Vascular cognitive impairment assessment method and device, electronic equipment and storage medium

Abstract

The present application relates to the technical field of data processing, and discloses a vascular cognitive impairment evaluation method and device, electronic equipment and a storage medium, the method comprising: acquiring diffusion tensor imaging data of a target object; based on the diffusion tensor imaging data, calculating a target diffusion tensor parameter and generating a target diffusion tensor parameter map; inputting the target diffusion tensor parameter map into a pre-constructed deep learning model to obtain a target saliency weight map of the target object; based on the target saliency weight map and a pre-established mutual information atlas set, determining a structural similarity score between the target saliency weight map and each mutual information atlas in the mutual information atlas set; and based on the structural similarity score, determining a cognitive impairment risk level represented by a target neuropsychological scale. The present application can accurately identify the dysfunction of a patient in multiple cognitive domains and provide personalized cognitive impairment risk level evaluation for each patient.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to a method, device, electronic device, and storage medium for assessing vascular cognitive impairment. Background Technology

[0002] Vascular cognitive impairment (VCI) is stroke or subclinical vascular brain injury caused by cerebrovascular lesions and their risk factors. VCI encompasses all stages of the disease, from mild cognitive impairment originating from cerebrovascular disease to vascular dementia. Based on its clinical manifestations, it is classified into three subtypes: non-dementia vascular cognitive impairment, vascular dementia, and mixed dementia, with non-dementia vascular cognitive impairment being the most common subtype. With the increasing prevalence of VCI, early identification of VCI can facilitate early intervention for patients and has significant clinical implications.

[0003] Currently, the diagnosis of VCI mainly relies on multimodal assessments of clinical manifestations, magnetic resonance imaging (MRI), and neuropsychological scales. This assessment method suffers from high subjectivity and dependence on specialized cognitive physicians for evaluation. Studies have reported differences in brain structure, function, and network characteristics between VCI patients and cognitively normal subjects. Based on this, some researchers have begun manually extracting features from MRI images to build traditional machine learning models for VCI identification. Existing machine learning-based VCI identification methods rely on manual feature engineering, making their results susceptible to subjective influences, and model performance largely depends on the quality and sufficiency of feature extraction. Furthermore, simple image indicators may not reflect changes in brain function and structure, limiting the performance of diagnostic models. Deep learning-based methods, however, can automatically extract image features without requiring manual feature selection based on experience, capturing the differences in brain image features between VCI and normal individuals in a more objective way, thus enabling VCI identification.

[0004] However, existing machine learning and deep learning research mainly focuses on the overall identification of VCI, while relatively neglecting the heterogeneous characteristics of cognitive impairment in VCI patients. That is, different patients may exhibit different patterns of cognitive dysfunction; for example, some patients mainly have memory impairment, while others show more prominent executive dysfunction. Current research cannot individualize the risk of impairment in multiple cognitive domains in VCI patients. Summary of the Invention

[0005] This invention provides a method, device, electronic device, and storage medium for assessing vascular cognitive impairment (VCI) to address the lack of individualized assessment of the risk of functional impairment in multiple cognitive domains in VCI patients in the prior art.

[0006] In a first aspect, the present invention provides a method for assessing vascular cognitive impairment, the method comprising: Acquire diffusion tensor imaging data of the target object; Based on diffusion tensor imaging data, the target diffusion tensor parameters are calculated, and a target diffusion tensor parameter map is generated. Input the target diffusion tensor parameter map into a pre-built deep learning model to obtain the target saliency weight map of the target object; Based on the target saliency weight map and a pre-established mutual information graph set, the structural similarity score between the target saliency weight map and each mutual information graph in the mutual information graph set is determined; the mutual information graph represents the degree of statistical association between the diffusion tensor parameter map and the target neuropsychological scale. Based on structural similarity scores, the risk level of cognitive impairment represented by the target neuropsychological scale is determined.

[0007] In one alternative implementation, the target dispersion tensor parameter includes at least one of the following: anisotropy parameter, mean diffusivity, axial diffusion coefficient, and radial diffusion coefficient. The target neuropsychological scale includes at least one of the following: MMSE, MoCA, immediate recall, delayed recall, link test-A, and link test-B.

[0008] In one alternative implementation, the mutual information graph is established through the following steps: Obtain the first sample dataset, which includes diffusion tensor parameter graphs of sample objects and multiple preset types of neuropsychological scales; the neuropsychological scales include test data of sample objects, and there are multiple sample objects; Standardize all the diffusion tensor parameter graphs; Based on the diffuse tensor parameter map of all sample objects after standardization, the same diffuse tensor parameter values ​​under the same voxel are used to form a diffuse tensor parameter matrix. Based on the neuropsychological scales of all sample subjects, the scores of the same scale are used to form a score matrix. Based on the diffusion tensor parameter matrix and the score matrix, mutual information is calculated to obtain a mutual information map corresponding to each preset type of neuropsychological scale.

