducing the requirements for scanning accuracy. Furthermore, it can also accurately predict the differentiation risk and differentiation path of neurodegenerative diseases in the prodromal stage, predict the window period in advance, improve the accuracy of prediction, and reduce the difficulty and cost of examination and prediction.

[0109] See Figure 2-6 The second embodiment of the method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging described in this invention is an information processing method in which all steps are implemented by a computer or other device. The specific scheme includes:

[0110] S201, Data Acquisition and Processing;

[0111] In this embodiment, samples were collected from multiple research centers and public databases as training and testing samples. Based on the specific circumstances of the training samples, they were labeled with diagnostic categories of different neurodegenerative diseases, including healthy controls (HD), Parkinson's disease (PD), and multiple system atrophy (MSA). Patients with Parkinson's disease (PD) and multiple system atrophy (MSA) who had previously presented with iRBD were also labeled.

[0112] During the sample collection process, the ethics committees of all participating research centers approved the research protocol, and all research subjects signed written informed consent forms. The data obtained was licensed data that did not contain sensitive personal information. All research subjects met the following inclusion criteria: (1) age 40-80 years; (2) no signs of dementia; (3) no history of intracranial surgery or traumatic brain injury; (4) no mental illness; (5) no alcohol use disorder; and (6) no history of other neurological diseases.

[0113] During the collection of training samples, the healthy control group (HC) was recruited from community members without neurological diseases. These individuals also signed written informed consent forms, and the data obtained was not sensitive to personal information. Patients with typical iRBD symptoms underwent video polysomnography using the Compumedics E-series EEG / polysomnography recording system (Compumedics Ltd). The diagnosis of iRBD was confirmed by both clinical assessment and polysomnographic evidence according to the International Classification of Sleep Disorders-3 (ICSD-3). These patients were also evaluated by neurologists to rule out potential Parkinson's syndrome. Parkinson's disease (PD) patients were diagnosed according to the Movement Disorders Society's clinical diagnostic criteria, while multiple system atrophy (MSA) patients were diagnosed and labeled according to existing clinical standards. All PD / MSA patients were evaluated by at least two senior movement disorder specialists using the Movement Disorders Society Unified Parkinson's Disease Rating Scale Part III (MDS-UPDRS III) and HY staging, with early to mid-stage disease (Hoehn and Yahr stages ≤3) as the valid sample.

[0114] In one specific implementation, the collected samples were divided into several cohorts, including an exploratory cohort, an external research-grade cohort, and an external clinical MRI cohort. The total research cohort (n=969) adopted a stratified discovery-validation-translation framework. The exploratory cohort at site A (n=332) included healthy controls, Parkinson's disease patients, and multiple system atrophy patients. All participants underwent T1-weighted imaging, DTI, and rs-fMRI high-resolution multimodal MRI scans, and were randomly divided into a training set and an internal independent test set at an 8:2 ratio; where n represents the number of subjects in the cohort. Furthermore, two independent cohorts were used to evaluate the efficacy of the fixed-parameter model: one was the external research-grade cohort (n=227), with data from centers B and C, using 1mm slice thickness 3D isotropic T1-weighted imaging; the other was the external clinical MRI cohort (n=303), with data from centers A, C, D, and E, using 5.0-6.0mm slice thickness 2D T1-weighted imaging.

[0115] In this study, the subjects provided by centers A and C were sub-protocol, with no overlap, and all subjects were included in a single cohort. To expand the application value of the model in the prodromal phase, this study additionally included 107 iRBD patients to form an exploratory cohort, which included a longitudinal subgroup of 25 phenotypic conversion patients (20 from the PPMI database and 5 from follow-up at center A). All 25 subjects were diagnosed with PD, and this was used as a clinical reference standard to verify the structural imaging evolution trajectory of PD-like diseases.

[0116] Images were acquired using a 3.0T MRI system. The specific scanning protocols for each cohort were as follows: the exploratory cohort used high-resolution multimodal 3D T1-weighted imaging, diffusion tensor imaging (DTI), and resting-state functional magnetic resonance imaging (rs-fMRI); the external research-grade cohort used high-resolution 3D T1-weighted imaging; and the external clinical MRI cohort used standard 2D thick-slice T1-weighted imaging. Detailed acquisition parameters are shown in Tables 1 and 2 below.

[0117] Table 1: Standardized High-Resolution Protocols

[0118]

[0119] Table 2: Heterogeneous clinical thick-slice T1 scanning protocols in external clinical MRI cohorts

[0120]

[0121] In Tables 1 and 2 above, T1-weighted refers to T1-weighted imaging; rs-fMRI (resting-state functional MRI) refers to resting-state functional magnetic resonance imaging; DTI (Diffusion Tensor Imaging) refers to diffusion tensor imaging; TR (repetition time) is the repetition time in milliseconds (ms); TE (echo time) is the echo time in milliseconds (ms); FOV (field of view) is the field of view; Voxel size is the voxel size in millimeters (mm); Matrix is ​​the matrix; Slices is the number of slices; Slice thickness is the slice thickness in millimeters (mm); Flip angle is the flip angle; Volumes are the number of volumes; Diffusion direction is the diffusion direction; and b value is the b-value.

[0122] S202. Construct a dual-stream structural data algorithm for extracting quantitative values ​​of brain structural morphology with dimensionality reduction and simplification.

