Construction Method and Device of Auxiliary Diagnosis Model for Vascular Cognitive Impairment

Abstract

The present invention relates to the technical field of brain MRI medical image processing. Specifically, a method for constructing an auxiliary diagnosis model for vascular cognitive impairment includes: acquiring multi-sequence magnetic resonance images of the brain and clinical data; preprocessing the multi-sequence magnetic resonance images of the brain and clinical data to obtain a heterogeneous multi-modal data set; the heterogeneous multi-modal data set includes multi-sequence magnetic resonance images and clinical data of the brains of a number of VCI patients, as well as multi-sequence magnetic resonance images and clinical data of the brains of a number of normal humans; constructing an auxiliary diagnosis model for vascular cognitive impairment based on the heterogeneous multi-modal data set. By using the multi-sequence magnetic resonance images and clinical data of VCI patients, as well as the multi-sequence magnetic resonance images and clinical data of the brains of normal humans, an auxiliary diagnosis model for vascular cognitive impairment is established, improving the diagnostic accuracy of vascular cognitive impairment. The present invention also provides a device for constructing an auxiliary diagnosis model for vascular cognitive impairment.

Description

Technical Field

[0001] The present invention relates to the technical field of brain MRI medical image processing, and particularly relates to a method and device for constructing an auxiliary diagnosis model for vascular cognitive impairment. Background Art

[0002] Vascular cognitive impairment (VCI) is a clinical syndrome of cerebrovascular lesions and their risk factors leading to clinical stroke or subclinical vascular brain injury, involving impairment of at least one cognitive domain, covering a spectrum from mild cognitive impairment to dementia, and also including various degrees of cognitive impairment caused by mixed pathologies such as coexisting Alzheimer's disease (AD). Vascular cognitive impairment is the second most common type of dementia after Alzheimer's disease, and its stages include vascular cognitive impairment without dementia (VCIND) and vascular dementia (VaD). Currently, artificial intelligence technology has been widely applied to various complex classifications in medical imaging and is increasingly used in the classification research of neurodegenerative diseases, and there are also various disease diagnosis models for various diseases.

[0003] However, there is no dedicated disease diagnosis model for vascular cognitive impairment in the prior art. The causes of vascular cognitive impairment are complex. Professor Li Shunwei from the Department of Neurology of Peking Union Medical College recorded in "Diagnosis and Treatment of Vascular Cognitive Impairment" that the diagnosis of vascular cognitive impairment must meet three conditions. First, there must be cerebrovascular lesions, regardless of their nature; second, there must be cognitive impairment, which may already show manifestations of dementia; third, there must be a causal relationship between the two. It can be seen that the diagnosis of vascular cognitive impairment is different from that of other diseases and requires a combination of clinical features, neurobehavior, neuroimaging, etc. to diagnose VCI. Moreover, vascular cognitive impairment is one of the diseases with the highest prevalence among various dementias. A survey by the Canadian Study of Health and Aging showed that the prevalence of VCI in the elderly over 65 years old is about 5%, and the incidence of VCI increases with age, with the highest incidence in the elderly aged 65 to 84 years. Therefore, there is an urgent need for a dedicated disease diagnosis model for vascular cognitive impairment to help improve the diagnostic accuracy of vascular cognitive impairment so as to detect it earlier and treat it as soon as possible. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a method and device for constructing an auxiliary diagnosis model for vascular cognitive impairment, which improves the diagnostic accuracy of vascular cognitive impairment.

[0005] In a first aspect, the present invention provides a method for constructing an auxiliary diagnosis model for vascular cognitive impairment.

[0006] In a first implementable manner, the method for constructing an auxiliary diagnosis model for vascular cognitive impairment includes: obtaining multi-sequence magnetic resonance images of the brain and clinical data; preprocessing the multi-sequence magnetic resonance images of the brain and clinical data to obtain a heterogeneous multi-modal data set; the heterogeneous multi-modal data set includes multi-sequence magnetic resonance images and clinical data of several VCI patients, as well as multi-sequence magnetic resonance images and clinical data of several normal humans; constructing an auxiliary diagnosis model for vascular cognitive impairment according to the heterogeneous multi-modal data set.