[0009] In one optional implementation, based on the target saliency weight map and a pre-established mutual information graph set, a structural similarity score is determined between the target saliency weight map and each mutual information graph in the mutual information graph set, including: Determine the first statistical data for the target significance weighting plot; The second statistical data for determining the target mutual information graph is any mutual information graph in the mutual information graph set; both the first and second statistical data include: mean and variance. Determine the covariance between the target saliency weight map and the target mutual information map; Based on the first statistical data, the second statistical data, and the covariance, the structural similarity score between the target saliency weight map and the target mutual information map is determined.

[0010] In one alternative implementation, the risk level of cognitive impairment represented by the target neuropsychological scale is determined based on structural similarity scores, including: Determine the corresponding rating scale for the target neuropsychological scale; Based on structural similarity scores and a rating scale comparison table, the risk level of cognitive impairment represented by the target neuropsychological scale was determined.

[0011] In one optional implementation, determining a rating scale corresponding to the target neuropsychological scale includes: Obtain a second sample dataset, which includes a saliency weight map of the sample objects and multiple preset types of neuropsychological scales; the neuropsychological scales include test data of the sample objects, and there are multiple sample objects; Calculate the structural similarity score between the saliency weight map of each sample object and each preset category of neuropsychological scale to obtain a set of structural similarity scores; Clustering is performed based on the structural similarity score set to obtain a one-to-one correspondence table of score levels for each preset category of neuropsychological scale.

[0012] In one alternative implementation, the target diffusion tensor parameters are the optimal diffusion tensor parameters determined when constructing the deep learning model.

[0013] In a second aspect, the present invention provides a vascular cognitive impairment assessment device, the device comprising: The acquisition module is used to acquire diffusion tensor imaging data of the target object; The generation module is used to calculate the target diffusion tensor parameters based on diffusion tensor imaging data and generate a target diffusion tensor parameter map. The learning module is used to input the target diffusion tensor parameter map into a pre-built deep learning model to obtain the target saliency weight map of the target object; The scoring module is used to determine the structural similarity score between the target saliency weight map and each mutual information graph in the mutual information graph set, based on the target saliency weight map and a pre-established mutual information graph set; the mutual information graph represents the degree of statistical association between the diffusion tensor parameter map and the target neuropsychological scale. The determination module is used to determine the risk level of cognitive impairment represented by the target neuropsychological scale based on structural similarity scores.

[0014] Thirdly, the present invention provides an electronic device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the vascular cognitive impairment assessment method of the first aspect or any corresponding embodiment described above.

[0015] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the vascular cognitive impairment assessment method of the first aspect or any corresponding embodiment described above.

[0016] Fifthly, the present invention provides a computer program product, including computer instructions for causing a computer to execute the vascular cognitive impairment assessment method of the first aspect or any corresponding embodiment described above.

[0017] The beneficial effects of this invention are as follows: This invention, by combining diffusion tensor imaging and deep learning models, generates more refined saliency weight maps and compares them with mutual information maps. This method not only reveals the correlation between diffusion tensor parameters and various neuropsychological scales but also calculates structural similarity scores, thus providing each patient with a personalized assessment of their cognitive impairment risk level. This individualized assessment helps to more accurately identify functional impairments in specific cognitive domains when users cannot take multiple cognitive scales, enabling individualized assessment of multidimensional cognitive function and ultimately leading to more effective intervention strategies. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the first process of the assessment method for vascular cognitive impairment according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the MMSE clustering hierarchical results according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the MoCA clustering hierarchical results according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the instant recall clustering hierarchical results according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the delayed recall clustering hierarchical results according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the connection test-A clustering hierarchical results according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the connection test-B clustering hierarchical results according to an embodiment of the present invention; Figure 8 This is a structural block diagram of a vascular cognitive impairment assessment device according to an embodiment of the present invention; Figure 9 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0022] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0023] To overcome the limitations of existing research, which is limited to VCI identification and makes it difficult to achieve individualized cognitive impairment risk assessment for patients, this invention provides an embodiment of a method for assessing vascular cognitive impairment. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than that shown here.

[0024] This embodiment provides a method for assessing vascular cognitive impairment, which can be used on servers, terminals, and mobile terminals, such as mobile phones and tablets. Figure 1 This is a flowchart of a method for assessing vascular cognitive impairment according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain diffusion tensor imaging data of the target object.

[0025] Diffusion tensor imaging (DTI) is a magnetic resonance imaging technique used to characterize the diffusion properties of water molecules in tissues. During DTI, diffusion-sensitive sequences with multiple gradient directions are used to acquire magnetic resonance images. Typically, at least six different directions are selected to measure the diffusion of water molecules, thus constructing a complete diffusion tensor.

[0026] In this embodiment, a Siemens 3T magnetic resonance imaging system can be used to scan the target object and acquire diffusion tensor imaging data. The scanning parameters for diffusion tensor imaging are as follows: diffusion tensor imaging is acquired using a diffusion-weighted dual-spin echo-echo planar imaging sequence with a repetition time of 8000 ms, an echo time of 96 ms, 64 diffusion-weighted directions, a b-value of 1000 s / mm², a deflection angle of 90°, a field of view of 224×224 mm², a matrix of 128×128, and a planar resolution of 1.75×1.75 mm² / voxel. The diffusion tensor imaging data includes anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD). To facilitate further analysis and localization, the diffusion tensor parameter map can be registered from the individual space to the standard space to obtain the diffusion tensor parameter map in the standard space (Montreal Neurological Institute space, MNI).