[0123] To accommodate the differences in spatial resolution of the samples mentioned above, this embodiment constructs a dual-stream structured data algorithm processing flow. The dual-stream structured data includes a first extraction branch suitable for high-resolution scan images and a second extraction branch for lower-resolution and anisotropic clinical scans. The first extraction branch can be implemented using a standard recon-all process, for example, using FreeSurfer (version 8.0.0); the second extraction branch can be implemented using the "recon-all-clinical" processing flow developed by the Martinos Center for Biomedical Imaging at Harvard Medical School. This "recon-all-clinical" combines the signed distance function (SDF) predicted by a convolutional neural network (CNN) with geometric processing, thereby achieving geometrically accurate cortical reconstruction and reliable thickness estimation across heterogeneous resolutions and contrasts without retraining. This flow has been validated on large clinical datasets.

[0124] In this embodiment, FreeSurfer is used to implement the standard recon-all workflow as the first extraction branch. The Schaefer 400 partition map is used to extract cortical thickness and surface area, and the subcortical volume is estimated based on the existing NextBrain probabilistic histological atlas. For anisotropic clinical scans (slice thickness 5-6 mm), the recon-all-clinical workflow is used to reconstruct the cortex, obtain cortical thickness and surface area, and estimate the subcortical volume.

[0125] In a preferred embodiment, to reduce the multicenter effect across all datasets (including the validation cohort and iRBD), age and sex are used as biological covariates, and features are harmonized using ComBat. To strictly prevent information leakage within the machine learning framework, a hierarchical strategy is employed in the harmonization process: for the internal test set and iRBD cohorts at the same site, ComBat parameters are learned and directly applied using the discovery training set as a reference batch, without refitting. Crucially, to evaluate the system's "plug-and-play" generalization ability in real-world routine clinical settings, data from external clinical thick-slice MRI cohorts (2D 5-6mm thick T1 sequences) are not subject to any centralized ComBat harmonization; the raw features extracted based on recon-all-clinical analysis are directly input into the locked model for independent inference. ComBat is a common batch effect correction method in medical imaging and omics data. For the external validation cohort using independent scanning equipment, its specific site effects (location and scale parameters) are independently estimated using data from within the cohort, and then its distribution baseline is uniformly aligned to the discovery training set.

[0126] For multimodal benchmarking in the exploration cohort, both rs-fMRI and DTI data underwent standardized preprocessing. rs-fMRI preprocessing included slice timing correction, motion realignment, ICA-FIX denoising, and bandpass filtering, followed by the construction of a Functional Connectivity (FC) matrix using Schaefer 400 partition atlas. DTI data was processed in MRTrix3, including denoising, eddy current correction, and biasfield correction, and a Structural Connectivity (SC) matrix was generated using anatomically constrained tractography. Finally, in this embodiment, global and local graph metrics for the FC and SC networks were calculated using the Brain Connectivity Toolbox. Specifically, a weighted functional connectivity (FC) matrix was constructed using Pearson correlation coefficients (excluding negative values), while the structural connectivity (SC) matrix was defined by aggregate streamline weights connecting pairs of regions of interest. Topological metrics were quantified using the BrainConnectivity Toolbox. Global efficiency and modularity metrics were calculated to assess the degree of integration and separation. At the node level, degree centrality, clustering coefficient, local efficiency, and betweenness centrality were calculated. Following established specifications, the hemispherical SC metric was calculated independently to eliminate long-range tracking bias, while the FC metric was calculated based on the whole-brain network. This approach enabled the use of the Schaefer 400 brain map to define network nodes, ensuring spatial consistency across different modalities.

[0127] S203, Machine learning model building and development;

[0128] In this embodiment, a step-by-step modeling strategy is adopted to construct the model. The specific steps include: (1) constructing a multimodal benchmark model that integrates T1, DTI, and rs-fMRI features. This model is used to evaluate the upper limit of the effectiveness of the three-category discrimination of healthy controls, PD, and MSA; (2) constructing a first-classification model that only includes T1-weighted structural features to fix parameters and improve clinical applicability; (3) selecting 20 brain regions with the most discriminative value and constructing a simplified T1-Top20 feature model as the second-classification model. Figure 3 The specific implementation steps of the step-by-step modeling strategy shown are as follows:

[0129] The diagnostic modeling framework for all three models adopts the Elastic Network Regularized Generalized Linear Model algorithm, which can be implemented based on GLMNET technology. The details will be explained below:

[0130] (1) Construct a multimodal model to provide accurate anchoring references;

[0131] In the exploratory cohort, this study first centered and normalized the high-resolution T1-weighted structural features (covering cortical thickness, surface area, and subcortical volume), the structural connectivity features derived from diffusion tensor imaging (DTI), and the functional connectivity features derived from resting-state functional magnetic resonance imaging (rs-fMRI) to construct a multimodal high-dimensional feature matrix.

[0132] In terms of model architecture design, this framework introduces a hybrid L1 (Lasso) and L2 (Ridge) penalty term. The L1 penalty term forces the coefficients of redundant or irrelevant image features to shrink to zero, achieving embedded, strongly guided feature selection. The L2 penalty term effectively addresses the high collinearity problem among multimodal brain region features, ensuring the model's statistical robustness on small medical sample data. Model parameter tuning is rigorously performed within the exploration queue, employing a 10-fold hierarchical cross-validation strategy, and performing a grid search in the two-dimensional hyperparameter space for the regularization mixture parameters and regularization strength parameters.

[0133] Finally, the optimal parameter combination that maximizes the area under the receiver operating characteristic (AUROC) curve of the cross-validation validation set was selected to establish the multimodal baseline classifier. This model rigorously defines the theoretical upper limit of performance for the three-class classification of healthy controls (HC), Parkinson's disease (PD), and multiple system atrophy (MSA), providing a precise anchoring benchmark for subsequent single-modal dimensionality reduction models for routine clinical use.

