Depression subtype classification system based on cross-species multi-modal brain image

Abstract

The invention discloses a depression subtype classification system based on a cross-species multi-mode brain image. The method comprises the following steps: firstly, obtaining cross-species multi-mode brain image data and depression brain image data and carrying out preprocessing; constructing a multi-modal mapping process between the human brain and the chimpanzee brain to obtain a homologous brain map of the human brain and the chimpanzee brain; the method comprises the following steps: constructing connection characteristics of a human brain and a chimpanzee brain by using whole-brain-based probability fiber tracking, and extracting specific connection characteristics of the human brain through cross-species comparison; constructing a depression typing model based on the specific connection characteristics of the human beings; and verifying the typing result based on the functional image data index and the brain function connection. According to the invention, from the perspective of species evolution, specific connection characteristics of human beings are constructed through cross-species comparison, depression subtype typing is realized, and a new perspective and a new method are provided for understanding depression subtypes and biological basis thereof.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of neuroimaging technology, and in particular to a method for classifying depression using cross-species brain imaging features, and in particular to a depression subtype classification system based on cross-species multimodal brain imaging. Background Art

[0002] Major Depressive Disorder (MDD) is a common debilitating mental illness caused by a combination of factors, and its etiology and pathophysiology are still unclear. Usually, the diagnosis of MDD is guided by subjective behavioral symptoms. Currently, doctors are influenced by their professional experience in the selection of scales and diagnostic criteria, and there is no unified standard in the industry. In addition, some questions in the scale involve personal privacy, and some patients may conceal the actual situation when answering questions, which leads to diagnostic deviations by doctors. Influenced by these factors, it is difficult to determine the type of depression, and there are no objective indicators with significant heterogeneity in judging phenotypic manifestations, etiology, and longitudinal trajectories.

[0003] Studies have shown that patients with depression have abnormalities in the occipital lobe and are accompanied by visual cortex dysfunction. The occipital lobe is related to the perception and processing of visual information and complex visual perception processes, and is also strongly connected to the limbic system (especially the hippocampus), parietal lobe, and temporal lobe. At the same time, a large number of studies have shown that MDD is also associated with histological changes in the higher cortex of the brain, such as the frontal lobe, parietal lobe, temporal lobe, and default mode network. The higher-level connection cortex is a structural and functional system in the cerebral cortex that plays a connecting and comprehensive role. It is the part of the cerebral cortex that performs higher-level cognitive functions. It controls and organizes human speech and thinking, plans human purposeful behaviors, and regulates conscious activities. The changes in these higher-level areas are particularly obvious in evolution, so it is particularly important to analyze the pathogenesis and classification of depression from an evolutionary perspective.

[0004] In the study of the potential mechanism of MDD, there are many difficulties in cross-species classification of depression. First, there is a lack of reliable homologous brain maps between species for cross-species comparison; second, the accuracy of the cross-species brain map model is an important factor affecting the subsequent classification; finally, domestic and foreign researchers have found that MDD may include multiple subtypes and subgroups, but there is no effective solution for how to subdivide these subtypes and subgroups from a cross-species perspective. Summary of the invention

[0005] In order to solve the above technical problems, the present invention aims to propose a depression subtype classification system based on cross-species multimodal brain imaging, comprising the following steps:

[0006] The data preprocessing module preprocesses the cross-species multimodal brain imaging data and depression brain imaging data to obtain the preprocessed structural image data, diffusion-weighted image data and functional image data;

[0007] The cross-species mapping module builds a multimodal mapping process between the human brain and the chimpanzee brain to obtain homologous brain maps of the chimpanzee brain and the human brain;

[0008] The feature extraction module obtains the whole-brain fiber connections of the human brain and chimpanzee brain based on probabilistic fiber tracking, screens the groups to obtain human-specific connection features, and obtains the brain areas associated with these human-specific structural connections;

[0009] Depression classification module: extracts the resting-state fMRI time series corresponding to human-specific connections and calculates the Pearson correlation to obtain the functional connection feature matrix of each sample. The functional connections of patients with depression are clustered unsupervisedly using clustering methods to obtain the classification of depression.

[0010] The verification module compares the classification results with the control group based on functional imaging data indicators and brain functional connectivity, and uses the significant differences in the comparison results as the verification standard for each classification.

