Mild cognitive impairment development process prediction method, device and computer equipment

Abstract

The application relates to a mild cognitive impairment development process prediction method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: acquiring first diffusion tensor imaging images of a mild cognitive impairment patient in multiple follow-ups; processing each first diffusion tensor imaging image, extracting white matter fibers between all hippocampal voxels and other brain voxels; performing regional segmentation on the hippocampus according to the white matter fibers between all hippocampal voxels and other brain voxels, and constructing hippocampal subregions; constructing a first gradient change feature of the white matter fibers of each hippocampal subregion according to the gradient change of the target feature of the white matter fibers of the hippocampal subregion; and predicting the cognitive impairment development process of the measured object according to the basic information, clinical information and first gradient change feature of the white matter fibers of the hippocampal subregion of the patient. The method can improve disease classification accuracy and specificity, and can predict the cognitive impairment development process of the measured object.

Description

TECHNICAL FIELD

[0001] The present application relates to the fields of image processing, artificial intelligence and medical technology, in particular to a cognitive impairment development process prediction method and device, computer equipment, storage medium and computer program product. BACKGROUND

[0002] Alzheimer disease (AD) is a neurodegenerative disease characterized by progressive decline in memory and other cognitive functions, which is irreversible and accounts for 60-80% of all dementia patients. Mild cognitive impairment (MCI) is a transitional stage between normal aging and AD.

[0003] In recent years, with the rapid development of neuroimaging, structural magnetic resonance imaging (MRI) and functional MRI can be used to diagnose mild cognitive impairment. Studies have found that changes in the hippocampus of the brain are closely related to mild cognitive impairment. Previous studies analyzed the hippocampus as a whole or extracted sub-regions and analyzed each sub-region as a whole. However, the brain is not composed of discrete parts, and its function is organized in a gradual transition gradient. One manifestation of this gradient change is that the anterior hippocampus is mainly connected with the frontal lobe and the anterior temporal lobe, and the posterior hippocampus is connected with the posterior parietal lobe and the occipital lobe. The function of the brain is supported by its structural basis, so it is reasonable to assume that the hippocampus-related fiber connections also have a front-back gradient change.

[0004] Compared with other major white matter pathways in the brain, the white matter fiber connections between the hippocampus and other regions of the brain are narrower and more curved, which limits the ability to study the microstructure and connectivity of these pathways, and therefore our understanding of them is relatively limited. The current analysis of the hippocampus as a whole assumes that all regions have consistent white matter fiber connections, which is not consistent with the actual situation of the gradient change of the hippocampus structure, resulting in inaccurate detection results. SUMMARY

[0005] Therefore, it is necessary to provide a mild cognitive impairment development process prediction method, device, computer equipment, computer readable storage medium and computer program product with high detection accuracy.

[0006] A mild cognitive impairment development process prediction method, the method comprising:

[0007] obtaining a first diffusion tensor imaging image of a mild cognitive impairment patient at multiple follow-up times;

[0008] processing each of the first diffusion tensor imaging images to extract white matter fibers between all voxels of the hippocampus and other voxels of the brain;

[0009] segmenting the hippocampus according to the white matter fibers between all voxels of the hippocampus and other voxels of the brain to construct hippocampal subregions;

[0010] constructing first gradient change features of the white matter fibers of each of the hippocampal subregions according to gradient changes of target features of the white matter fibers of the hippocampal subregions;

[0011] predicting a cognitive impairment development process of the subject according to the basic information and the clinical information of the patient and the first gradient change features of the white matter fibers of the hippocampal subregions.

[0012] In one of the embodiments, processing each of the first diffusion tensor imaging image sequences to extract white matter fibers between all voxels of the hippocampus and other voxels of the brain includes:

[0013] preprocessing the first diffusion tensor imaging images;

[0014] analyzing the preprocessed first diffusion tensor imaging images to obtain diffusion eigenvalues and diffusion directions;

[0015] performing fiber tracking from each voxel in the hippocampus to other voxels of the brain according to the diffusion eigenvalues and the diffusion directions to extract the white matter fibers between all voxels of the hippocampus and other voxels of the brain.

[0016] In one of the embodiments, segmenting the hippocampus according to the white matter fibers between all voxels of the hippocampus and other voxels of the brain to construct hippocampal subregions includes:

[0017] constructing a white matter fiber connection matrix between all voxels of the hippocampus and other voxels of the brain according to the white matter fibers between all voxels of the hippocampus and other voxels of the brain;

[0018] interchanging rows and columns of the all-voxel-all-voxel intercorrelation matrix of the hippocampus according to the white matter fiber connection matrix and obtaining the hippocampal subregions by using a segmentation algorithm.

[0019] In one of the embodiments, constructing first gradient change features of the white matter fibers of each of the hippocampal subregions according to gradient changes of target features of the white matter fibers of the hippocampal subregions includes:

[0020] segmenting the white matter fibers of the hippocampal subregions into multiple segments and extracting average features of target features of each segment of the white matter fibers of each of the hippocampal subregions to obtain the first gradient change features of the white matter fibers of the hippocampal subregions.

[0021] In one of the embodiments, the cognitive impairment development process of the subject is predicted according to the basic information, clinical information of the subject and the first gradient change feature of the white matter fiber of the hippocampal subregion.

[0022] The basic information, clinical information of the subject and the first gradient change feature of the white matter fiber of the hippocampal subregion are input into a pre-trained cognitive impairment development process prediction model to predict the cognitive impairment development process of the subject.