[0134] (2) Construct a first classification model that only includes T1-weighted structural features;

[0135] In this embodiment, a first classification model is also constructed to reduce and optimize multimodal magnetic resonance imaging to single-modal brain morphological structural images. This first classification model simplifies the complex information from multimodal magnetic resonance imaging. The input of the first classification model is simplified to a single T1-weighted structural image feature. Subsequently, based on this T1-weighted structural feature, quantitative values ​​of the first brain structure morphology are extracted, specifically including cortical thickness, surface area, and subcortical volume. On this basis, a disease-specific core feature set is constructed.

[0136] In the feature selection phase, to more accurately capture the specific brain structural evolution patterns of different Parkinson's syndromes, this study did not directly rely on the overall feature output of the three-classification model, but instead adopted a "one-vs-rest" feature extraction strategy. Specifically, two independent GLMNET binary classification models were trained in the exploration cohort: a "PD vs (HC+MSA)" model and a "MSA vs (HC+PD)" model. This design aimed to allow the models to focus on the specific imaging boundaries that distinguish a single target disease from all other clinical states. To strictly avoid selection bias and data leakage that might be introduced during the feature selection process, this study employed a rigorous nested cross-validation framework in the internal exploration cohort. The inner loop was dedicated to high-weight selection of one-to-many features and GLMNET hyperparameter tuning, while the outer loop was used only for unbiased performance evaluation of unseen samples. Furthermore, permutation tests were performed under equal data distribution to statistically confirm that the extracted core feature set did not originate from data noise or overfitting. After model tuning, the coefficients matrices of the two independent models under optimal hyperparameters are obtained.

[0137] (3) Select 20 brain regions with the most discriminative value and construct a simplified feature model based on T1-Top20 as a second classification model;

[0138] After obtaining the coefficients matrix of the two independent models contained in the first classification model under the optimal hyperparameters, the core driving brain regions corresponding to different diagnostic categories are extracted in order to construct a second classification model that can predict the diagnostic category based on the core driving brain regions.

[0139] Specifically, after removing intercepts and zero-coefficient features, the feature coefficients of each brain region included in the first classification model are sorted in descending order according to the absolute value of the non-zero coefficients. Under the elastic network regularization framework, the magnitude of the absolute value of the coefficients objectively quantifies the independent contribution of the morphological changes of the brain region to the differential diagnosis of a specific disease. Based on this objective ranking, this embodiment extracts the top 10 PD core driving brain regions and the top 10 MSA core driving brain regions. After merging the two groups of brain regions and removing overlapping areas, a simplified T1-Top20 space containing no more than 20 features is finally constructed.

[0140] In subsequent multi-class prediction of diagnostic categories and spatial mapping of the prodromal period (iRBD) diagnosis, this system does not require the introduction of additional manual weight settings. Instead, it directly uses the original non-zero coefficients generated by data-driven methods from the aforementioned independent binary classification model as the feature weights for each brain region. This weighting mechanism based on original coefficient values ​​fully accommodates the heterogeneity of disease pathology: even if a specific brain region overlaps in the core feature lists of PD and MSA, the system will assign coefficient values ​​of completely different magnitudes and directions (positive / negative correlation) based on the actual statistical role of that brain region in the evolutionary trajectory of different disease pathologies. This method achieves extreme feature dimension compression while ensuring high accuracy and clinical interpretability in cross-disease identification with rigorous mathematical logic. Finally, only these 20 features are used to construct the final GLMNET classifier, and all model parameters are frozen before evaluation by external research-grade and external clinical MRI cohorts to obtain the second classification model.

[0141] S204, Model Performance Verification;

[0142] In this embodiment, all statistical analysis of the data was performed using R software version 4.4.1. After construction, the efficacy of each model was evaluated using the Area Under the Receiver Operating Characteristic (AUROC), sensitivity, specificity, positive predictive value, and negative predictive value. The 95% confidence interval for AUROC was calculated using the DeLong method, and the 95% confidence intervals for other efficacy indicators were calculated using the Clopper-Pearson exact method. The DeLong test was used to compare the AUROC differences between the multimodal baseline model and the T1-Top20 model in the internal test set. All statistical tests were two-tailed, with a significance level set at P < 0.05. In the iRBD cohort, the sensitivity of identifying patients diagnosed with phenotypic transformation using the PD-like evolution trajectory was assessed to evaluate its prognostic effectiveness. The effectiveness of the constructed models was evaluated in specific internal test sets, external test sets, and clinical validation sets.

[0143] like Figure 4 As shown in Part A, the multimodal benchmark model integrating T1, DTI, and rs-fMRI features achieved an area under the receiver operating characteristic (AUROC) of 0.98 (95% confidence interval: 0.939–1.000) in the discrimination of healthy controls (HC), PD, and MSA. Figure 4 As shown in Part B, the single-modal Full T1 Model achieved an AUROC of 0.96 (95% confidence interval: 0.914–1.000). And as... Figure 4As shown in Figure C, the AUROC of the simplified T1-Top20 model also reaches 0.95 (95% confidence interval: 0.894-1.000). This indicates that all three models exhibit considerable diagnostic efficacy on the internal test set. Figure 4 As shown in Figure D, for the distinction between PD and MSA categories, the T1-Top20 generalization model achieved an AUROC of 0.91 (95% confidence interval: 0.859-0.952) on the external test set (which can be constructed based on data from an external research-grade cohort). According to... Figure 4 As shown in Figure E, the T1-Top20 generalization model achieved an AUROC of 0.85 (95% confidence interval: 0.810-0.892) on the clinical validation set (which can be constructed based on an external clinical MRI cohort) for classifying PD and MSA. Based on the above, it can be seen that the simplified second classification model provided in this embodiment can achieve good diagnostic results on internal test sets, external data, and clinical validation test sets, and it can determine the early diagnostic category of neurodegenerative diseases based on low-resolution MRI images in clinical practice.