[0007] Combined with the first implementable manner, in a second implementable manner, obtaining multi-sequence magnetic resonance images of the brain and clinical data includes: obtaining functional magnetic resonance images, gray matter images, white matter images of the brain and clinical data of several VCI patients, and obtaining functional magnetic resonance images, gray matter images, white matter images of the brain and clinical data of several normal humans; determining the functional magnetic resonance images, gray matter images, and white matter images as multi-sequence magnetic resonance images of the brain; labeling the multi-sequence magnetic resonance images of the brain and clinical data, and the label is used to characterize VCI patients or normals.

[0008] Combined with the second implementable manner, in a third implementable manner, preprocessing the multi-sequence magnetic resonance images of the brain and clinical data to obtain a heterogeneous multi-modal data set includes: obtaining multi-modal registration data according to the multi-sequence magnetic resonance images of the brain; performing normalization processing on the multi-modal registration data and clinical data to obtain a heterogeneous multi-modal data set.

[0009] Combined with the third implementable manner, in a fourth implementable manner, obtaining multi-modal registration data according to the multi-sequence magnetic resonance images of the brain includes: resampling the multi-sequence magnetic resonance images of the brain to obtain multi-sequence magnetic resonance images of the brain with a unified three-dimensional shape; performing image registration on the multi-sequence magnetic resonance images of the brain with a unified three-dimensional shape to obtain multi-modal registration data.

[0010] Combined with the fourth implementable manner, in a fifth implementable manner, resampling the multi-sequence magnetic resonance images of the brain to obtain multi-sequence magnetic resonance images of the brain with a unified three-dimensional shape includes: resampling all the multi-sequence magnetic resonance images of the brain with a preset three-dimensional shape to obtain resampled data consistent with the preset three-dimensional shape; using a cubic spline interpolation algorithm for resampling calculation to obtain multi-sequence magnetic resonance images of the brain with a unified three-dimensional shape.

[0011] Combined with the fourth implementation method, in the sixth implementation method, image registration is performed on the multi-sequence magnetic resonance images of the brain with a unified three-dimensional shape to obtain multi-modal registration data, including: using the open-source ANTS registration library, taking the gray matter image as the fixed-domain image of the benchmark, and the white matter image and functional magnetic resonance image of the brain as the moving-domain images, and aligning the moving-domain images anatomically to the fixed-domain image.

[0012] Combined with the first implementation method, in the seventh implementation method, the heterogeneous multi-modal dataset includes training samples and test samples. An auxiliary diagnosis model for vascular cognitive impairment is constructed based on the heterogeneous multi-modal dataset, including: building an initial diagnosis model; training the initial diagnosis model according to the training samples; obtaining the accuracy rate of the trained initial diagnosis model according to the test samples; and stopping training and obtaining the auxiliary diagnosis model for vascular cognitive impairment when the accuracy rate is greater than the preset threshold.

[0013] Combined with the seventh implementation method, in the eighth implementation method, building the initial diagnosis model includes: using HRNET as the network backbone of the deep learning neural network and adopting a three-time sequence fusion method of shallow layer - middle layer - tail layer to build the initial diagnosis model.

[0014] As can be seen from the above technical solutions, the beneficial technical effects of the present invention are as follows: The optimized HRNET is used to extract the depth information of the images. HRNET adopts a parallel scale network of high resolution and low resolution, which can maintain the input-to-output of high-resolution features and is more conducive to the preservation of important fine-grained features. A multi-fusion method of three-time sequence fusion of shallow layer - middle layer - tail layer is adopted. On the basis of better retaining the features of the shallow layer, the problem of overfitting easily caused by the multi-sequence model is avoided.

[0015] In the second aspect, the present invention provides a device for constructing an auxiliary diagnosis model for vascular cognitive impairment.