[0023] In one of the embodiments, the training method of the cognitive impairment development process prediction model comprises:

[0024] A first training set is obtained, which includes a plurality of second diffusion tensor imaging images of multiple follow-ups of a plurality of patients, corresponding basic information, clinical information of the patients, and actual cognitive impairment development process labeled on the second diffusion tensor imaging images;

[0025] Each of the second diffusion tensor imaging image sequences is processed to extract the white matter fiber between all the voxels of the hippocampus and other voxels of the brain;

[0026] The hippocampus is regionally segmented according to the white matter fiber to construct a hippocampal subregion;

[0027] The second gradient change feature of the white matter fiber of each hippocampal subregion is constructed according to the gradient change of the target feature of the white matter fiber of the hippocampal subregion;

[0028] The basic information, clinical information of the subject and the second gradient change feature of the white matter fiber of the hippocampal subregion are input into the cognitive impairment development process prediction model to be trained to predict the cognitive impairment development process of the subject;

[0029] The cognitive impairment development process prediction model is adjusted according to the difference between the predicted cognitive impairment development process and the labeled actual cognitive impairment development process, and when the training end condition is met, a trained cognitive impairment development process prediction model is obtained.

[0030] In one of the embodiments, the target feature of the white matter fiber of the hippocampal subregion is a feature of the white matter fiber of the hippocampal subregion related to mild cognitive impairment determined based on a classification model training.

[0031] A mild cognitive impairment development process prediction device, the device comprises:

[0032] The acquisition module is configured to acquire a first diffusion tensor imaging image of a plurality of follow-ups of a mild cognitive impairment patient;

[0033] a white matter fiber extraction module configured to process each of the first diffusion tensor imaging images to extract white matter fibers between all voxels of the hippocampus and other voxels of the brain;

[0034] a hippocampal sub-region segmentation module configured to segment the hippocampus into sub-regions according to the white matter fibers between all voxels of the hippocampus and other voxels of the brain;

[0035] a gradient change module configured to construct a first gradient change feature of the white matter fibers of each of the hippocampal sub-regions according to a gradient change of a target feature of the white matter fibers of the hippocampal sub-regions;

[0036] a prediction module configured to predict a cognitive impairment development process of the subject according to the basic information and the clinical information of the patient and the first gradient change feature of the white matter fibers of the hippocampal sub-regions.

[0037] A computer device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the following steps when executing the computer program:

[0038] acquiring first diffusion tensor imaging images of a mild cognitive impairment patient in multiple follow-ups;

[0039] processing each of the first diffusion tensor imaging images to extract white matter fibers between all voxels of the hippocampus and other voxels of the brain;

[0040] segmenting the hippocampus into sub-regions according to the white matter fibers between all voxels of the hippocampus and other voxels of the brain to construct hippocampal sub-regions;

[0041] constructing a first gradient change feature of the white matter fibers of each of the hippocampal sub-regions according to a gradient change of a target feature of the white matter fibers of the hippocampal sub-regions;

[0042] predicting a cognitive impairment development process of the subject according to the basic information and the clinical information of the patient and the first gradient change feature of the white matter fibers of the hippocampal sub-regions.

[0043] A computer readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the following steps:

[0044] acquiring first diffusion tensor imaging images of a mild cognitive impairment patient in multiple follow-ups;

[0045] processing each of the first diffusion tensor imaging images to extract white matter fibers between all voxels of the hippocampus and other voxels of the brain;

[0046] segmenting the hippocampus into sub-regions according to the white matter fibers between all voxels of the hippocampus and other voxels of the brain to construct hippocampal sub-regions;

[0047] constructing a first gradient change feature of the white matter fiber of each of the hippocampal subregions according to the gradient change of the target feature of the white matter fiber of the hippocampal subregion;

[0048] predicting the cognitive impairment development process of the measured object according to the basic information, the clinical information of the patient and the first gradient change feature of the white matter fiber of the hippocampal subregion.

[0049] A computer program product comprises a computer program which, when executed by a processor, implements the following steps:

[0050] obtaining a first diffusion tensor imaging image of a mild cognitive impairment patient in multiple follow-ups;

[0051] processing each of the first diffusion tensor imaging images to extract the white matter fiber between all voxels of the hippocampus and other voxels of the brain;

[0052] segmenting the hippocampus into regions according to the white matter fiber between all voxels of the hippocampus and other voxels of the brain to construct hippocampal subregions;

[0053] constructing a first gradient change feature of the white matter fiber of each of the hippocampal subregions according to the gradient change of the target feature of the white matter fiber of the hippocampal subregion;

[0054] predicting the cognitive impairment development process of the measured object according to the basic information, the clinical information of the patient and the first gradient change feature of the white matter fiber of the hippocampal subregion.

[0055] The cognitive impairment development process prediction method, device, computer equipment, storage medium and computer program product can calculate the first gradient change feature of the white matter fiber of the hippocampal subregion according to the diffusion tensor imaging image of the measured object in multiple follow-ups, and predict according to the basic information, the clinical information and the first gradient change feature of the white matter fiber of the hippocampal subregion. Since the first gradient change feature of the white matter fiber of the hippocampal subregion is used, the disease classification accuracy and specificity can be improved, the cognitive impairment development process of the measured object can be predicted, and the doctor and the patient can be helped to intervene in advance to slow down the development of the disease. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 It is an application environment diagram of the mild cognitive impairment development process prediction method in one embodiment;

[0057] Figure 2 It is a flowchart of the mild cognitive impairment development process prediction method in one embodiment;

[0058] Figure 3 It is a schematic diagram of the hippocampal structure in one embodiment;