[0011] In the data preprocessing module, the cross-species data used are standard brain magnetic resonance data of human and chimpanzee species, including T1w images and diffusion-weighted imaging, and the depression brain imaging data are T1w images and human resting-state fMRI data; the preprocessing process of the data preprocessing module specifically includes denoising, grayscale uniform field correction, brain shell removal, gray-white matter segmentation, and surface reconstruction of the T1w structural image to obtain characteristics such as sulcus depth, curvature, thickness, surface area, and volume, and aligning the individual voxel spatial structural image to the MNI standard space; denoising, head motion correction, eddy current correction, brain shell removal, and diffusion tensor fitting are performed on the diffusion magnetic resonance data; time correction, head motion correction, spatial standardization, time series detrending, spatial smoothing, and regression to exclude head motion parameters, white matter, cerebrospinal fluid and other signals of the resting-state magnetic resonance data;

[0012] In the cross-species mapping module, the process of constructing a multimodal mapping between a human brain map and an orangutan brain map mainly includes the following steps:

[0013] A) Projecting the standard human brain atlas from voxels to the surface of the human brain, thereby obtaining a surface-based human brain atlas and extracting the corresponding brain regions;

[0014] B) The myelin distribution, groove depth, curvature, thickness, surface area, volume, etc. of the human and chimpanzee brains are used as multimodal features to drive surface-based multimodal brain mapping and calculate the transformation relationship between the two species;

[0015] C) applying the above transformation relationship to the surface-based human brain atlas to achieve mapping of brain maps between the two species and obtain a homologous whole-brain atlas corresponding to chimpanzees;

[0016] In the feature extraction module, probabilistic fiber tracking based on the whole brain is used in humans and chimpanzees respectively, and the whole brain fiber connections of the human brain and chimpanzee brain are obtained by standardization, denoising, and removal of false positives; based on the homologous brain maps between the two species obtained in S2-3, whole brain structural connection networks based on brain regions are respectively constructed to facilitate cross-species comparative analysis; the whole brain structural connections constructed for the two species are statistically analyzed in groups, and the proportion of subjects with structural connections between brain regions is calculated; in the human brain, connections with a group threshold greater than 60% are selected, and in the chimpanzee, connections with a group threshold of 0% are selected, so as to screen out human-specific connection features and obtain brain regions associated with these human-specific structural connections;

[0017] In the depression classification module, constructing a feature-based depression classification model mainly includes the following steps:

[0018] A) Based on the brain regions with unique structural connections of the human brain obtained above, extract the resting-state fMRI time series corresponding to the human-specific connections;

[0019] B) Pearson correlations were calculated between time series corresponding to human-specific connections and converted into corresponding Z values ​​using Fisher transformation, thereby obtaining the correlation coefficient of each connection for each subject to obtain the functional connection feature matrix for each sample;

[0020] C) using a dimensionality reduction method on the functional connectivity feature matrix to select all features whose eigenvalues ​​are greater than a preset value to form a feature extraction matrix, and using a clustering method to perform unsupervised clustering on the functional connectivity of patients with depression to obtain a classification of depression;

[0021] D) Using the intra-cluster error variance, profile value and the graphical tool elbow method as the judgment criteria, the best classification model was obtained;

[0022] In the verification module, the verification standard is resting-state functional magnetic resonance imaging indicators, such as local consistency (Reho), low-frequency fluctuation amplitude (ALFF), fractional low-frequency fluctuation amplitude (fALFF), functional connection density (FCD), etc., which is characterized in that SPM is used to extract the resting-state functional magnetic resonance imaging indicators of each subject, and the typing of each indicator is subjected to a two-sample t-test with the control group. After voxel-based multiple comparison correction, the comparison results of different typings are obtained, and the significant differences in the comparison results are used as the verification standard for each typing. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the implementation of the present invention or the existing technical solutions, the following briefly introduces the drawings required to be used in the embodiments or the description of the prior art.

[0024] Figure 1 It is a flowchart of a depression subtype classification system based on cross-species multimodal brain imaging provided by an embodiment of the present invention;

[0025] Figure 2 It is a schematic diagram of a process of constructing a multimodal mapping between a human brain map and an orangutan brain map by a cross-species mapping module provided by an embodiment of the present invention;

[0026] Figure 3 It is a schematic diagram of a flow chart of feature-based depression classification provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] In one embodiment, Figure 1 As shown, a depression subtype classification system based on cross-species multimodal brain imaging is provided, including a preprocessing module 10, a cross-species mapping module 20, a feature extraction module 30, a depression classification module 40 and a verification module 50, wherein the preprocessing module 10 uses cross-species data of standard brain magnetic resonance data of two species, human and chimpanzee, including T1w images and diffusion-weighted imaging, and the depression brain imaging data are T1w images and human resting-state fMRI data, and the T1w images, diffusion-weighted images and resting-state fMRI data are preprocessed respectively. The cross-species mapping module 20 uses the brain features of humans and chimpanzees to construct a multimodal mapping that drives the human brain and the chimpanzee brain, and obtains the corresponding atlas of the chimpanzee brain and the human brain. The feature extraction module 30 uses probabilistic fiber tracking based on the whole brain, and after standardization, denoising, and removal of false positives, the structural connection differences between the human brain and the chimpanzee brain are compared in groups to obtain the unique features of the human brain. The depression classification module 40 uses the extracted human-specific connections as features to reduce the dimension and cluster the depression data to obtain a classification model. The verification module 50 uses resting-state brain functional magnetic resonance imaging indicators and brain functional connections to verify the classification model.