[0016] In the ninth implementation method, the device for constructing an auxiliary diagnosis model for vascular cognitive impairment includes: a data acquisition module configured to acquire multi-sequence magnetic resonance images of the brain and clinical data; a heterogeneous multi-modal dataset acquisition module configured to preprocess the multi-sequence magnetic resonance images of the brain and clinical data to obtain a heterogeneous multi-modal dataset; the heterogeneous multi-modal dataset includes multi-sequence magnetic resonance images and clinical data of several VCI patients, as well as multi-sequence magnetic resonance images and clinical data of several normal humans; and a model construction module configured to construct an auxiliary diagnosis model for vascular cognitive impairment based on the heterogeneous multi-modal dataset.

[0017] As can be seen from the above technical solutions, the beneficial technical effects of the present invention are as follows: By preprocessing multi-sequence magnetic resonance images of the brain and clinical data, a heterogeneous multi-modal data set is obtained, so that the auxiliary diagnosis model for vascular cognitive impairment can comprehensively learn data from different sources and of different types, increasing the reliability of the auxiliary diagnosis model for vascular cognitive impairment. Using the multi-sequence magnetic resonance images of the brain and clinical data of the VCI patient group, as well as the multi-sequence magnetic resonance images of the brain and clinical data of the normal human group, an auxiliary diagnosis model for vascular cognitive impairment can be established, improving the diagnostic accuracy of vascular cognitive impairment, helping to detect vascular cognitive impairment earlier, and treating it as soon as possible. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0019] Figure 1 It is a schematic diagram of a method for constructing an auxiliary diagnosis model for vascular cognitive impairment provided by the present invention;

[0020] Figure 2 It is a schematic structural diagram of a device for constructing an auxiliary diagnosis model for vascular cognitive impairment provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] The following will describe in detail the embodiments of the technical solutions of the present invention with reference to the drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, so they are only examples and cannot be used to limit the protection scope of the present invention.

[0022] It should be noted that unless otherwise specified, the technical terms or scientific terms used in this application should have the ordinary meaning understood by those skilled in the art to which the present invention belongs.

[0023] Combined with Figure 1 As shown, this embodiment provides a method for constructing an auxiliary diagnosis model for vascular cognitive impairment, including:

[0024] Step S01, obtaining multi-sequence magnetic resonance images of the brain and clinical data;

[0025] Step S02, preprocessing the multi-sequence magnetic resonance images of the brain and clinical data to obtain a heterogeneous multi-modal data set; the heterogeneous multi-modal data set includes multi-sequence magnetic resonance images of the brain and clinical data of several VCI patients, as well as multi-sequence magnetic resonance images of the brain and clinical data of several normal humans;

[0026] Step S03: Construct an auxiliary diagnostic model for vascular cognitive impairment based on the heterogeneous multimodal dataset.

[0027] By preprocessing the multi-sequence magnetic resonance images of the brain and clinical data, a heterogeneous multimodal dataset is obtained, enabling the auxiliary diagnostic model for vascular cognitive impairment to comprehensively learn data from different sources and of heterogeneous types, thereby increasing the reliability of the auxiliary diagnostic model for vascular cognitive impairment. Using the multi-sequence magnetic resonance images of the brain and clinical data of the VCI patient group, as well as the multi-sequence magnetic resonance images of the brain and clinical data of the normal human group, an auxiliary diagnostic model for vascular cognitive impairment can be established, improving the diagnostic accuracy of vascular cognitive impairment, helping to detect vascular cognitive impairment early, and enabling treatment as soon as possible.

[0028] Optionally, obtaining the multi-sequence magnetic resonance images of the brain and clinical data includes: obtaining the functional magnetic resonance images, gray matter images, white matter images, and clinical data of several VCI patients, as well as obtaining the functional magnetic resonance images, gray matter images, white matter images, and clinical data of several normal humans; determining the functional magnetic resonance images, gray matter images, and white matter images as the multi-sequence magnetic resonance images of the brain; and labeling the multi-sequence magnetic resonance images of the brain and clinical data with labels used to represent VCI patients or normals.