[0059] Figure 4 An explanatory diagram of the cognitive impairment development process of a mild cognitive impairment patient predicted in an embodiment;

[0060] Figure 5 A structural block diagram of a mild cognitive impairment development process prediction device in an embodiment;

[0061] Figure 6 An internal structural diagram of a computer device in an embodiment. DETAILED DESCRIPTION

[0062] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application is further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0063] The mild cognitive impairment development process prediction method provided by the embodiments of the present application can be applied in an application environment as shown in Figure 1 . The detection terminal 102 communicates with the processing terminal 104. The detection terminal can be provided with a data storage system, such as detection images collected by the detection terminal. The detection terminal 102 sends the detection images to the processing terminal 104 for processing. The detection terminal can be an MRI (magnetic resonance imaging) device. The detection terminal collects brain image data of the measured object, such as a brain MRI image, and sends it to the processing terminal 104. The processing terminal acquires first diffusion tensor imaging images of the mild cognitive impairment patient in multiple follow-ups. Each of the first diffusion tensor imaging images is processed to extract white matter fibers between all hippocampal voxels and other brain voxels. The hippocampus is regionally segmented according to the white matter fibers between the all hippocampal voxels and the other brain voxels, and the hippocampal subregion is constructed. The first gradient change feature of the white matter fibers of each of the hippocampal subregions is constructed according to the gradient change of the target feature of the white matter fibers of the hippocampal subregion. The cognitive impairment development process of the measured object is predicted according to the basic information and clinical information of the patient and the first gradient change feature of the white matter fibers of the hippocampal subregion.

[0064] In an embodiment, as shown in Figure 2 , a mild cognitive impairment development process prediction method is provided. Taking the processing terminal in Figure 1 as an example, the method includes the following steps:

[0065] Step 202, acquiring first diffusion tensor imaging images of a mild cognitive impairment patient in multiple follow-ups.

[0066] Diffusion tensor imaging (DTI) is a new method of depicting brain structure that is an extension and refinement of diffusion weighted imaging (DWI) and can effectively observe and track white matter fibers. For example, if one says that MRI traces hydrogen atoms in water molecules, then DTI maps the direction of water molecule movement. DTI can reveal how brain tumors affect neural cell connections, guiding medical personnel in brain surgery. It can also reveal subtle abnormal changes associated with stroke, multiple sclerosis, schizophrenia, and dyslexia.

[0067] Among them, the patient diagnosed as mild cognitive development disorder is taken as the measured object. The multiple follow-up diffusion tensor imaging of the measured object is used for prediction. The multiple follow-up diffusion tensor imaging, i.e. the diffusion tensor imaging of each follow-up, predicts the development process of mild cognitive impairment by mining the changes of multiple follow-up diffusion tensor imaging.

[0068] Diffusion tensor imaging has high resolution, in addition to higher spatial resolution (sub-millimeter level, up to 760 μm anisotropy, even 300 μm), it also has multiple diffusion intensity ("multi-shell") and multiple diffusion direction (for example, 94 cases of 391 direction diffusion tensor data).

[0069] In step 204, each of the first diffusion tensor images is processed to extract white matter fibers between all voxels of the hippocampus and other voxels of the brain.

[0070] The white matter of the brain is composed of myelinated fiber bundles, called tracts or fiber pathways. These fibers form a complex three-dimensional structure in the hemispheres or brain stem. DTI diffusion tensor imaging is a new method of depicting brain structure that is a special form of magnetic resonance imaging (MRI). It can display the orientation of neural fiber bundles in three dimensions! It can non-invasively display the white matter and white matter tracts in vivo, and achieve fine imaging of human neural fiber bundles.

[0071] Specifically, the hippocampus of the brain is a brain area that helps humans process long-term learning and memory of sound, light, taste, and other events. In medicine, the "hippocampus" is an inner fold area of the cerebral cortex, forming an arc-shaped protrusion at the bottom of the "lateral ventricle" around the "choroid fissure", which is composed of two fan-shaped parts, and sometimes the two are referred to as hippocampal structures. The hippocampal structure in one embodiment is shown as the area 301 in the brain image shown in Figure 3

[0072] ​The first diffusion tensor imaging image can be pre-processed, and the pre-processing includes registration processing and eddy current correction processing. After the pre-processing, the pre-processed first diffusion tensor imaging image is analyzed to extract diffusion eigenvalues and diffusion directions. According to the diffusion eigenvalues and the diffusion directions, the whole brain fibers are tracked. Specifically, probabilistic tractography is performed from each voxel in the hippocampus to all vertices of the gray matter / white matter interface of the brain to evaluate the connectivity of each brain voxel of the hippocampus, and a model considering that each voxel can have multiple fiber directions is used.

[0073] In step 206, the hippocampus is regionally segmented according to the white matter fibers between all voxels of the hippocampus and other voxels of the brain, and hippocampal subregions are constructed.

[0074] By classifying the brain region voxels according to the cortical regions of their highest connection probability, the hippocampus can be divided into different connection-defined regions, which are assumed to correspond to different hippocampal subregions, and these subregions have gradient changes. In this way, a white matter fiber connection matrix between all voxels of the hippocampus and other voxels of the brain is obtained, and the hippocampus is segmented according to the white matter fiber connection matrix to construct hippocampal subregions.

[0075] In step 208, a first gradient change feature of the white matter fibers of each hippocampal subregion is constructed according to the gradient changes of the target features of the white matter fibers of the hippocampal subregion.