[0144] In addition, please continue to refer to Figure 5 In the actual internal test set, the T1-Top20 feature model exhibited a clear coefficient distribution pattern. Among the PD-related feature components, the features with the largest positive coefficients were mainly concentrated in the area of ​​the right frontal tectum / sula cortex, the left parietal cortex, and the right posterior occipital cortex, suggesting that an increase in the cortical area of ​​these brain regions would significantly improve the discrimination probability of PD classification. In the MSA-related feature components, the brain regions dominated by negative coefficients were mainly the infratentorial structures, including the left pons, left pontine nucleus, and left inferior cerebellar peduncle, suggesting that a decrease in the volume of these regions would increase the discrimination probability of MSA classification. Two other brain regions showed opposite coefficient directions in different components: the area of ​​the right posterior occipital cortex was positive for PD classification and negative for MSA classification; while the thickness of the left visual cortex was negative for PD classification and positive for MSA classification.

[0145] After the model constructed above has been validated, to evaluate the application value of the T1-Top20 feature model in the prodromal stage of PD, please refer to [reference needed]. Figure 6This study applied the model to the aforementioned exploratory iRBD cohort (n=107) to verify its predictive accuracy. In 25 iRBD phenotype converts diagnosed with Parkinson's disease (PD), this example further evaluated the prognostic value of the model. Results showed that 19 (76%) of the confirmed PD converts were concentrated in the "typical PD quadrant," with a relatively short average phenotype conversion time (1.8 years); 5 (20%) converts were located in the "uncertain quadrant," with a longer average phenotype conversion time (4.4 years); and only 1 (4%) convert was mapped to the "atypical / complex quadrant," achieving a relatively accurate predictive effect. Figure 6 A scatter plot of 107 iRBD patients on the PD-MSA neuroanatomical coordinate system is shown. Dashed lines represent diagnostic thresholds determined based on the discovery stage. Confirmed Parkinson's disease conversion cases (circles) are primarily clustered in typical Parkinson's disease regions, clearly distinct from the indeterminate (diamond) and atypical / complex (triangle) subtypes. Parkinson's disease phenotypic scores and atypical / MSA-like phenotypic scores are derived by calculating the weighted sum of the top 10 regions driving the Parkinson's disease classification and the top 10 driving the MSA classification from the T1-Top20 features, where y represents the number of years required for phenotypic conversion.

[0146] S205, Determination of diagnostic category and prediction of iRBD differentiation type;

[0147] In this embodiment, the second quantitative value of brain structure morphology is the feature vector of two Top 10 core driving brain regions. The second classification model includes a simplified Parkinson's disease classifier and a simplified multiple system atrophy (MSA) classifier. The simplified Parkinson's disease classifier calculates a probability score indicating the subject has Parkinson's disease and a linear score indicating the subject is healthy, based on the 10 frozen weights and the feature vectors of the Top 10 core driving brain regions for Parkinson's disease. The simplified MSA classifier calculates a probability score indicating the subject has multiple system atrophy and a linear score indicating the subject is healthy, based on the 10 frozen weights and the feature vectors of the Top 10 core driving brain regions for multiple system atrophy. After obtaining the two linear scores, the three scores are converted into probability scores for different diagnostic categories using the softmax function in the second classification model, and the diagnostic category corresponding to the highest probability score is used as the auxiliary diagnostic result.

[0148] In a preferred embodiment, to overcome the clinical challenge of objectively quantifying and predicting the trajectory of transformation from the prodromal stage of neurodegenerative diseases (such as examinees with iRBD) to a clear phenotype, this embodiment proposes a "two-dimensional feature space mapping and risk stratification" mechanism based on mature disease benchmarks.

[0149] In constructing the radiographic evolution trajectory of the precursor expectant, this study did not introduce additional complex neural networks or nonlinear mapping algorithms, but directly called the simplified T1-Top20 feature model (i.e., the second classification model described in this embodiment) that had already completed parameter freezing.

[0150] The specific mapping algorithm is as follows: The system automatically extracts quantitative values ​​of brain structural morphology from the baseline MRI of the prodromal subjects, corresponding to the core brain regions, and performs linear weighted summation with the "10 PD driving characteristic coefficients" and "10 MSA driving characteristic coefficients" frozen in the exploration cohort. This calculation process directly outputs two independent quantitative indicators—the "PD characteristic score," reflecting the severity of structural atrophy in typical Parkinson's disease, and the "MSA characteristic score," reflecting the structural damage load in atypical or multi-system atrophy. These two independent scores constitute the horizontal and vertical axes of the reference coordinate system, respectively, thereby accurately and dimensionally reducing the projection of the subject's multi-scale brain structural state onto this two-dimensional diagnostic space.

[0151] Regarding the critical decision thresholds determining subtype differentiation and evolutionary trajectories, the algorithm design rigorously avoids the risks of circular argument bias and data leakage caused by data fitting within the exploratory prodromal cohort. The supervised stratified cut-off values ​​used by the system are calculated entirely independently of the iRBD cohort; they are calculated based on the receiver operating characteristic (ROC) curves of the internal exploratory cohort (based on mature PD and MSA patients with deterministic clinical gold standards), and the optimal classification threshold corresponding to the Youden Index (the maximum point of the sum of sensitivity and specificity) is taken. This anchoring threshold is absolutely frozen before being applied to any external prodromal data.