[0029] Optionally, preprocessing the multi-sequence magnetic resonance images of the brain and clinical data to obtain a heterogeneous multimodal dataset includes: obtaining multi-modal registration data based on the multi-sequence magnetic resonance images of the brain; and performing normalization processing on the multi-modal registration data and clinical data to obtain a heterogeneous multimodal dataset.

[0030] In some embodiments, resampling and registration are performed on the multi-sequence magnetic resonance images to ensure that the positions of the multi-sequence magnetic resonance images correspond and the slice thickness is consistent, and then normalization processing is performed on the multi-modal registration data and clinical data to obtain a heterogeneous multimodal dataset. In this way, after preprocessing the multi-sequence magnetic resonance images of the brain and clinical data, the difficulty of deep learning network training is reduced.

[0031] Optionally, obtaining multi-modal registration data based on the multi-sequence magnetic resonance images of the brain includes: resampling the multi-sequence magnetic resonance images of the brain to obtain multi-sequence magnetic resonance images of the brain with a unified three-dimensional shape; and performing image registration on the multi-sequence magnetic resonance images of the brain with a unified three-dimensional shape to obtain multi-modal registration data.

[0032] Optionally, resample the multi-sequence magnetic resonance images of the brain to obtain multi-sequence magnetic resonance images of the brain with a unified three-dimensional shape, including: resample all the multi-sequence magnetic resonance images of the brain using a preset three-dimensional shape to obtain resampled data consistent with the preset shape; use the cubic spline interpolation algorithm to perform resampling calculations to obtain multi-sequence magnetic resonance images of the brain with a unified three-dimensional shape.

[0033] In some embodiments, uniformly use a three-dimensional shape of H:W:D (height: width: depth) = 73:61:61 to perform target shape sampling on all the multi-sequence magnetic resonance images of the brain to obtain a sampled shape, and then use the cubic spline interpolation algorithm to calculate the sampled shape. In this way, all the multi-sequence magnetic resonance images of the brain are standardized and unified, reducing the difficulty of model construction. Using the cubic spline interpolation algorithm can better maintain the continuity of image details and can better preserve the key features that affect the auxiliary diagnosis efficiency.

[0034] Optionally, perform image registration on the multi-sequence magnetic resonance images of the brain with a unified three-dimensional shape to obtain multi-modal registration data, including: using the open-source ANTS registration library, with the gray matter image of the brain as the fixed-domain image and the white matter and fMRI images as the moving-domain images, align the moving-domain images anatomically to the fixed-domain images.

[0035] In some embodiments, G(MOVE IMAGE) → FIXED IMAGE*, where G() is the mapping function for structural image registration. MOVE IMAGE is the fixed-domain image, i.e., the gray matter image of the brain, and FIXED IMAGE is the moving-domain image, i.e., the white matter and fMRI (functional magnetic resonance imaging) images of the brain.

[0036] Optionally, the heterogeneous multi-modal dataset includes training samples and test samples. Construct an auxiliary diagnosis model for vascular cognitive impairment based on the heterogeneous multi-modal dataset, including: build an initial diagnosis model; train the initial diagnosis model according to the training samples; obtain the accuracy rate of the trained initial diagnosis model according to the test samples; in the case where the accuracy rate is greater than the preset threshold, stop training and obtain the auxiliary diagnosis model for vascular cognitive impairment.