[0076] Specifically, the hippocampal subregions have gradient changes. The target features are features of the white matter fibers of the hippocampal subregions related to mild cognitive impairment determined based on training of a classification model. According to the gradient changes of these features, a first gradient change feature of the white matter fibers of each hippocampal subregion is constructed.

[0077] In step 210, the cognitive impairment progression of the measured object is predicted according to the basic information, clinical information, and the first gradient change feature of the white matter fibers of the hippocampal subregion of the patient.

[0078] The basic information includes age, gender, and education level, etc.

[0079] The clinical information includes clinical test information, including but not limited to MRS indicators, ADAS-Cog13, FAQ, MMSE, etc.

[0080] In this embodiment, the first gradient change characteristics of the white matter fibers in the hippocampal subregion of the mild cognitive impairment patient in multiple follow-ups are used to mine the white matter micro-change process of the mild cognitive impairment patient, and then the cognitive impairment development process of the measured object is predicted. For example, in the first follow-up, the hippocampal subregion A is responsible for working memory, and the hippocampal subregion B is responsible for episodic memory. However, in the second follow-up, the white matter fibers from the hippocampal subregion A to the hippocampal subregion B are interrupted, and the disease affects the memory function. Therefore, it can be predicted that the patient's disease course has changed between the first follow-up and the second follow-up. By using the changes in multiple follow-ups, the cognitive impairment development process of the measured object is predicted.

[0081] The cognitive impairment development process includes the probabilities of developing into different degrees of cognitive impairment at each time point in a future period of time. It should be noted that this cognitive impairment development process is not a diagnosis, but a probability prediction, which can help patients understand the disease development process and help doctors make medical interventions as an auxiliary means.

[0082] The cognitive impairment includes three different degrees of cognitive impairment, namely mild cognitive impairment, moderate cognitive impairment, and severe cognitive impairment. An embodiment of the mild cognitive impairment development process predicted for the measured object is shown in FIG. 1, which respectively shows the probability of developing into mild cognitive impairment at each time point in the future, the probability of developing into moderate cognitive impairment at each time point in the future, and the probability of developing into severe cognitive impairment at each time point in the future. Figure 4

[0083] The cognitive impairment development prediction method described above calculates the first gradient change characteristics of the white matter fibers in the hippocampal subregion by using the diffusion tensor imaging images of the measured object in multiple follow-ups, and predicts according to the basic information, clinical information, and first gradient change characteristics of the white matter fibers in the hippocampal subregion. Since more abundant information of the changes in the first gradient change characteristics of the white matter fibers in the hippocampal subregion is used, the disease classification accuracy and specificity can be improved, the cognitive impairment development process of the measured object can be predicted, and thus the doctor and the patient can intervene in advance to slow down the disease development.

[0084] In another embodiment, the white matter fibers between all voxels of the hippocampus and other voxels of the brain are extracted by processing each of the first diffusion tensor imaging image sequences, including: preprocessing the first diffusion tensor imaging image; analyzing the preprocessed first diffusion tensor imaging image to obtain diffusion eigenvalues and diffusion directions; and performing whole-brain fiber tracking from each voxel in the hippocampus to other voxels of the brain according to the diffusion eigenvalues and diffusion directions to extract the white matter fibers between all voxels of the hippocampus and other voxels of the brain.

[0085] ​In gradient coils, diffusion weighted images are prone to distortions due to susceptibility induced distortions and eddy currents. EDDY provides a method to correct for eddy current distortions and subject motion. It can also use a field map (estimated by TOPUP) to account for and correct for susceptibility induced distortions. It also allows for detection of outliers, data points corrupted by motion induced signal loss. Finally, it allows for layer based outlier detection and replacement, and layer to volume registration.

[0086] One optional pre-processing step is to run mri_degibbs, which removes Gibbs ring artifacts from the data. These artifacts look like ripples in a pond, and are most visible in images with a b-value of 0. Inspect the data before and after pre-processing with mrview to determine if the pre-processing step made the data better, worse, or had no effect. If no Gibbs artifacts are visible in the data, omit this step.

[0087] Where the pre-processed first diffusion tensor imaging image is analyzed, a diffusion tensor model can be used. Specifically, a diffusion tensor model is fit to the data that has been pre-processed and eddy current corrected at each voxel.

[0088] For diffusion weighted images, the diffusion characteristics of each voxel in the brain can be modeled as a combination of eigenvectors and eigenvalues: the eigenvectors represent the direction of diffusion, and the eigenvalues represent the magnitude of diffusion. Then, a number of different equations can be used to calculate different properties of the diffusion in that voxel. The most commonly used equation for this diffusion tensor imaging is called fractional anisotropy, or FA. Fractional anisotropy is a weighted sum of the eigenvalues in each voxel. A higher FA value indicates that diffusion is greater in one direction, and a lower FA value indicates either that there is little diffusion, or that the diffusion is unconstrained.

[0089] In addition to FA, there are a few other measures: mean diffusivity (MD): the average of the eigenvalues, calculated by adding the eigenvalues and dividing by 3. This is useful for identifying brain pathologies such as edema; axial diffusivity (AD): the value of the largest eigenvalue; radial diffusivity (RD): the average of the two smallest eigenvalues. This is commonly used to analyze large fiber tracts that are oriented the same, such as the corpus callosum.

[0090] To determine the direction of diffusion in each voxel, a basis function is established based on the data from the subject: by extracting the diffusion signal from representative gray matter, white matter, and cerebrospinal fluid voxels, a model is established to estimate the signal at different directions and different b-values, and the basis function is adjusted until it matches the observed data.