[0152] Based on the aforementioned fixed two-dimensional coordinate system and freezing threshold, the system performs objective quadrant division and trajectory recognition on the prognostic candidates, differentiating their prognosis into three subtypes:

[0153] (1) Typical PD type (PD score is higher than the preset threshold and MSA score is lower than the preset threshold). The system identifies that the brain structure of this type of examinee has shown a classic PD-specific atrophy pattern and is on a high-risk track to evolve into typical PD.

[0154] (2) Atypical or complex PD type (MSA score higher than its specific threshold, regardless of PD score), the system indicates that the subject's brain structure has deviated from the classic PD unidirectional evolutionary trajectory, presenting a more extensive or MSA-like invasive neurodegenerative burden;

[0155] (3) Uncertain type (both axis scores are below the threshold), indicating that the current macroscopic structure has not undergone specific irreversible changes. To verify the effectiveness of the system in identifying differentiation trajectories, this study introduced 25 iRBD patients with confirmed phenotype transformation to PD through long-term longitudinal follow-up as independent clinical reference criteria. By projecting this confirmed transformation subgroup into the above-defined two-dimensional coordinate and threshold system, the model's prospective identification sensitivity for "typical PD evolution trajectory" was evaluated.

[0156] In this step, when targeting specific clinical applications, the frozen training model parameters are used. This eliminates the need for repeated training and validation of the model in actual applications, greatly saving computing power. Furthermore, the T1 structure sequence used is suitable for use in hospitals at all levels across the country, achieving true "plug-and-play".

[0157] S206, Data Visualization and Report Generation;

[0158] In this embodiment, after determining the diagnostic category and predicting the iRBD differentiation type, a visual report can be generated based on the score values ​​of different core driving brain regions obtained during model construction, outputting information such as the physiological principles and lesion areas of the disease diagnosis results.

[0159] Based on the above, this invention constructs and verifies a machine learning model based on T1-weighted structural MRI, which can be used for the differential diagnosis of early and mid-stage Parkinson's syndrome. This model is generalized to anisotropic two-dimensional (2D) thick-slice (5-6 mm) clinical MRI, enabling early prediction of differentiation direction, improving the generalization ability of the testing process, saving costs, and having good scalability. It can also simultaneously output information such as the physiological principles and lesion areas of the disease diagnosis results, improving interpretability and facilitating doctors and patients to view the classification reasons of the assessment results. This ensures that the model's output results are reasonably and directly related to the pathophysiological principles of medicine, avoiding the "black box" of the model's principles.

[0160] This embodiment shifts from multimodal analysis to unimodal analysis and from whole-brain morphological measurement to features of 20 core brain regions. While maintaining discriminative performance, it streamlines and optimizes the neuroimaging-based diagnostic model. The provided scheme is based on a T1-Top20 model with locked multi-category image baseline thresholds, enabling accurate diagnostic category differentiation using routine clinical MRI images. It can also identify phenotypic transitions in the prodromal stage of disease; that is, the method provided in this embodiment can not only diagnose the diagnostic category of patients already suffering from the disease but also perform risk warning classification for patients not yet suffering from the disease. Under real clinical conditions, the application of the diagnostic model can be advanced to achieve early diagnosis and risk prediction of neurodegenerative diseases, demonstrating its potential value in the early diagnosis and differential diagnosis of Parkinson's syndrome. Given that structural MRI of the brain is a routine examination in the initial diagnostic assessment of Parkinson's syndrome patients, the generalized model constructed in this embodiment is expected to significantly promote the early diagnosis and differential diagnosis of neurodegenerative diseases.

[0161] The foregoing description of the method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging in embodiments of the present invention has been presented. The following description of the system for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging in embodiments of the present invention is provided. Figure 7 One embodiment of the early auxiliary diagnostic system and prediction for neurodegenerative diseases using clinical magnetic resonance imaging in this invention includes:

[0162] The data acquisition module 701 is used to acquire multimodal magnetic resonance imaging as training samples and to simplify the extraction of the first brain structure morphological quantitative values ​​of each training sample by dimensionality reduction; wherein, the training samples are labeled with diagnostic category tags for different neurodegenerative diseases.

[0163] The region extraction module 702 is used to train a first classification model based on the first brain structure morphological quantitative values ​​and the diagnostic category labels of each training sample, and to extract the feature coefficient matrix corresponding to the trained first classification model; based on the feature weight coefficients of different diagnostic categories contained in the feature coefficient matrix, the core driving brain regions corresponding to different diagnostic categories are selected.

[0164] Model building module 703 is used to build a second classification model based on the core driving brain regions and the corresponding feature weight coefficients;

[0165] The auxiliary diagnostic module 704 is used to acquire clinical magnetic resonance images of the examinee in the early or prodromal stage, extract the second brain structure morphological quantitative value of the clinical magnetic resonance image, call the second classification model to calculate the probability score of different diagnostic categories based on the second brain structure morphological quantitative value, and output the auxiliary diagnosis or progression prediction result based on the probability score.

[0166] The system provided in this invention can use routine clinical magnetic resonance imaging to achieve accurate and automated identification of neurodegenerative diseases in the early stages, thereby improving the generalization ability of diagnostic aid technology.