[0027] Step S102: Based on the diffusion tensor imaging data, calculate the target diffusion tensor parameters and generate a target diffusion tensor parameter map.

[0028] Specifically, the acquired raw diffusion tensor imaging data undergoes preprocessing, including image format conversion, resampling, brain tissue segmentation, magnetic field eddy current correction, and head motion correction. A tensor model is used to model and analyze the diffusion characteristics of water molecules in the tissue to obtain a tensor matrix, thereby calculating the diffusion tensor parameter image. Each diffusion tensor parameter corresponds to a diffusion tensor parameter map.

[0029] In some optional implementations, the target dispersion tensor parameter includes at least one of the following: anisotropy parameter, average diffusivity, axial diffusion coefficient, and radial diffusion coefficient. That is, the target dispersion tensor parameter in this embodiment can be one or more, and can be one of the following combinations: (1) FA; (2) MD; (3) AD; (4) RD; (5) FA and MD; (6) MD and AD; (7) AD and RD; (8) FA and AD; (9) MD and RD; (10) FA and RD; (11) FA, MD, and AD; (12) MD, AD, and RD; (13) FA, AD, and RD; (14) FA, MD, and RD; (15) FA, MD, AD, and RD.

[0030] In some alternative implementations, the target diffusion tensor parameters are the optimal diffusion tensor parameters determined when constructing the deep learning model.

[0031] The specific target diffusion tensor parameters can be selected based on the optimal combination determined by the deep learning model. That is, in order to find the optimal combination of target diffusion tensor parameters to identify VCI, the deep learning model will use the above 15 parameter combinations for training in the early stages of training.

[0032] Step S103: Input the target diffusion tensor parameter map into the pre-built deep learning model to obtain the target saliency weight map of the target object.

[0033] When there are multiple target diffusion tensor parameter maps, the multiple target diffusion tensor parameter maps are connected in the channel dimension and then input into the deep learning model.

[0034] Regarding the construction of the deep learning model in this embodiment, firstly, diffusion tensor imaging data of cognitively normal subjects and patients with vascular cognitive impairment are acquired, and diffusion tensor parameter maps are calculated; then, a generalizable deep learning model is constructed. Specifically, DenseNet can be used as the feature extraction backbone network structure, combined with a VCI classifier, a domain classifier, and a gradient inversion layer to construct a generalizable deep learning model for VCI recognition.

[0035] Specifically, the deep learning model in this embodiment consists of three parts: a feature extractor, a VCI (Visual Identity Graph) classifier, and an unsupervised domain adaptation module. The feature extractor comprises four densely connected modules, three transition layers, one global average pooling layer, and one fully connected layer. Each densely connected module contains one densely connected layer, and each dense layer consists of two convolutional layers. BatchNorm is used for feature normalization between the convolutional layers, and the ReLU activation function adds non-linearity to the features. The transition layer includes a convolutional layer combining BatchNorm batch normalization and the ReLU activation function. Following the transition layer is a stride average pooling layer of 2, used to downsample the feature map. The first convolutional layer can be designed with 64 filters to expand the feature dimension, and each subsequent convolutional layer adds 32 filters. At the output of the model's feature extractor, a random feature dropout layer with a probability of 0.5 is applied to alleviate overfitting. The VCI classifier consists of one fully connected layer and uses the softmax function to output the VCI probability. The unsupervised domain adaptation module adopts a framework based on a domain adversarial neural network strategy. The unsupervised domain adaptive module structure consists of a domain classifier and a gradient inversion layer. The domain classifier can be composed of three fully connected layers (dimensions: 256→128→2), combined with a batch normalization layer and a ReLU activation function.

[0036] Furthermore, in this embodiment, the deep learning model uses a pre-collected internal dataset (including diffusion tensor imaging data of normal subjects and patients with vascular cognitive impairment, as well as VCI data, etc.) to train its VCI recognition capability. 20% of the dataset is randomly assigned as the test set, and the remaining 80% is used as the training set for five-fold cross-validation during model training. After cross-validation, the optimal training parameters are determined and used to retrain the model on the complete training set.

[0037] During model training, data augmentation was performed using intensity and spatial transformations. Specifically, the intensity was randomly shifted from 0.9 to 1.1 with a 30% probability, the image was flipped along the first axis, or the image range was randomly scaled from 0.9 to 1.1 with a 20% probability. The model was trained with a batch size of 16 and a learning rate of 0.00005 for 50 epochs. Cosine annealing was used to dynamically adjust the learning rate. Cross-entropy loss was chosen as the loss function, and the model parameters were optimized using the Adaptive Moments Estimation (Adam) optimizer with a weight decay of 0.001.