[0091] The FODs are created using the generated basis functions. These are estimates of the diffusion quantities in each of the three orthogonal directions. These tensors are similar to the tensors used in traditional diffusion tensor studies, but FODs allow for the estimation of multiple crossing fibers within a single voxel and can decompose the diffusion signal into multiple directions. Finally, to allow for comparisons across subjects, the FODs need to be registered to a standard space.

[0092] Correctly estimating the FODs for each tissue type lays the foundation for fiber tracking. The next key step is to determine the boundary between gray matter and white matter, which will be used as the starting point for fiber tracking.

[0093] To create the boundary between gray matter and white matter, the gray-white matter boundary segmented using freesurfer based on the Tl image can be used to restrict the location of seed placement, but this boundary needs to be transformed from the Tl structural space to the diffusion space by registration. Care must be taken to ensure that the Tl structural boundary is aligned with the boundary of the diffusion weighted images; even small differences in the location of the two scans can affect the results of the fiber tracking.

[0094] Seeds are placed at random locations along the boundary between gray matter and white matter. Fibers will be tracked from the seeds, following a path until it terminates in another region. In this approach, the region of seeds along the gray matter and white matter boundary will produce multiple fibers, with the direction of the fibers most likely along the FODs that dominate the location. The default method is to use the algorithm from iFOD2, which is a probabilistic tracking method.

[0095] In another embodiment, the hippocampus is regionally segmented according to white matter fibers between all voxels of the hippocampus and other voxels of the brain, comprising: constructing a white matter fiber connection matrix between all voxels of the hippocampus and other voxels of the brain according to white matter fibers between all voxels of the hippocampus and other voxels of the brain; and obtaining hippocampal subregions according to the white matter fiber connection matrix by exchanging rows and columns of a hippocampal voxel-hippocampal voxel intercorrelation matrix and using a segmentation algorithm.

[0096] By classifying the voxels of a brain region according to their highest connection probability cortical region, the hippocampus can be divided into different connection-defined regions, which are assumed to correspond to different partitions of the hippocampus, and these partitions have the characteristic of gradient variation.

[0097] For each subject, probabilistic tractography is performed from each voxel in the hippocampus to all vertices of the gray / white matter interface to assess the connectivity of each brain voxel of the hippocampus, using a model that takes into account the possibility of multiple fiber directions for each voxel. In this way, a white matter fiber connection matrix between all voxels of the hippocampus and other voxels of the brain is obtained. For these extracted hippocampal subregion-related fiber tracts, they are all converted to a standard space by registration, serving as fiber tract templates for subsequent analysis.

[0098] The rows or intercorrelation matrix are permuted to automatically cluster the definition of different clusters using k-means partitioning. The goal of clustering the intercorrelation matrix is to group together regions that share the same connections with other parts of the brain.

[0099] The rows and columns of the intercorrelation matrix of all voxels in the hippocampus - hippocampus are permuted and the k-means partitioning algorithm is used to automatically obtain different partitions of the hippocampus.

[0100] To increase the chance of obtaining contiguous (non-scattered) partitions, distance constraints are added so that each hippocampal partition is limited to consist of spatially contiguous vertices.

[0101] The spatial relationship between each hippocampal voxel of the intercorrelation matrix of all voxels in the hippocampus - hippocampus is applied to diffusion map embedding, a nonlinear dimensionality reduction technique used to identify spatial axes of varying connectivity across different regions. In contrast to block-based or border mapping techniques, diffusion map embedding does not define discrete networks but extracts continuous gradients that depict rapid transitions. The embedding of the diffusion map results in a principal gradient that explains the main variability of the spatial distribution. Diffusion map embedding uses a nonlinear transformation provided by the BrainSpace toolbox.

[0102] In another embodiment, a first gradient change feature of white matter fibers of each of the hippocampal subregions is constructed according to the gradient change of a target feature of the white matter fibers of the hippocampal subregion, comprising: segmenting the white matter fibers of the hippocampal subregion into multiple segments, and extracting an average feature of the target feature of each segment of the white matter fibers of the hippocampal subregion to obtain the first gradient change feature of the white matter fibers of the hippocampal subregion.

[0103] Compared with other major white matter pathways in the brain, the white matter fiber connections between the hippocampus and other regions of the brain are narrower and more curved, thus limiting the ability to study the microstructure and connectivity of these pathways, and thus relatively less is known about them. To extract fibers related to the hippocampus, especially its subregions, higher resolution of diffusion tensor images is required. Therefore, the present scheme intends to extract hippocampal subregion-related fibers based on high-resolution diffusion tensor images. These data have higher spatial resolution, multiple diffusion intensities ("multi-shell"), and multiple diffusion directions (thus greater "angular resolution"). In addition to the improvement of diffusion image acquisition, fiber density imaging (TDI) and "super-resolution" analysis, all of which can provide better anatomical detail description than standard atlases based on eigenvalues, thus more clearly depicting white matter fibers.

[0104] For these extracted hippocampal subregion-related fiber tracts, each is converted to standard space by registration as a fiber tract template for subsequent analysis.

[0105] Super-resolution TDI technology can not only play a role in the direct visualization of brain substructures, but also serve as an important complementary tool to fiber tracing studies (e.g., helping to place seed / target regions due to fine anatomical details and high spatial resolution), intersubjective spatial normalization, assessment of changes in brain connectivity (defined by whole-brain fiber tracing results), and voxel-based analysis.