[0167] In another embodiment of this application, the diagnostic categories include Parkinson's disease and multiple system atrophy; the second classification model consists of a simplified Parkinson's disease classifier and a simplified multiple system atrophy classifier;

[0168] The auxiliary diagnostic module 704 is specifically used for:

[0169] Based on the second quantitative value of brain structure morphology, the probability scores of Parkinson's disease and multiple system atrophy are calculated by calling the simplified Parkinson's disease classifier and the simplified multiple system atrophy classifier, respectively.

[0170] The diagnostic category corresponding to the highest probability score is output as the auxiliary diagnostic result.

[0171] In another embodiment of this application, the first classification model consists of a full-feature Parkinson's disease classifier and a full-feature multisystem atrophy classifier;

[0172] The region extraction module 702 is specifically used for:

[0173] Extract the Parkinson's disease feature coefficient matrix of the full-feature Parkinson's disease classifier under optimal hyperparameters and the multi-system atrophy feature coefficient matrix of the full-feature multi-system atrophy classifier under optimal hyperparameters respectively;

[0174] After removing the intercept terms and zero coefficients from the Parkinson's disease characteristic coefficient matrix and the multisystem atrophy characteristic coefficient matrix, multiple non-zero characteristic coefficients of Parkinson's disease and multiple non-zero characteristic coefficients of multisystem atrophy are obtained.

[0175] The absolute values ​​of the non-zero characteristic coefficients of Parkinson's disease and the non-zero characteristic coefficients of multiple system atrophy are sorted in descending order, and the corresponding brain regions within the preset ranking range are extracted to obtain the core driving brain regions of Parkinson's disease and multiple system atrophy.

[0176] In another embodiment of this application, the data acquisition module 701 is specifically used for:

[0177] Based on the spatial resolution characteristics of each training sample, the first extraction branch and the second extraction branch in the dual-stream structured data algorithm are used to reconstruct the cortex of training samples with different resolutions, and the quantitative values ​​of the first brain structure morphology of each training sample are obtained by measurement and calculation.

[0178] The auxiliary diagnostic module 704 is further used for:

[0179] Based on the second processing branch of the dual-stream structured data algorithm, cortical reconstruction, topological correction, and thickness estimation are performed on the clinical magnetic resonance images to extract quantitative morphological values ​​of the second brain structure of the core driving brain regions corresponding to different diagnostic categories in the clinical magnetic resonance images.

[0180] In another embodiment of this application, both the data acquisition module 701 and the auxiliary diagnosis module 704 call a morphology extraction unit. The morphology extraction unit extracts quantitative values ​​of brain structure morphology from magnetic resonance images of different resolutions through the first extraction branch and the second extraction branch in the dual-stream structured data algorithm.

[0181] In another embodiment of this application, the auxiliary diagnostic module 704 is also used to predict the differentiation risk of examinees in the prodromal stage of neurodegenerative diseases.

[0182] The auxiliary diagnostic module 704 is specifically used for:

[0183] The second classification model is invoked to calculate the total feature score corresponding to different diagnostic categories based on the quantitative value of the second brain structure morphology, and the differentiation risk prediction result is generated.

[0184] Among them, the examinee in the prodromal stage of the neurodegenerative disease is the examinee who has symptoms of idiopathic rapid eye movement sleep behavior disorder.

[0185] In another embodiment of this application, the total feature scores corresponding to the different diagnostic categories include the Parkinson's disease feature score and the multiple system atrophy feature score;

[0186] The auxiliary diagnostic module 704 is specifically used to: call the second classification model to calculate the total score of Parkinson's disease characteristics and the total score of multiple system atrophy characteristics based on the second quantitative value of brain structure morphology;

[0187] Obtain a preset differentiation prediction threshold, and divide the differentiation trajectory based on the total score of Parkinson's disease features and the total score of multiple system atrophy features;

[0188] If the total score of Parkinson's disease features is higher than the differentiation prediction threshold and the total score of multiple system atrophy features is lower than the differentiation prediction threshold, it indicates that there is a typical risk of Parkinson's disease.

[0189] If the total score of the multi-system atrophy feature is higher than the differentiation prediction threshold, it indicates that there is a risk of multi-system atrophy.

[0190] If both the total score for Parkinson's disease characteristics and the total score for multiple system atrophy characteristics are lower than the differentiation prediction threshold, it indicates that the current macrostructure has not undergone specific irreversible changes, and the risk of differentiation is uncertain.

[0191] In another embodiment of this application, the step of calling the second classification model to calculate the total score of Parkinson's disease features and the total score of multiple system atrophy features based on the second quantitative value of brain structure morphology includes:

[0192] Obtain the weight coefficients of the core driving brain regions related to Parkinson's disease frozen in the second classification model, and obtain the total score of Parkinson's disease characteristics by weighted summation based on the second brain structure morphological quantitative value and the weight coefficients.

[0193] Obtain the weight coefficients of the core driving brain regions frozen and associated with multiple system atrophy in the second classification model, call the second classification model to perform a weighted summation based on the second brain structure morphological quantitative value and the weight coefficients, and obtain the total score of multiple system atrophy features.

[0194] In another embodiment of this application, the early auxiliary diagnosis and prediction system for neurodegenerative diseases using clinical magnetic resonance imaging further includes an effect evaluation module, which is specifically used to: acquire the structural connectivity features derived from diffusion tensor imaging and the functional connectivity features derived from resting-state functional magnetic resonance imaging of the training samples, and construct a multimodal high-dimensional feature matrix by combining the high-resolution T1-weighted structural features of the training samples.

[0195] A multimodal benchmark classifier is established by combining a mixed penalty term and a 10-fold hierarchical cross-validation strategy, and the classification performance of the multimodal benchmark classifier is evaluated based on a pre-set test set.