[0037] In some embodiments, multi-sequence magnetic resonance images of the brain and clinical data are used as the input of the model, and the labels corresponding to the multi-sequence magnetic resonance images of the brain and clinical data are used as the output of the model. The initial diagnosis model is trained by deep learning. The multi-sequence magnetic resonance images of the brain and clinical data in the test samples are input into the trained initial diagnosis model to obtain the output values of the initial diagnosis model; the output values corresponding to each multi-sequence magnetic resonance image of the brain and clinical data are respectively compared with the corresponding labels. If they are consistent, the result is correct; if not, the result is wrong. The total number of test samples and the number of correct results are counted, and the number of correct results is divided by the total number of test samples to obtain the accuracy rate of the trained initial diagnosis model. When the accuracy rate is greater than the preset threshold, the training is stopped and the initial diagnosis model corresponding to this accuracy rate is determined as the final auxiliary diagnosis model for vascular cognitive impairment.

[0038] Optionally, HRNET is used as the network backbone of the deep learning neural network, and the initial diagnosis model is built by using the three - sequence fusion method of shallow - middle - tail layer.

[0039] Optionally, in the shallow fusion, multiple modal data are fused together before being input into the model for training. In the middle fusion, the features in the middle sections of different modal networks are concatenated, and the concatenated multi-modal features are classified. In the tail fusion, the features before the last activation function in different modal networks are concatenated using the Concatenate function, and the concatenated multi-modal features are classified. The HRNET network adopts the multi-fusion method of three - sequence fusion of shallow - middle - tail layer, which better preserves the features of the shallow layer and avoids the problem of overfitting easily caused by multi-sequence models.

[0040] In some embodiments, the output layer of the HRNET network predicts whether it is a VCI patient through the sogmoid activation function, where the output for VCI patients is 1 and the output for normal cases is 0.

[0041] In some embodiments, shallow fusion fuses multiple modalities of data together before inputting them into the model for training, so that all image data goes through the complete neural network, and deep features related to the diagnostic efficacy of the images are mined. Tail layer fusion takes the features before the last activation function in different modality networks and performs Concatenate splicing, and a fusion method for classifying the spliced multi-modal features is used to mine the shallow features affecting the diagnosis. The multi-fusion method that adopts three sequential fusions of shallow-middle-tail layers not only mines the deep features related to the diagnostic efficacy of the images, but also does not lose the shallow features affecting the diagnosis. By using a parallel scale network with high and low resolutions, the input-to-output preservation of high-resolution features can be achieved, which is more conducive to the preservation of important fine-grained features and improves the accuracy of the auxiliary diagnosis model for vascular cognitive impairment. At the same time, by integrating deep learning technology and radiomics technology, a deep learning feature extraction network is used to extract depth information, and an imaging genomics model is comprehensively established by combining the depth information, radiomics features and clinical features for clinical auxiliary diagnosis, which improves the accuracy of the diagnosis of vascular cognitive impairment.

[0042] Combined with Figure 2 As shown, this embodiment provides a device for constructing an auxiliary diagnosis model for vascular cognitive impairment, including: a data acquisition module 101, a heterogeneous multi-modal dataset acquisition module 102, and a model construction module 103. The data acquisition module 101 is configured to acquire multi-sequence magnetic resonance images of the brain and clinical data; the heterogeneous multi-modal dataset acquisition module 102 is configured to preprocess the multi-sequence magnetic resonance images of the brain and clinical data to obtain a heterogeneous multi-modal dataset; the heterogeneous multi-modal dataset includes multi-sequence magnetic resonance images and clinical data of the brains of several VCI patients, as well as multi-sequence magnetic resonance images and clinical data of the brains of several normal humans; the model construction module 103 is configured to construct an auxiliary diagnosis model for vascular cognitive impairment according to the heterogeneous multi-modal dataset.

[0043] Optionally, the data acquisition module acquires multi-sequence magnetic resonance images of the brain and clinical data in the following manner, including: acquiring the brain fmri, gray matter images, white matter images and clinical data of several VCI patients, and acquiring the brain fmri images, gray matter images, white matter images and clinical data of several normal humans; determining the brain fmri images, gray matter images, and white matter images as multi-sequence magnetic resonance images of the brain; and labeling the multi-sequence magnetic resonance images of the brain and clinical data with labels for characterizing VCI patients or normals.