[0106] The hippocampus has six main associated fiber tracts: 1) Inferior longitudinal fasciculus-related fibers, composed of short and long conjugate fibers, originating from the middle and inferior temporal gyrus and parahippocampal gyrus, hippocampus, amygdala and temporal pole, terminating in the extrastriate region; 2) Spinal limbic pathway: this pathway connects the brainstem, hippocampus and subcortical nuclei: thalamus, putamen and caudate nucleus; 3) Cingulate fasciculus-related fibers: fibers from the hippocampus pass through the cingulate fasciculus, mainly reaching the medial frontal lobe and parietal lobe; 4) Anterior commissure-related tract: the anterior commissure is revealed to be a narrow and dense pathway between the anterior hippocampus and amygdala; 5) Tapetum-related fiber tract; 6) Fornix-related tract;

[0107] By combining super-resolution TDI technology, the relevant regions of interest (ROIs) needed for extracting hippocampal fiber tracts can be defined more precisely. After whole-brain fiber tracing, the extraction of each hippocampal fiber tract is completed using tckedit in mrtrix3.

[0108] For MCI or normal subjects in the MCI-related database, the average values ​​of FA, AD, MD, and RD of the relevant fiber bundles in the hippocampal subregion in the aforementioned standard space can be obtained as one of the target characteristics of the subjects.

[0109] During extraction, the white matter fibers in the hippocampal subregion are divided into multiple segments, and the average value of the target features (FA, AD, MD, and RD) of each segment of the white matter fibers in the hippocampal subregion is extracted to obtain the first gradient change features of the white matter fibers in the hippocampal subregion.

[0110] In one embodiment, predicting the cognitive impairment development process of the tested subject based on the patient's basic information, clinical information, and the first gradient change characteristics of white matter fibers in the hippocampus subregion includes: inputting the patient's basic information, clinical information, and the first gradient change characteristics of white matter fibers in the hippocampus subregion into a pre-trained cognitive impairment development process prediction model to predict the cognitive impairment development process of the tested subject.

[0111] Specifically, the predictive model for the development of cognitive impairment can be trained using the LASSO regularized Cox regression model analysis mode. This model is used to study the relationship between predictor variables and survival time. The approach involves first training the model, then constructing the model using the optimal parameters, and finally evaluating the prediction accuracy (C-index value) to determine the quality of the prediction.

[0112] The LASSO regularized Cox regression model is provided by the glmnet toolbox.

[0113] Cox proportional hazard model is often used to study the relationship between the predictive variables and the survival time. Cox proportional hazard model can analyze the influence of multiple factors on survival time, and allows the existence of censored data, which is the most important multivariate analysis method in survival analysis.

[0114] The main purpose of survival analysis is to study the relationship between covariates (independent variables) and observations, i.e. survival function, when affected by covariates, the traditional method is to consider regression analysis, i.e. the influence of each covariate pair. Because survival data contains censored data, it is difficult to solve the above problems by general regression analysis. An important part of survival analysis is to explore the risk factors that affect survival time or survival rate. These risk factors can affect survival rate by affecting the mortality risk (i.e. risk rate) at each time. The risk rate function of different characteristics of the population is different at different times, and the risk rate function is usually expressed as the product of the baseline risk rate function and the corresponding covariate function.

[0115] Therefore, Cox regression model is different from general regression analysis, and the influence of covariates on survival time is reflected by the ratio of risk function and baseline risk function. The risk function and baseline risk function are unknown. After completing parameter estimation, the baseline risk function and risk function can be estimated, and the survival rate at each time can be calculated.

[0116] The basic form of Cox regression model is:

[0117]

[0118] Where β1, β2, …, β m are the partial regression coefficients of independent variables.

[0119] The meaning of partial regression coefficient β j in the model is that when other covariates are fixed, the change amount of log hazard ratio of covariate x j (j = 1, 2, …, m) changes one unit. It can be seen that the regression coefficient β j can also be explained as the log value of the relative risk obtained by changing one unit of independent variable x j when other independent variables are fixed. This explanation is more commonly used in the analysis of risk factors of survival time.

[0120] Specifically, the training method of the cognitive impairment development process prediction model comprises:

[0121] Obtain a first training set; the first training set includes multiple second diffusion tensor imaging images of multiple patients in multiple follow-ups, corresponding patient basic information, clinical information, and actual cognitive impairment development process labeled on the second diffusion tensor imaging images; process each of the second diffusion tensor imaging image sequences, extract white matter fibers between all hippocampus voxels and other brain voxels; regionally segment the hippocampus according to the white matter fibers, and construct hippocampus subregions; construct second gradient change characteristics of white matter fibers of each of the hippocampus subregions according to gradient changes of target characteristics of the white matter fibers of the hippocampus subregions; input the basic information, clinical information, and the second gradient change characteristics of the white matter fibers of the hippocampus subregions of the measured object into a cognitive impairment development process prediction model to be trained, and predict the cognitive impairment development process of the measured object; adjust the cognitive impairment development process prediction model according to differences between the predicted cognitive impairment development process and the labeled actual cognitive impairment development process, and obtain a trained cognitive impairment development process prediction model when a training end condition is met.

[0122] The first training set can use an MCI database disclosed by ADNI, and contains baseline and multiple tracking data; contains T1w, BOLD, and other image data, and cognitive, blood, and cerebrospinal fluid biochemical index data of the subjects. The second diffusion tensor imaging image labels the actual cognitive impairment development process. The steps of processing the second diffusion tensor imaging image, extracting white matter fibers between all hippocampus voxels and other brain voxels, regionally segmenting the hippocampus according to the white matter fibers, and constructing hippocampus subregions are the same as the processing of the prediction process, and will not be repeated here.