[0038] When faced with unknown external data, the model undergoes transfer training, using labeled data from the training set (source domain) and unlabeled data from the external data (target domain). During training, the model's unsupervised domain adaptation module multiplies the gradient by a negative scalar during backpropagation, forcing the feature extractor to learn domain-independent features. Therefore, this unsupervised domain adaptation module jointly optimizes two objectives: minimizing the VCI recognition loss on the source domain data and maximizing the confusion of the domain classifier when recognizing data from the source and target domains. All classifiers are trained using cross-entropy loss. The scheduling parameter λ of the domain classifier is dynamically increased according to the following formula: ; in, The weights represent the training progress, varying linearly from 0 to 1. These adaptive weights help achieve stable adversarial training in the early stages of transfer training and enhance domain adaptability in later stages. The domain classifier and the VCI recognition classifier are jointly trained using a weighted sum of losses from the following formula: ; in, It is the VCI classification loss of the source domain. and These are the domain classification losses for the source and target domains, respectively. Set it to 0.1.

[0039] In this embodiment, the Adam optimizer is used to optimize the model with a learning rate of 0.00001 and a weight decay of 0.001. During training, batch data is sampled alternately from the two domains.

[0040] To find the optimal combination of DTI data to identify VCI, the deep learning model was trained using the following 15 combinations of DTI data: (1) FA; (2) MD; (3) AD; (4) RD; (5) FA and MD; (6) MD and AD; (7) AD and RD; (8) FA and AD; (9) MD and RD; (10) FA and RD; (11) FA, MD and AD; (12) MD, AD and RD; (13) FA, AD and RD; (14) FA, MD and RD; (15) FA, MD, AD and RD.

[0041] After inputting the target diffusion tensor parameter map into the deep learning model, guided backpropagation is used to visualize the model's areas of interest, generating an individual-level saliency weight map for the target object to obtain the deep learning model's attention patterns to the target object's brain map. This saliency weight map shows the saliency of VCI patients in specific brain regions. For example, in this saliency weight map, brain regions with high weights represent areas most closely associated with the pathological mechanisms of VCI (such as the frontal lobe and hippocampus with severe white matter osteoporosis), reflecting abnormal features at the brain structural level.

[0042] Step S104: Based on the target saliency weight map and the pre-established mutual information graph set, determine the structural similarity score between the target saliency weight map and each mutual information graph in the mutual information graph set; the mutual information graph represents the degree of statistical association between the diffusion tensor parameter map and the target neuropsychological scale.

[0043] In some alternative implementations, the target neuropsychological scale includes at least one of the following: MMSE, MoCA, immediate recall, delayed recall, linking test-A, and linking test-B.

[0044] Among them, the MMSE (Mini-Mental State Examination) can assess aspects including orientation (time and place), attention and calculation ability, memory, language ability, and visuospatial ability; the MoCA (Montreal Cognitive Assessment), compared to the MMSE, is better able to identify mild cognitive impairment and covers multiple cognitive domains, including attention, executive function, visuospatial ability, language, memory, abstract thinking, and orientation; Immediate recall usually involves the ability to remember phrases or words in a short period of time, reflecting the function of working memory and short-term memory; Delayed recall asks the subject to recall previously presented information after a period of time (usually several minutes), usually after immediate recall, and can assess long-term memory ability; the Connect-A Test (TMT-A) is a test used to assess attention, visual search, and executive function, mainly examining attention and visual-motor coordination; the Connect-B Test (TMT-B) is an extension of the TMT-A, adding complexity, and can assess an individual's flexibility, task switching ability, and higher-level executive function.

[0045] MMSE and MoCA are generally overall cognitive or global cognitive function indicators. Immediate recall and delayed recall are classified into the memory domain. Among them, immediate recall reflects more of learning ability and short-term / working memory components, while delayed recall is more biased towards episodic memory and memory consolidation ability. Trail Making Test-A is generally classified into the attention and processing speed domain, which mainly reflects visual search, sustained attention and mental processing speed. Trail Making Test-B is usually classified into the executive function domain.

[0046] Mutual information maps are obtained by calculating mutual information between diffusion tensor parameter maps and each neuropsychological scale respectively. When the target neuropsychological scales include the above six neuropsychological scales (MMSE, MoCA, immediate recall, delayed recall, Trail Making Test-A, Trail Making Test-B), the mutual information map set includes six mutual information maps, which are respectively: MMSE mutual information map, MoCA mutual information map, immediate recall mutual information map, delayed recall mutual information map, Trail Making Test-A mutual information map, and Trail Making Test-B mutual information map. The target neuropsychological scale is any one of the six neuropsychological scales.

[0047] In this embodiment, the target saliency weight map needs to calculate the structural similarity index measurement score with each mutual information map respectively, so as to obtain the structural similarity index measurement score corresponding to each neuropsychological scale one by one.

[0048] For example, the SSIM score of the target saliency weight map and the MMSE mutual information map is 0.88; the SSIM score of the target saliency weight map and the MoCA mutual information map is 0.75, and so on.