[0044] Optionally, the heterogeneous multimodal dataset acquisition module preprocesses brain multi-sequence magnetic resonance images and clinical data to obtain a heterogeneous multimodal dataset in the following ways, including: obtaining multimodal registration data from the brain multi-sequence magnetic resonance images; normalizing the multimodal registration data and clinical data to obtain a heterogeneous multimodal dataset.

[0045] Optionally, the heterogeneous multimodal dataset acquisition module obtains multimodal registration data from the brain multi-sequence magnetic resonance images in the following ways, including: resampling the brain multi-sequence magnetic resonance images to obtain brain multi-sequence magnetic resonance images with a unified three-dimensional shape; performing image registration on the brain multi-sequence magnetic resonance images with a unified three-dimensional shape to obtain multimodal registration data.

[0046] Optionally, the heterogeneous multimodal dataset acquisition module resamples the brain multi-sequence magnetic resonance images to obtain brain multi-sequence magnetic resonance images with a unified three-dimensional shape in the following ways, including: sampling all the brain multi-sequence magnetic resonance images with a preset three-dimensional shape to obtain a sampled shape; using a cubic spline interpolation algorithm to calculate the sampled shape to obtain brain multi-sequence magnetic resonance images with a unified three-dimensional shape.

[0047] Optionally, the heterogeneous multimodal dataset acquisition module performs image registration on the brain multi-sequence magnetic resonance images with a unified three-dimensional shape to obtain multimodal registration data in the following ways, including: using an open-source ANTS registration library, taking the gray matter image as the fixed-domain image of the benchmark, and the white matter and fmri images as the moving-domain images, and aligning the moving-domain images anatomically to the fixed-domain image.

[0048] Optionally, the heterogeneous multimodal dataset includes training samples and test samples. The model construction module constructs an auxiliary diagnosis model for vascular cognitive impairment based on the heterogeneous multimodal dataset in the following ways, including: building an initial diagnosis model; training the initial diagnosis model according to the training samples; obtaining the accuracy rate of the trained initial diagnosis model according to the test samples; and stopping training and obtaining an auxiliary diagnosis model for vascular cognitive impairment when the accuracy rate is greater than a preset threshold.

[0049] Optionally, the model construction module builds an initial diagnosis model in the following ways, including: using HRNET as the network backbone of the deep learning neural network and adopting a three-time sequence fusion method of shallow-middle-tail to build the initial diagnosis model.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the various embodiments of the present invention, and they should all be covered within the scope of the claims and the description of the present invention.

Claims

1. A method for constructing an auxiliary diagnosis model for vascular cognitive impairment, characterized in that Including: Obtain multi - sequence magnetic resonance images of the brain and clinical data; the multi - sequence magnetic resonance images of the brain include functional magnetic resonance images of the brain, gray matter images of the brain, and white matter images of the brain; Pre - process the multi - sequence magnetic resonance images of the brain and the clinical data to obtain a heterogeneous multi - modal dataset, including: resample the multi - sequence magnetic resonance images of the brain to obtain multi - sequence magnetic resonance images of the brain with a unified three - dimensional shape, perform image registration on the multi - sequence magnetic resonance images of the brain with the unified three - dimensional shape to obtain multi - modal registration data; the image registration includes using the open - source ANTS registration library, taking the gray matter image of the brain as the fixed - domain image, and the white matter image and the functional magnetic resonance image of the brain as the moving - domain images, and aligning the moving - domain images anatomically to the fixed - domain image; The heterogeneous multi - modal dataset includes multi - sequence magnetic resonance images and clinical data of several VCI patients, as well as multi - sequence magnetic resonance images and clinical data of several normal humans; Construct an auxiliary diagnosis model for vascular cognitive impairment based on the heterogeneous multi - modal dataset, including: using HRNET as the network backbone of the deep - learning neural network, and adopting a three - stage sequence fusion method of shallow - middle - tail to build an initial diagnosis model; shallow fusion fuses multiple modal data together before inputting them into the model for training, middle fusion concatenates the features in the middle sections of different modal networks and classifies the concatenated multi - modal features; tail fusion concatenates the features before the last activation function in different modal networks using the Concatenate function and classifies the concatenated multi - modal features.