[0196] The effect evaluation module is also used to evaluate the classification effect of the second classification model using a preset test set. When the classification effect of the second classification model is within the feasibility evaluation range, the model building module 703 is called to freeze the parameters of the second classification model.

[0197] The system and method provided in this embodiment of the invention can use routine clinical magnetic resonance images to achieve accurate and automated identification of neurodegenerative diseases in the early stages of neurodegenerative diseases, as well as to achieve early prediction of differentiation direction. This improves the generalization ability of the testing process, saves costs, and has good scalability.

[0198] Based on the same inventive concept, this specification also provides an electronic device for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging. The electronic device for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging in this embodiment of the invention will be described in detail below from the perspective of hardware processing.

[0199] Figure 8 This is a schematic diagram of an electronic device provided as an embodiment of this specification. Refer to the following... Figure 8 To describe the electronic device 800 according to this embodiment of the invention. Figure 8 The electronic device 800 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0200] like Figure 8 As shown, the electronic device 800 is presented in the form of a general-purpose computing device. The components of the electronic device 800 may include, but are not limited to: at least one processing unit 810, at least one storage unit 820, a bus 830 connecting different system components (including storage unit 820 and processing unit 810), a display unit 840, etc.

[0201] The storage unit stores program code that can be executed by the processing unit 810, causing the processing unit 810 to perform the steps described in the processing method section of this specification according to various exemplary embodiments of the present invention. For example, the processing unit 810 can perform, as follows: Figure 1 or Figure 2 The steps are shown.

[0202] The storage unit 820 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 8201 and / or a cache storage unit 8202, and may further include a read-only memory unit (ROM) 8203.

[0203] The storage unit 820 may also include a program / utility 8204 having a set (at least one) program module 8205, such program module 8205 including but not limited to: an operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0204] Bus 830 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0205] Electronic device 800 can also communicate with one or more external devices 100 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with the electronic device 800, and / or with any device that enables the electronic device 800 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 850. Furthermore, electronic device 800 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 860. Network adapter 860 can communicate with other modules of electronic device 800 via bus 830. It should be understood that, although... Figure 8 As not shown in the diagram, other hardware and / or software modules may be used in conjunction with the electronic device 800, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0206] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described in this invention can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this invention can be embodied in the form of a software product, which can be stored in a computer-readable storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, or network device, etc.) to execute the method described above according to this invention. When the computer program is executed by a data processing device, it enables the computer-readable medium to implement the method described above, i.e.: as... Figure 1 or Figure 2 The method shown.

[0207] Figure 9 This is a schematic diagram of a computer-readable medium provided for embodiments of this specification.

[0208] accomplish Figure 1 or Figure 2The computer program of the method shown can be stored on one or more computer-readable media. A computer-readable medium can be a readable signal medium or a readable storage medium. A readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0209] The computer-readable storage medium may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The readable storage medium may also be any readable medium other than a readable storage medium, capable of transmitting, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0210] In addition, the present invention also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the method for early auxiliary diagnosis and prediction of neurodegenerative diseases for clinical magnetic resonance imaging as described in any of the above embodiments.

[0211] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0212] In summary, the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that in practice, general-purpose data processing devices such as microprocessors or digital signal processors (DSPs) can be used to implement some or all of the functions of some or all of the components according to the embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0213] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the present invention is not inherently related to any specific computer, virtual device, or electronic device, and various general-purpose devices can also implement the present invention. The above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0214] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0215] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging, characterized in that, include: Multimodal magnetic resonance imaging (MRI) images were collected as training samples, and the quantitative values ​​of the first brain structure morphology of each training sample were extracted by dimensionality reduction and simplification; wherein, the training samples were labeled with diagnostic category tags for different neurodegenerative diseases; A first classification model is trained based on the quantitative values ​​of the first brain structure morphology of each training sample and the diagnostic category label, and the feature coefficient matrix corresponding to the trained first classification model is extracted. Based on the feature weight coefficients of different diagnostic categories contained in the feature coefficient matrix, the core driving brain regions corresponding to different diagnostic categories are selected. A second classification model is constructed based on the core driving brain regions and their corresponding feature weight coefficients. Acquire clinical magnetic resonance images of subjects in the early or prodromal stage, and extract quantitative values ​​of the second brain structure morphology from the clinical magnetic resonance images; The second classification model is invoked to calculate the probability score of different diagnostic categories based on the quantitative value of the second brain structure morphology, and the auxiliary diagnosis or progression prediction result is output based on the probability score.

2. The method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging according to claim 1, characterized in that, The diagnostic categories include Parkinson's disease and multiple system atrophy; the second classification model consists of a simplified Parkinson's disease classifier and a simplified multiple system atrophy classifier; The step of calling the second classification model to calculate probability scores for different diagnostic categories based on the quantitative values ​​of the second brain structure morphology, and outputting auxiliary diagnostic and progression prediction results based on the probability scores includes: Based on the second quantitative value of brain structure morphology, the probability scores of Parkinson's disease and multiple system atrophy are calculated by calling the simplified Parkinson's disease classifier and the simplified multiple system atrophy classifier, respectively. The diagnostic category corresponding to the highest probability score is output as the auxiliary diagnostic result.