[0049] Step S105: determining the cognitive impairment risk level represented by the target neuropsychological scale based on the structural similarity index measurement scores.

[0050] An example is given below for illustration: according to the structural similarity index measurement scores, the following thresholds are set to classify cognitive impairment risk levels: High risk: SSIM score > 0.80; Medium risk: 0.70 < SSIM score ≤ 0.80; Low risk: SSIM score ≤ 0.70; For example, when the SSIM score of the target saliency weight map and the MMSE mutual information map is 0.88, it indicates that the brain regions related to MMSE may have obvious problems in the cognitive function assessment, therefore the cognitive impairment risk level of the target subject is high risk. If immediate recall and delayed recall show high structural similarity index measurement scores, it indicates that the memory-related cognitive risk is relatively high.

[0051] High-risk patients may require immediate further evaluation and intervention. Intermediate-risk patients need regular monitoring and cognitive training should be considered. Low-risk patients may not require immediate intervention; simply maintaining a healthy lifestyle and regular checkups are sufficient.

[0052] Current research faces numerous challenges in assessing the risk of functional impairment in multiple cognitive domains in patients with mild vascular cognitive impairment, particularly in terms of individualized assessment. Traditional methods often rely on single neuropsychological scales, which cannot comprehensively reflect the patient's actual functional status across different cognitive domains. This limitation makes it difficult for clinicians to develop targeted interventions, thereby affecting patients' treatment outcomes and quality of life.

[0053] This invention, by combining diffusion tensor imaging and deep learning models, generates more refined saliency weight maps and compares them with mutual information maps. This method not only reveals the correlation between diffusion tensor parameters and various neuropsychological scales but also calculates structural similarity scores, thus providing each patient with a personalized assessment of their cognitive impairment risk level. This individualized assessment helps to more accurately identify functional impairments in specific cognitive domains when users cannot take multiple cognitive scales, enabling individualized assessment of multidimensional cognitive function and ultimately leading to more effective intervention strategies.

[0054] In some alternative implementations, the mutual information graph is established through the following steps: Step a1: Obtain the first sample dataset, which includes diffusion tensor parameter maps of the sample objects and multiple preset types of neuropsychological scales. The neuropsychological scales include test data of the sample objects, and there are multiple sample objects. Specifically, data from all sample objects that simultaneously possess diffusion tensor imaging derived parameter maps (such as FA, MD, RD, AD) and data from six neuropsychological scales (MMSE, MoCA, immediate recall, delayed recall, TMT-A, TMT-B) can be collected.

[0055] Step a2: Standardize all diffuse tensor parameter graphs.

[0056] All diffuse tensor parametric maps undergo standard image processing procedures, such as registration, normalization, and smoothing, to ensure that each location (voxel) is comparable across different objects.

[0057] Step a3: Based on the standardized diffusion tensor parameter map of all sample objects, the same diffusion tensor parameter values ​​under the same voxel are combined to form a diffusion tensor parameter matrix. For each voxel position, the diffusion tensor parameter values ​​of all sample objects at that voxel can be selected to form a column of data, which is the diffusion tensor parameter matrix in this embodiment.

[0058] Step a4: Based on the neuropsychological scales used by all sample subjects, the scores for the same item on the scale are combined to form a score matrix. This can be achieved by selecting the scores of all sample subjects for the same test item on the same neuropsychological scale, forming a single data column, which is the score matrix in this embodiment.

[0059] Step a5: Based on the diffusion tensor parameter matrix and the score matrix, perform mutual information calculation to obtain the mutual information map corresponding to each preset type of neuropsychological scale.

[0060] Using two columns of data as an example, calculate the mutual information score between the two columns of data according to the following formula to obtain a mutual information (MI) value.

[0061] ; in, and It is a variable and variables The boundary probability distribution, This is the joint probability distribution of the two. For each voxel, the MI was calculated using a permutation test of 100 times; statistical significance was assessed with p < 0.05, and false positive correction for multiple comparisons was performed using a Gaussian random field (GRF).

[0062] The spatial location of this MI value is the same as the spatial location of the voxels used in the diffusion tensor parametric image. This process is repeated for all voxels throughout the entire brain, and all voxels are pieced together to form a mutual information map. This is used to determine the neuroanatomical relevance of the neuropsychological scale. The mutual information map quantifies the statistical dependence between the diffusion tensor parametric image and neuropsychological performance. Subsequent permutation tests and multiple comparison corrections are performed to test statistical significance.

[0063] In some optional implementations, step S104 above, namely, determining the structural similarity score between the target saliency weight map and each mutual information graph in the mutual information graph set based on the target saliency weight map and the pre-established mutual information graph set, includes: Step S1041: Determine the first statistical data of the target significance weight map.

[0064] Step S1042: Determine the second statistical data of the target mutual information graph. The target mutual information graph is any mutual information graph in the mutual information graph set. Both the first and second statistical data include: mean and variance.

[0065] Step S1043: Determine the covariance between the target saliency weight map and the target mutual information map; Step S1044: Based on the first statistical data, the second statistical data, and the covariance, determine the structural similarity score between the target saliency weight map and the target mutual information map.