2. The method according to claim 1, wherein Obtain multi - sequence magnetic resonance images of the brain and clinical data, including: Obtain functional magnetic resonance images of the brain, gray matter images of the brain, white matter images of the brain, and clinical data of several VCI patients, and obtain functional magnetic resonance images of the brain, gray matter images of the brain, white matter images of the brain, and clinical data of several normal humans; determine the functional magnetic resonance images of the brain, gray matter images of the brain, and white matter images of the brain as multi - sequence magnetic resonance images of the brain; Label the multi - sequence magnetic resonance images of the brain and the clinical data, and the label is used to represent VCI patients or normals.

3. The method according to claim 2, characterized in that, Pre - process the multi - sequence magnetic resonance images of the brain and the clinical data to obtain a heterogeneous multi - modal dataset, including: Obtain multi - modal registration data based on the multi - sequence magnetic resonance images of the brain; Perform normalization processing on the multi - modal registration data and the clinical data to obtain a heterogeneous multi - modal dataset.

4. The method according to claim 1, wherein Resample the multi - sequence magnetic resonance images of the brain to obtain multi - sequence magnetic resonance images of the brain with a unified three - dimensional shape, including: Resample all the multi - sequence magnetic resonance images of the brain using a preset three - dimensional shape to obtain resampled data consistent with the preset three - dimensional shape; Use the cubic spline interpolation algorithm for resampling calculation to obtain multi - sequence magnetic resonance images of the brain with a unified three - dimensional shape.

5. The method according to claim 1, wherein The heterogeneous multi - modal dataset includes training samples and test samples. Construct an auxiliary diagnosis model for vascular cognitive impairment based on the heterogeneous multi - modal dataset, including: Build an initial diagnosis model; Train the initial diagnosis model according to the training samples; Obtain the accuracy rate of the trained initial diagnosis model according to the test samples; When the accuracy rate is greater than the preset threshold, stop training and obtain an auxiliary diagnosis model for vascular cognitive impairment.

6. An apparatus for constructing an auxiliary diagnosis model for vascular cognitive impairment, characterized in that, Including: A data acquisition module configured to acquire multi-sequence magnetic resonance images of the brain and clinical data; the multi-sequence magnetic resonance images of the brain include functional magnetic resonance images of the brain, gray matter images of the brain, and white matter images of the brain; A heterogeneous multi-modal data set acquisition module configured to preprocess the multi-sequence magnetic resonance images of the brain and the clinical data to obtain a heterogeneous multi-modal data set; including: resampling the multi-sequence magnetic resonance images of the brain to obtain multi-sequence magnetic resonance images of the brain with a unified three-dimensional shape, and performing image registration on the multi-sequence magnetic resonance images of the brain with the unified three-dimensional shape to obtain multi-modal registration data; the performing image registration includes using an open-source ANTS registration library, with the gray matter image of the brain as the fixed-domain image, and the white matter image and the functional magnetic resonance image of the brain as the moving-domain images, and aligning the moving-domain images anatomically to the fixed-domain image; the heterogeneous multi-modal data set includes multi-sequence magnetic resonance images and clinical data of several VCI patients, as well as multi-sequence magnetic resonance images and clinical data of several normal human bodies; A model construction module configured to construct an auxiliary diagnosis model for vascular cognitive impairment according to the heterogeneous multi-modal data set, including: using HRNET as the network backbone of the deep learning neural network, and adopting a shallow-middle-tail three-time sequence fusion method to build an initial diagnosis model; the shallow fusion fuses multiple modal data together before inputting the model and then inputs the model for training, the middle fusion splices the features in the middle sections of different modal networks, and classifies the spliced multi-modal features; the tail fusion splices the features before the last activation function in different modal networks using the Concatenate function, and classifies the spliced multi-modal features.

Images