3. The method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging according to claim 2, characterized in that, The first classification model consists of a full-feature Parkinson's disease classifier and a full-feature multisystem atrophy classifier; The process of selecting the core driving brain regions corresponding to different diagnostic categories based on the feature weight coefficients of different diagnostic categories contained in the feature coefficient matrix includes: Extract the Parkinson's disease feature coefficient matrix of the full-feature Parkinson's disease classifier under optimal hyperparameters and the multi-system atrophy feature coefficient matrix of the full-feature multi-system atrophy classifier under optimal hyperparameters, respectively. After removing the intercept terms and zero coefficients from the Parkinson's disease characteristic coefficient matrix and the multisystem atrophy characteristic coefficient matrix, multiple non-zero characteristic coefficients of Parkinson's disease and multiple non-zero characteristic coefficients of multisystem atrophy are obtained. The absolute values ​​of the non-zero characteristic coefficients of Parkinson's disease and the non-zero characteristic coefficients of multiple system atrophy are sorted in descending order, and the corresponding brain regions within the preset ranking range are extracted to obtain the core driving brain regions of Parkinson's disease and multiple system atrophy.

4. The method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging according to claim 2, characterized in that, The dimensionality reduction and simplified extraction of the first brain structure morphological quantitative values ​​of each training sample includes: Based on the spatial resolution characteristics of each training sample, the first extraction branch and the second extraction branch in the dual-stream structured data algorithm are used to reconstruct the cortex of training samples with different resolutions. The quantitative values ​​of the first brain structure morphology of each training sample are obtained through dimensionality reduction simplification, measurement and calculation. The second quantitative brain structural morphology values ​​extracted from the clinical magnetic resonance images include: Based on the second processing branch of the dual-stream structured data algorithm, cortical reconstruction, topological correction, and thickness estimation are performed on the clinical magnetic resonance images to extract quantitative values ​​of the second brain structure morphology of the core driving brain regions corresponding to different diagnostic categories in the clinical magnetic resonance images.

5. The method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging according to any one of claims 1-4, characterized in that, The early auxiliary diagnosis also includes predicting the differentiation risk of candidates in the prodromal stage of neurodegenerative diseases; When the subject is in the prodromal stage of a neurodegenerative disease, after acquiring the subject's clinical magnetic resonance imaging (MRI) images and extracting the quantitative values ​​of the second brain structure morphology from the MRI images, the procedure further includes: The second classification model is invoked to calculate the total feature score corresponding to different diagnostic categories based on the quantitative value of the second brain structure morphology, and the differentiation risk prediction result is generated. Among them, the examinee in the prodromal stage of the neurodegenerative disease is the examinee who has symptoms of idiopathic rapid eye movement sleep behavior disorder.

6. The method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging according to claim 5, characterized in that, The total feature scores corresponding to the different diagnostic categories include the Parkinson's disease feature score and the multiple system atrophy feature score; The step of calling the second classification model to calculate the feature total score corresponding to different diagnostic categories based on the second brain structure morphological quantitative value, and generating differentiation risk prediction results includes: Obtain the weight coefficients of the core driving brain regions related to Parkinson's disease frozen in the second classification model, and obtain the total score of Parkinson's disease characteristics by weighted summation based on the second brain structure morphological quantitative value and the weight coefficients. Obtain the weight coefficients of the core driving brain regions frozen and associated with multiple system atrophy in the second classification model, call the second classification model to perform a weighted summation based on the second brain structure morphological quantitative value and the weight coefficients, and obtain the total score of multiple system atrophy features; Obtain a preset differentiation prediction threshold, and divide the differentiation trajectory based on the total score of Parkinson's disease features and the total score of multiple system atrophy features; If the total score of Parkinson's disease features is higher than the differentiation prediction threshold and the total score of multiple system atrophy features is lower than the differentiation prediction threshold, it indicates that there is a typical risk of Parkinson's disease. If the total score of the multi-system atrophy feature is higher than the differentiation prediction threshold, it indicates that there is a risk of multi-system atrophy. If both the total score for Parkinson's disease characteristics and the total score for multiple system atrophy characteristics are lower than the differentiation prediction threshold, it indicates that the current macrostructure has not undergone specific irreversible changes, and the risk of differentiation is uncertain.

7. A system for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging, characterized in that, include: The data acquisition module is used to acquire multimodal magnetic resonance images as training samples and to simplify the extraction of quantitative values ​​of the first brain structure morphology of each training sample by dimensionality reduction; wherein, the training samples are labeled with diagnostic category tags for different neurodegenerative diseases; The region extraction module is used to train a first classification model based on the quantitative values ​​of the first brain structure morphology of each training sample and the diagnostic category label, and to extract the feature coefficient matrix corresponding to the trained first classification model; based on the feature weight coefficients of different diagnostic categories contained in the feature coefficient matrix, the core driving brain regions corresponding to different diagnostic categories are selected. The model building module is used to build a second classification model based on the core driving brain regions and their corresponding feature weight coefficients; The auxiliary diagnostic module is used to acquire clinical magnetic resonance images of the examinee in the early or prodromal stage, extract the second brain structure morphological quantitative value of the clinical magnetic resonance image, call the second classification model to calculate the probability score of different diagnostic categories based on the second brain structure morphological quantitative value, and output the auxiliary diagnosis or progression prediction result based on the probability score.

8. A device for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging, characterized in that, The device for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor invokes the instructions in the memory to cause the device for early auxiliary diagnosis and prediction of neurodegenerative diseases for clinical magnetic resonance imaging to perform the steps of the method for early auxiliary diagnosis and prediction of neurodegenerative diseases for clinical magnetic resonance imaging as described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program / instructions thereon, characterized in that, When the program / instructions are executed by the processor, they implement the steps of the method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging as described in any one of claims 1-6.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method for early auxiliary diagnosis and prediction of neurodegenerative diseases using clinical magnetic resonance imaging as described in any one of claims 1-6.

Images