[0066] Specifically, the structural similarity score (SSIM) between the saliency weighted graph and the mutual information graph can be calculated using the following formula: ; Where C1 and C2 are local constants to avoid the denominator being zero, μ x and μ y These are the mean values ​​of the significance weight plot and the mutual information plot, respectively, σ x and σ y These are the variances of the significance weighting plot and the mutual information plot, respectively, σ xy It is the covariance between the saliency weight map and the mutual information map.

[0067] The target mutual information graph can be any one of the following: MMSE mutual information graph, MoCA mutual information graph, immediate recall mutual information graph, delayed recall mutual information graph, connection test-A mutual information graph, and connection test-B mutual information graph. The target saliency weight graph needs to be used to calculate a structural similarity score with each target mutual information graph separately.

[0068] In some optional implementations, step S105 above, namely determining the risk level of cognitive impairment represented by the target neuropsychological scale based on structural similarity scoring, includes: Step S1051: Determine the rating scale corresponding to the target neuropsychological scale.

[0069] In some optional implementations, step S1051 above includes: Step b1: Obtain the second sample dataset, which includes a saliency weight map of the sample objects and multiple preset types of neuropsychological scales; the neuropsychological scales include test data of the sample objects, and there are multiple sample objects.

[0070] Step b2: Calculate the structural similarity score between the saliency weight map of each sample object and each preset type of neuropsychological scale to obtain a set of structural similarity scores.

[0071] Step b3: Clustering is performed based on the structural similarity score set to obtain a one-to-one rating level comparison table for each preset category of neuropsychological scale.

[0072] Step S1052: Based on the structural similarity score and the score level comparison table, determine the cognitive impairment risk level represented by the target neuropsychological scale.

[0073] Specifically, saliency weight maps for VCI patients are collected. These maps can be generated according to steps S101 to S103 above, in 3D voxel maps format, consistent with the image space. Then, mutual information maps (MI value maps calculated from all samples at the voxel level and obtained after statistical thresholding / correction) are determined for each patient under six neuropsychological scales. The saliency weight map is a pixel / voxel brightness map that reflects the most important pixel / voxel brightness for a particular VCI patient in the model's judgment. The mutual information map statistically shows which voxel diffusion tensor parameter values ​​are strongly correlated with a certain cognitive scale (group results). Then, the SSIM score between the two maps is calculated. The SSIM score is an indicator that measures the similarity between two images in terms of brightness, contrast, and local structure; the value range is typically [0,1] (closer to 1 indicates greater similarity in image patterns). The resulting similarity quantifies the degree of matching between the model-generated saliency weight map of VCI patients and specific anatomical relevance in neuropsychology.

[0074] For each cognitive domain, scale scores were clustered using an unsupervised K-means clustering algorithm based on patients' SSIM scores, and the optimal number of clusters (K) was determined using the elbow method. The resulting clusters were interpreted as representing different degrees of structural-neuropsychological similarity, thus representing different cognitive domains and varying degrees of cognitive impairment risk. Similarity scores between groups were statistically compared using the Kruskal-Wallis test, followed by a Dunn test with Bonferroni correction. The results of multi-cognitive domain impairment risk stratification are referenced. Figures 2 to 7 As shown. Among them, Figure 2 For MMSE clustering hierarchical results, Figure 3 For MoCA clustering hierarchical results, Figure 4 To instantly recall clustering and stratification results, Figure 5 To delay recall of clustering stratification results, Figure 6 For the connection test - A clustering hierarchical results, Figure 7 This is the result of the B-cluster stratification test. The vertical axis represents the similarity score, and the horizontal axis represents the risk level.

[0075] A higher SSIM score reflects a greater structural consistency between individual white matter changes and neuropsychological scales, indicating a higher risk of cognitive impairment. Statistical differences between clusters were assessed using the Kruskal-Wallis test and the Dunn test (with Bonferroni correction). * indicates p < 0.05; *** indicates p < 0.001.

[0076] In this embodiment, an unsupervised clustering method is proposed. This method can stratify the impairment risk of VCI patients in six core cognitive domains and provide three levels of risk rating: low, medium, and high, thereby achieving individualized assessment of the patient's multidimensional cognitive function status.

[0077] This invention innovatively combines diffusion tensor imaging, multi-cognitive domain neuropsychological scales, and generalizable deep learning methods to propose an individualized cognitive impairment risk analysis method for vascular cognitive impairment based on a salient weight map of a deep learning model. This method can automatically and efficiently extract cognitive impairment-related features from imaging data, overcoming the limitation of existing methods that can only identify VCI, and completing a multi-cognitive domain impairment risk assessment for VCI patients.

[0078] This invention not only analyzes the decision-making basis of deep learning models through saliency visualization, but also maps multiple neurocognitive scale scores to the mutual information space of diffusion tensor images, thereby obtaining quantifiable and interpretable brain structure-cognitive function correspondence patterns. Based on this mechanistic association, it enables individualized assessment of the risk of impairment in multiple cognitive domains in VCI patients, possessing both interpretability and clinical usability. It overcomes the technical bottleneck of existing methods that cannot perform individualized cognitive risk assessment for vascular cognitive impairment.