[0029] In the above-mentioned depression subtype classification system based on cross-species multimodal brain imaging, the structural image data and functional image data in the cross-species multimodal brain imaging data and the depression brain imaging data are preprocessed respectively by the preprocessing module, and the corresponding maps of the chimpanzee brain and the human brain are obtained by the cross-species mapping module. Then, the feature extraction module is used to extract the unique features of cross-species probabilistic fiber tracking in humans, and then the depression classification module is used to classify depression based on the above-extracted features. Finally, the verification module is used to further confirm the accuracy of the classification, thereby realizing the depression subtype classification of cross-species multimodal brain imaging and improving the accuracy and reliability of the depression subtype classification of cross-species multimodal brain imaging.

[0030] In the embodiment of the present invention, Figure 2 As shown, further, the cross-species mapping module 20 specifically includes a human brain map conversion module 201, an inter-species mapping module 202 and a chimpanzee brain map output module 203:

[0031] The human brain atlas projection module 201 projects the standard human brain template from voxels to the surface of the human brain, thereby obtaining the sub-region of interest corresponding to the human brain template and extracting the corresponding brain region;

[0032] The inter-species mapping module 202 uses the myelin distribution, groove depth, curvature, thickness, surface area, volume, etc. of the brains of human and chimpanzee species as features to drive multimodal brain atlas mapping, calculates the transformation relationship between the two species, and obtains a mapping model between the human brain and the chimpanzee brain;

[0033] The chimpanzee brain atlas output module 203 uses the mapping model in 202 to obtain the mapping of the designated area of ​​the human brain in the chimpanzee brain, realizes the mapping of the corresponding area, and obtains the corresponding whole brain atlas of the chimpanzee;

[0034] In the embodiment of the present invention, Figure 3 As shown, further, the depression classification module 40 specifically includes depression data feature sequence extraction 401, feature matrix construction 402, cluster classification 403 and optimal classification model determination 404:

[0035] Depression data feature sequence extraction 401, based on the brain regions of human brain-specific structural connections obtained by the feature extraction module 30, extracting the resting state fMRI time series corresponding to the human-specific connections;

[0036] Feature matrix construction 402, Pearson correlations are calculated between time series corresponding to human-specific connections and converted into corresponding Z values ​​using Fisher transformation, thereby obtaining the correlation coefficient of each connection of each subject to obtain the functional connection feature matrix of each sample;

[0037] Cluster classification 403, using a dimensionality reduction method on the functional connection feature matrix, selecting all features with feature values ​​greater than a preset value to form a feature extraction matrix, and using a clustering method to perform unsupervised clustering on the functional connections of the depression patients to obtain a classification of depression;

[0038] The best classification model is determined 404, using the intra-cluster error variance, the profile value and the elbow method of the graphical tool as the judgment criteria to obtain the best classification model;

[0039] See also Figure 1-Figure 3 Based on the above scheme, in order to better understand the depression subtype classification system based on cross-species multimodal brain imaging provided by the embodiment of the present invention, the steps of the scheme are described in detail below:

[0040] (1) Data Acquisition

[0041] Magnetic resonance imaging data of 40 healthy human brains, Magnetic resonance imaging data of 46 healthy chimpanzees, and brain imaging data of 1,279 people with depression from multiple centers were obtained.

[0042] (2) Data preprocessing

[0043] The preprocessing steps of structural image data mainly include: denoising, grayscale uniform field correction, brain shell removal, gray-white matter segmentation, surface reconstruction, obtaining characteristics such as groove depth, curvature, thickness, surface area, volume, and registering the individual voxel space structural image to the MNI standard space; the preprocessing steps of diffusion MRI data mainly include: denoising, head motion correction, eddy current correction, brain shell removal and diffusion tensor fitting; the preprocessing steps of resting state MRI data mainly include: time correction, head motion correction, spatial standardization, time series detrending, spatial smoothing, and excluding head motion parameters, white matter and cerebrospinal fluid signals through regression.