[0123] The subjects are divided into a training set and a test set. The cognitive impairment development process prediction model is established by using Cox regression. The cognitive impairment development process prediction model is trained based on the training set, and the prognosis performance is evaluated based on the test set. Based on the training data, the regularization parameter of the LASSO model is optimized by using 10-fold cross-validation. The time-to-event prognosis model estimates the overall risk score of an individual progressing to AD dementia. People with a high risk score develop AD dementia earlier than people with a low risk score. According to the risk score, the probability value of developing AD dementia per year can be estimated, given the baseline risk (developing AD dementia) function, which is estimated based on the training set. In combination with age, gender, education, MRS indicators, ADAS-Cog13, FAQ, MMSE, and other clinical variables, the cognitive impairment development process prediction model is trained.

[0124] During the training process, the cognitive impairment development process prediction model is trained by continuously adjusting the difference between the predicted cognitive impairment development process and the actually labeled cognitive impairment development process. During the training, more rich information of functional connectivity gradient changes is used, so that the disease classification accuracy and specificity can be improved, so that the prediction model can predict the mild cognitive impairment development trajectory and give specific data and charts for display.

[0125] In another embodiment, the target features of the hippocampal subregion white matter fibers are features of the hippocampal subregion white matter tracts determined based on the classification model training to be related to mild cognitive impairment.

[0126] In the classification model training stage, sparse feature selection is used to select part of the features into the SVM classifier model. The main purpose of the SVM classifier is to classify the subjects. The relationship between the subjects and the features is that each subject has multiple features. The first part of feature selection is to avoid too many features entering the SVM, which affects the classification accuracy.

[0127] The SVM also assigns weights to the input features during classification, such as assigning a large weight to features that contribute greatly to correct classification, and vice versa. Then, according to the weight of each feature, the features with small weights are removed, and the features of the hippocampal subregion white matter tracts related to mild cognitive impairment are obtained. The target features include but are not limited to the FA, AD, MD and RD of the white matter fibers in the hippocampal subregion.

[0128] It should be understood that although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but can be alternately or alternately executed with at least part of other steps or steps or stages in other steps.

[0129] Based on the same inventive concept, the embodiments of the present application also provide a mild cognitive impairment development process prediction device for implementing the above-mentioned mild cognitive impairment development process prediction method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, and therefore the specific limitations in one or more mild cognitive impairment development process prediction device embodiments provided below can refer to the limitations of the mild cognitive impairment development process prediction method described above, which will not be repeated here.

[0130] In one embodiment, as shown in Figure 5 a mild cognitive impairment development process prediction device is provided, comprising:

[0131] The acquisition module 502 is configured to acquire first diffusion tensor imaging images of a mild cognitive impairment patient in multiple follow-ups.

[0132] The white matter fiber extraction module 504 is configured to process each of the first diffusion tensor imaging images to extract white matter fibers between all hippocampal voxels and other brain voxels.

[0133] The hippocampal sub-region segmentation module 506 is configured to segment the hippocampus into regions according to the white matter fibers between all hippocampal voxels and other brain voxels, and construct hippocampal sub-regions.

[0134] The gradient change module 508 is configured to construct a first gradient change feature of the white matter fibers of each of the hippocampal sub-regions according to the gradient change of the target feature of the white matter fibers of the hippocampal sub-regions.

[0135] The prediction module 510 is configured to predict the cognitive impairment development process of the measured object according to the basic information, clinical information and the first gradient change feature of the white matter fibers of the hippocampal sub-regions.

[0136] The above cognitive impairment development process prediction device calculates the first gradient change feature of the white matter fibers of the hippocampal sub-regions through the diffusion tensor imaging images of the measured object in multiple follow-ups, and predicts according to the basic information, clinical information and the first gradient change feature of the white matter fibers of the hippocampal sub-regions. Since more abundant information about the change of the first gradient change feature of the white matter fibers of the hippocampal sub-regions is utilized, the disease classification accuracy and specificity can be improved, and the cognitive impairment development process of the measured object can be predicted, thereby helping doctors and patients to intervene in advance and slow down the development of the disease.

[0137] In another embodiment, the white matter fiber extraction module is configured to preprocess the first diffusion tensor imaging image; analyze the preprocessed first diffusion tensor imaging image to obtain diffusion eigenvalues and diffusion directions; and trace fibers from each voxel in the hippocampus to other voxels in the brain based on the diffusion eigenvalues and diffusion directions to extract white matter fibers between all voxels in the hippocampus and other voxels in the brain.

[0138] In another embodiment, the hippocampal subfield segmentation module is configured to construct a white matter fiber connection matrix between all voxels in the hippocampus and other voxels in the brain based on the white matter fibers between all voxels in the hippocampus and other voxels in the brain; and exchange rows and columns of the all-voxel-all-voxel intercorrelation matrix of the hippocampus based on the white matter fiber connection matrix, and obtain hippocampal subfields using a segmentation algorithm.

[0139] In another embodiment, the gradient change module is configured to segment the white matter fibers of the hippocampal subfields into multiple segments, and extract average features of target features of each segment of the white matter fibers of the hippocampal subfields to obtain first gradient change features of the white matter fibers of the hippocampal subfields.

[0140] In another embodiment, the prediction module is configured to input the basic information, clinical information, and the first gradient change features of the white matter fibers of the hippocampal subfields of the patient into a pre-trained cognitive impairment progression prediction model to predict the cognitive impairment progression of the subject.