[0079] This embodiment also provides a vascular cognitive impairment assessment device, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0080] This embodiment provides a device for assessing vascular cognitive impairment, such as... Figure 8 As shown, the device includes: The acquisition module 301 is used to acquire the diffusion tensor imaging data of the target object; The generation module 302 is used to calculate the target diffusion tensor parameters based on diffusion tensor imaging data and generate a target diffusion tensor parameter map; the target diffusion tensor parameters are the optimal diffusion tensor parameters determined when constructing the deep learning model.

[0081] Learning module 303 is used to input the target diffusion tensor parameter map into a pre-built deep learning model to obtain the target saliency weight map of the target object; Scoring module 304 is used to determine the structural similarity score between the target saliency weight map and each mutual information graph in the mutual information graph set based on the target saliency weight map and a pre-established mutual information graph set. The mutual information graph represents the degree of statistical association between the diffusion tensor parameter map and the target neuropsychological scale. The target diffusion tensor parameter includes at least one of the following: anisotropy parameter, average diffusion rate, axial diffusion coefficient, and radial diffusion coefficient. The target neuropsychological scale includes at least one of the following: MMSE, MoCA, immediate recall, delayed recall, connection test-A, and connection test-B.

[0082] Module 305 is used to determine the risk level of cognitive impairment represented by the target neuropsychological scale based on structural similarity scores.

[0083] In some alternative embodiments, the apparatus further includes: The module is used to acquire a first sample dataset, which includes diffusion tensor parameter maps of sample objects and multiple preset types of neuropsychological scales. The neuropsychological scales include test data of the sample objects, and there are multiple sample objects. All diffusion tensor parameter maps are standardized. Based on the standardized diffusion tensor parameter maps of all sample objects, the same diffusion tensor parameter values ​​under the same voxel are combined to form a diffusion tensor parameter matrix. Based on the neuropsychological scales of all sample objects, the scores of the same scale are combined to form a score matrix. Based on the diffusion tensor parameter matrix and the score matrix, mutual information is calculated to obtain a mutual information map corresponding to each preset type of neuropsychological scale.

[0084] In some alternative implementations, the scoring module 304 is specifically used for: Determine the first statistical data for the target significance weighting plot; The second statistical data for determining the target mutual information graph is any mutual information graph in the mutual information graph set; both the first and second statistical data include: mean and variance. Determine the covariance between the target saliency weight map and the target mutual information map; Based on the first statistical data, the second statistical data, and the covariance, the structural similarity score between the target saliency weight map and the target mutual information map is determined.

[0085] In some alternative implementations, the determining module 305 is specifically used for: The process involves determining a rating scale correspondence table for the target neuropsychological scale, including: acquiring a second sample dataset, which includes a saliency weight map of the sample objects and multiple pre-defined categories of neuropsychological scales; the neuropsychological scales include test data of the sample objects, and there are multiple sample objects; calculating the structural similarity score between the saliency weight map of each sample object and each pre-defined category of neuropsychological scale to obtain a set of structural similarity scores; and performing clustering based on the set of structural similarity scores to obtain a one-to-one rating scale correspondence table for each pre-defined category of neuropsychological scale.

[0086] Based on structural similarity scores and a rating scale comparison table, the risk level of cognitive impairment represented by the target neuropsychological scale was determined.

[0087] The vascular cognitive impairment assessment device provided in this embodiment of the invention can execute the vascular cognitive impairment assessment method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the above modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0088] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0089] The following is a detailed reference. Figure 9 This diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 901, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 902 or a program loaded from memory 908 into random access memory (RAM) 903. The RAM 903 also stores various programs and data required for the operation of the electronic device. The processor 901, ROM 902, and RAM 903 are interconnected via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0090] Typically, the following devices can be connected to I / O interface 905: input devices 906 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 907 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 908 including, for example, magnetic tapes, hard disks, etc.; and communication devices 909. Communication device 909 allows electronic devices to exchange data via wireless or wired communication with other devices. Although Figure 9 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0091] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 909, or installed from a memory 908, or installed from a ROM 902. When the computer program is executed by a processor 901, it performs the functions defined in the vascular cognitive impairment assessment method of the embodiments of the present invention.

[0092] Figure 9 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0093] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code originally stored on a remote storage medium or a non-transitory machine-readable storage medium and subsequently stored on a local storage medium after being downloaded via a network. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium may also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the vascular cognitive impairment assessment method shown in the above embodiments is implemented.