[0044] (3) Multimodal mapping between human and chimpanzee brain maps

[0045] In order to obtain the brain map of the human brain surface, the voxel-to-surface projection method was adopted. The Freesurfer and Connectome Workbench toolbox were used to realize the conversion of various brain file formats such as .annot, .func.gii, and .label.gii. The standard AAL template was projected onto the human brain surface to obtain the ROI corresponding to the AAL template. A total of 80 brain regions were extracted, and some overlapping areas under the cortex were selected as the standard, which were closer to the cortex.

[0046] The myelin distribution, groove depth, curvature, thickness, surface area, volume, etc. of the brains of human and chimpanzee species are used as features to drive multimodal brain mapping. The above features are used as multimodal input to map the human brain to the chimpanzee brain map, and the correspondence and transformation relationship between the two species is calculated, which is used for mapping the corresponding areas across species. Using this transformation relationship, the mapping of the specified area of ​​the human brain in the chimpanzee brain can be obtained, the mapping of the corresponding area can be achieved, and the AAL map of the whole chimpanzee brain can be obtained.

[0047] (4) Extraction of unique features of the human brain

[0048] Probabilistic fiber tracking based on the whole brain was used in humans and chimpanzees respectively. Based on the homologous brain maps between the two species, the whole brain structural connection network based on brain regions was constructed respectively, and the inter-ROI structural connection of the AAL brain maps of the two species was obtained, which was 80*80*40 for the human brain and 80*80*46 for the chimpanzee brain. For the connection matrix of the sample, the connection of each ROI of each sample was first divided by the total number of connections passing through the ROI for standardization, and then the connections with a probability greater than 0.05% were taken as effective connections for denoising, and then the connections between the ROIs were counted to obtain the upper triangular matrix of the unique connection probability between each ROI, and the average connection probability between the ROIs of the two species was obtained.

[0049] For the above structural connection results, the whole-brain structural connections constructed for the two species were counted in groups, and the proportion of subjects with structural connections between brain regions was calculated. Humans and chimpanzees were compared at the group level. In the human brain, connections with a group threshold greater than 60% were selected, and in chimpanzees, connections with a group threshold of 0% were selected to screen out human-specific connection features and obtain the brain regions associated with these human-specific structural connections. A total of 223 human-specific inter-ROI structural connections and corresponding coordinates were screened out, and human-specific connections were used as features for further analysis.

[0050] (5) Constructing a feature-based depression classification model

[0051] Based on the brain regions with unique structural connections of the human brain obtained above, the time series corresponding to the AAL template of the resting-state fMRI of the depression brain imaging dataset were extracted. The Pearson correlation between the time series corresponding to the human-specific connections was calculated and converted into the corresponding Z value using Fisher transformation, so as to obtain the correlation coefficient of each connection of each subject, and finally obtain the functional connection feature matrix of each sample.

[0052] First, PCA was used to reduce the dimension of the functional connection feature matrix obtained above, and all features with eigenvalues ​​greater than 1 were taken. The kmeans clustering method was used to perform unsupervised clustering of the functional connections of MDD patients to obtain 2-10 types of MDD.

[0053] Then, the optimal number of clusters is selected using the within-cluster error variance (SSE) and silhouette index as the judgment criteria, and the within-cluster error variance is visualized according to the number of clusters using the graphical tool elbow method.

[0054] Finally, since the inflection point of the decreasing curve tends to be flat when the intra-cluster error variance k is 3 and the profile value is the highest when k is 3, k is determined to be 3 as the number of clusters, and three types are obtained, corresponding to 486, 520, and 273 subjects respectively.

[0055] (6) Verification of typing results

[0056] Resting-state functional magnetic resonance imaging local consistency (Reho), low-frequency fluctuation amplitude (ALFF), fractional low-frequency fluctuation amplitude (fALFF) and functional connectivity density (FCD) were used as verification criteria for each classification. First, SPM was used to extract the resting-state functional magnetic resonance imaging index of each subject, and the classification of each index was compared with the control group by a two-sample t-test and voxel-based multiple comparison correction (FWE corrected p<0.05) to obtain the comparison results of different classifications. By comparing the significant differences between different classifications and the control group, the feasibility of the classification was further verified;

[0057] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0058] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A depression subtype classification system based on cross-species multimodal brain imaging, characterized by: The steps include: S1, data preprocessing module, preprocesses cross-species multimodal brain imaging data and depression brain imaging data respectively to obtain preprocessed structural image data, diffusion weighted image data and functional image data; S2, cross-species mapping module, constructs a multimodal mapping process between the human brain and the chimpanzee brain to obtain homologous brain maps of the chimpanzee brain and the human brain; S3, feature extraction module, obtains the whole-brain fiber connections of human and chimpanzee brains based on probabilistic fiber tracking, screens the groups to obtain human-specific connection features, and obtains the brain areas associated with these human-specific structural connections; S4, depression classification module, extracts the resting-state fMRI time series corresponding to human-specific connections and calculates the Pearson correlation to obtain the functional connection feature matrix of each sample. The functional connections of patients with depression are clustered unsupervisedly using clustering methods to obtain the classification of depression. S5, verification module, compares the classification results with the control group based on functional imaging data indicators and brain functional connectivity, and uses the significant difference in the comparison results as the verification standard for each classification.