[0141] In another embodiment, the training module is further configured to obtain a first training set; the first training set includes a plurality of second diffusion tensor imaging images of multiple follow-ups of a plurality of patients, corresponding basic information, clinical information of the patients, and actual cognitive impairment progression labeled on the second diffusion tensor imaging images; process each of the second diffusion tensor imaging image sequences to extract white matter fibers between all voxels in the hippocampus and other voxels in the brain; perform regional segmentation of the hippocampus based on the white matter fibers to construct hippocampal subfields; construct second gradient change features of the white matter fibers of each of the hippocampal subfields based on gradient changes of target features of the white matter fibers of the hippocampal subfields; input the basic information, clinical information, and the second gradient change features of the white matter fibers of the hippocampal subfields of the subject into a cognitive impairment progression prediction model to be trained to predict the cognitive impairment progression of the subject; adjust the cognitive impairment progression prediction model based on differences between the predicted cognitive impairment progression and the labeled actual cognitive impairment progression, and obtain the trained cognitive impairment progression prediction model when a training end condition is met.

[0142] In another embodiment, the target features of the white matter fibers of the hippocampal subfields are features of the white matter fibers of the hippocampal subfields related to mild cognitive impairment determined based on training of a classification model.

[0143] Each module in the aforementioned device for predicting the development of mild cognitive impairment can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0144] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores classification models. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements a method for predicting the developmental process of mild cognitive impairment.

[0145] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0146] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the mild cognitive impairment development process prediction method of the above embodiments.

[0147] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the mild cognitive impairment development process prediction method of the above embodiments.

[0148] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the mild cognitive impairment development process prediction method of the above embodiments.

[0149] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties.

[0150] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing related hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiments. Any reference to memory, database or other medium used in the embodiments provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided by the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided by the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0151] The technical features of the above embodiments can be combined in any way. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present disclosure.

[0152] The above-described embodiments are merely illustrative of several embodiments of the present application, and the description is relatively specific and detailed, but should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. A device for predicting the development of mild cognitive impairment, characterized in that, The device includes: The acquisition module is used to acquire the first diffusion tensor imaging images of patients with mild cognitive impairment during multiple follow-up visits; The white matter fiber extraction module is used to process each of the first diffusion tensor imaging images to extract white matter fibers between all voxels in the hippocampus and other voxels in the brain. The hippocampal subregion segmentation module is used to segment the hippocampus into regions based on the white matter fibers between all voxels of the hippocampus and other voxels of the brain, and to construct hippocampal subregions. The gradient change module is used to divide the white matter fibers in the hippocampus subregion into multiple segments, extract the average feature of the target feature of each segment of the white matter fibers in the hippocampus subregion, and obtain the first gradient change feature of the white matter fibers in the hippocampus subregion. The prediction module is used to predict the cognitive impairment development process of the patient based on the patient's basic information, clinical information, and the first gradient change characteristics of white matter fibers in the hippocampus subregion. The cognitive impairment development process includes the probability of developing into different degrees of cognitive impairment at various time points in the future.

2. The apparatus according to claim 1, characterized in that, The white matter fiber extraction module is used to preprocess the first diffusion tensor imaging image; analyze the preprocessed first diffusion tensor imaging image to obtain diffusion characteristic values ​​and diffusion direction; and, based on the diffusion characteristic values ​​and diffusion direction, perform fiber tracing from each voxel in the hippocampus to other voxels in the brain to extract white matter fibers between all voxels in the hippocampus and other voxels in the brain.

3. The apparatus according to claim 1, characterized in that, The hippocampal subregion segmentation module is used to construct a white matter fiber connection matrix between all hippocampal voxels and other brain voxels based on the white matter fibers between all hippocampal voxels and other brain voxels; based on the white matter fiber connection matrix, the rows and columns of the cross-correlation matrix between all hippocampal voxels are swapped, and the hippocampal subregions are obtained using a segmentation algorithm.

4. The apparatus according to claim 1, characterized in that, The prediction module is used to input the patient's basic information, clinical information, and the first gradient change characteristics of white matter fibers in the hippocampus subregion into a pre-trained cognitive impairment development prediction model to predict the patient's cognitive impairment development process.

5. The apparatus according to claim 4, characterized in that, The device further includes a training module for acquiring a first training set. The first training set includes multiple follow-up second diffusion tensor imaging images of various patients, corresponding to the patients' basic information, clinical information, and the actual cognitive impairment progression annotated on the second diffusion tensor imaging images. Each second diffusion tensor imaging image is processed to extract white matter fibers between all voxels in the hippocampus and other voxels in the brain. The hippocampus is segmented based on the white matter fibers to construct hippocampal subregions. Second gradient change features of white matter fibers in each hippocampal subregion are constructed based on the gradient changes of target features of white matter fibers in the hippocampal subregions. The patient's basic information, clinical information, and the second gradient change features of white matter fibers in the hippocampal subregions are input into a cognitive impairment progression prediction model to be trained to predict the patient's cognitive impairment progression. The cognitive impairment progression prediction model is adjusted based on the difference between the predicted cognitive impairment progression and the annotated actual cognitive impairment progression. When the training termination condition is met, a trained cognitive impairment progression prediction model is obtained.

6. The apparatus according to claim 1 or 5, characterized in that, The target features of white matter fibers in the hippocampus subregion are the features of white matter fibers in the hippocampus subregion associated with mild cognitive impairment, determined based on classification model training.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it performs the operations performed by the apparatus according to any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it performs the operations performed by the apparatus according to any one of claims 1 to 6.

Images