[0094] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0095] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method of assessing vascular cognitive impairment, characterized by, The method includes: Acquire diffusion tensor imaging data of the target object; Based on the diffusion tensor imaging data, the target diffusion tensor parameters are calculated, and a target diffusion tensor parameter map is generated. The target dispersion tensor parameter map is input into a pre-built deep learning model to obtain the target saliency weight map of the target object; Based on the target saliency weight map and a pre-established mutual information graph set, a structural similarity score is determined between the target saliency weight map and each mutual information graph in the mutual information graph set; wherein, the process includes: determining a first statistical data point of the target saliency weight map; determining a second statistical data point of the target mutual information graph, wherein the target mutual information graph is any mutual information graph in the mutual information graph set; both the first and second statistical data points include: mean and variance; determining the covariance between the target saliency weight map and the target mutual information graph; and determining the structural similarity score between the target saliency weight map and the target mutual information graph based on the first statistical data point, the second statistical data point, and the covariance; the mutual information graph characterizes the degree of statistical association between the diffusion tensor parameter map and the target neuropsychological scale; Based on the structural similarity score, the cognitive impairment risk level represented by the target neuropsychological scale is determined; The mutual information graph is established through the following steps: Obtain a first sample dataset, which includes diffusion tensor parameter graphs of sample objects and multiple preset types of neuropsychological scales; the neuropsychological scales include test data of the sample objects, and there are multiple sample objects; All the aforementioned diffusion tensor parameter graphs are standardized. Based on the diffusion tensor parameter map of all the sample objects after standardization, the same diffusion tensor parameter values ​​under the same voxel are combined to form a diffusion tensor parameter matrix. Based on the neuropsychological scales of all the sample subjects, the scores of the same item on the scale are used to form a score matrix. Based on the diffusion tensor parameter matrix and the score matrix, mutual information is calculated to obtain the mutual information map corresponding to each preset type of neuropsychological scale.

2. The method according to claim 1, characterized in that, The target dispersion tensor parameters include at least one of the following: anisotropic parameter, average diffusivity, axial diffusion coefficient, and radial diffusion coefficient; The target neuropsychological scale includes at least one of the following: MMSE, MoCA, immediate recall, delayed recall, link test-A, and link test-B.

3. The method according to claim 1, characterized in that, The determination of the cognitive impairment risk level represented by the target neuropsychological scale based on the structural similarity score includes: Determine the rating scale corresponding to the target neuropsychological scale; Based on the structural similarity score and the rating level comparison table, the cognitive impairment risk level represented by the target neuropsychological scale is determined.

4. The method according to claim 3, characterized in that, The determination of the rating scale corresponding to the target neuropsychological scale includes: Obtain a second sample dataset, which includes a saliency weight map of the sample objects and multiple preset types of neuropsychological scales; the neuropsychological scales include test data of the sample objects, and there are multiple sample objects; Calculate the structural similarity score between the saliency weight map of each sample object and the neuropsychological scale of each preset category to obtain a set of structural similarity scores; Clustering is performed based on the structural similarity score set to obtain a score level correspondence table for each preset category of the neuropsychological scale.

5. The method according to claim 1, characterized in that, The target diffusion tensor parameters are the optimal diffusion tensor parameters determined when constructing the deep learning model.

6. A device for assessing vascular cognitive impairment, characterized in that, The device includes: The acquisition module is used to acquire diffusion tensor imaging data of the target object; The generation module is used to calculate the target diffusion tensor parameters based on the diffusion tensor imaging data and generate a target diffusion tensor parameter map. The learning module is used to input the target diffusion tensor parameter map into a pre-built deep learning model to obtain the target saliency weight map of the target object; A scoring module is used to determine the structural similarity score between the target saliency weight map and each mutual information graph in the mutual information graph set, based on the target saliency weight map and a pre-established mutual information graph set. This includes: determining a first statistical data point for the target saliency weight map; determining a second statistical data point for the target mutual information graph, where the target mutual information graph is any mutual information graph in the mutual information graph set; both the first and second statistical data points include: mean and variance; determining the covariance between the target saliency weight map and the target mutual information graph; and determining the structural similarity score between the target saliency weight map and the target mutual information graph based on the first statistical data point, the second statistical data point, and the covariance; the mutual information graph represents the degree of statistical association between the diffusion tensor parameter map and the target neuropsychological scale. A determination module is used to determine the level of cognitive impairment risk represented by the target neuropsychological scale based on the structural similarity score; A construction module is used to acquire a first sample dataset, which includes diffusion tensor parameter maps of sample objects and multiple preset types of neuropsychological scales; the neuropsychological scales include test data of the sample objects, and there are multiple sample objects; all the diffusion tensor parameter maps are standardized; based on the standardized diffusion tensor parameter maps of all the sample objects, the same diffusion tensor parameter values ​​under the same voxel are combined into a diffusion tensor parameter matrix; based on the neuropsychological scales of all the sample objects, the scores of the same scale are combined into a score matrix; based on the diffusion tensor parameter matrix and the score matrix, mutual information is calculated to obtain the mutual information map corresponding to each preset type of neuropsychological scale.

7. An electronic device, characterized in that, include: A memory and a processor are communicatively connected, the memory storing computer instructions, and the processor executing the computer instructions to perform the vascular cognitive impairment assessment method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the vascular cognitive impairment assessment method according to any one of claims 1 to 5.

Images