2. A depression subtype classification system based on cross-species multimodal brain imaging according to claim 1, characterized in that: In S1, the cross-species data used are standard brain magnetic resonance data of human and chimpanzee species, including T1w images and diffusion-weighted imaging, and the brain imaging data of depression are T1w images and human resting-state fMRI data; the preprocessing process of the data preprocessing module is specifically to perform denoising, grayscale uniform field correction, brain shell removal, gray-white matter segmentation, and surface reconstruction on the T1w structural image to obtain characteristics such as sulcus depth, curvature, thickness, surface area, and volume, and align the individual voxel spatial structural image to the MNI standard space; perform denoising, head motion correction, eddy current correction, brain shell removal, and diffusion tensor fitting on the diffusion magnetic resonance data; perform time correction, head motion correction, spatial standardization, time series detrending, spatial smoothing, and exclude head motion parameters, white matter, cerebrospinal fluid and other signals through regression on the resting-state magnetic resonance data.

3. A depression subtype classification system based on cross-species multimodal brain imaging according to claim 1, characterized in that: In S2, the cross-species mapping module constructs a multimodal mapping process between the human brain map and the orangutan brain map, which mainly includes the following steps: S2-1, projecting the voxel-based human brain map onto the human brain surface template, thereby obtaining the surface-based human brain map; S2-2, the myelin distribution, groove depth, curvature, thickness, surface area, volume, etc. of the brains of humans and chimpanzees are used as multimodal features to drive surface-based multimodal brain mapping, and the transformation relationship between the two species is calculated; S2-3, using the transformation relationship in S2-2 to apply to the surface-based human brain map, to achieve the mapping of brain maps between the two species, and obtain the corresponding homologous whole-brain map of chimpanzees.

4. A depression subtype classification system based on cross-species multimodal brain imaging according to claim 1, characterized in that: In S3, whole-brain probabilistic fiber tracking was used in humans and chimpanzees, and normalization, denoising, and false positive removal were performed to obtain whole-brain fiber connections of the human and chimpanzee brains. Based on the homologous brain maps between the two species obtained by S2-3, whole-brain structural connection networks based on brain regions were constructed respectively for cross-species comparative analysis; the whole-brain structural connections constructed for the two species were statistically analyzed in groups, and the proportion of subjects with structural connections between brain regions was calculated; in the human brain, connections with a group threshold greater than 60% were selected, and in chimpanzees, connections with a group threshold of 0% were selected. Through comparative analysis, human-specific connection features were screened out, and the brain regions associated with these human-specific structural connections were obtained.

5. A depression subtype classification system based on cross-species multimodal brain imaging according to claim 1, characterized in that: In S4, the depression classification model module mainly includes the following steps: S4-1, based on the brain regions with unique structural connections of the human brain obtained in S3, extract the resting-state fMRI time series corresponding to the human-specific connections; S4-2, Pearson correlation was calculated between the time series of human-specific structural connection brain regions, and converted into corresponding Z values ​​using Fisher transformation, so as to obtain the correlation coefficient of each connection of each subject and obtain the functional connection feature matrix of each sample; S4-3, using a dimensionality reduction method on the functional connection feature matrix, selecting all features whose eigenvalues ​​are greater than a preset value to form a feature extraction matrix, and using a clustering method to perform unsupervised clustering on the functional connections of patients with depression to obtain a classification of depression; S4-4, using intra-cluster error variance, profile value and the graphical tool elbow method as the judgment criteria, the best classification model was obtained.

6. A depression subtype classification system based on cross-species multimodal brain imaging according to claim 1, characterized in that: In S5, the resting-state functional magnetic resonance imaging indicators of each subject were extracted, such as local consistency (Reho), amplitude of low-frequency fluctuations (ALFF), fractional amplitude of low-frequency fluctuations (fALFF), functional connectivity density (FCD), etc.; the classification of each indicator was subjected to a two-sample t-test with the control group, and the comparison results of different classifications were obtained after voxel-based multiple comparison correction, and the significant differences in the comparison results were used as the verification criteria for each classification.

Images