A biological subtype identification method based on brain injury mapping of multiple brain function states

Abstract

The application discloses a brain injury mapping biological subtype recognition method based on multiple brain function states, and belongs to the technical field of medical image processing and computational neuroscience. The application first constructs an epilepsy brain network vulnerability evaluation system based on multiple brain state functional connections, can systematically depict abnormal patterns of brain networks at the individual level in multiple task states and resting states, and breaks through the limitation that traditional single brain state analysis is difficult to capture dynamic propagation characteristics of diseases. Through spatial coupling analysis of the structural atrophy atlas and the multiple state functional connection template, the application can quantitatively identify epicenters in different functional states as disease transmission hubs, and realize multidimensional description of epilepsy network abnormalities. The dual clustering framework proposed by the application can integrate common information of multiple brain states, recognize epilepsy biological subtypes with robustness and clinical consistency, and provide technical support for revealing differences in epicenter distribution between primary cortex and high-level cortex of different subtypes. On this basis, the brain vulnerability model divides the brain region vulnerability level in two dimensions of atrophy and epicenter, explains the pathological mechanism of epilepsy from the network propagation perspective, and provides a quantitative analysis tool for verifying the network propagation hypothesis of the disease.

Description

Technical Field

[0001] This invention belongs to the field of medical image processing and computational neuroscience technology, specifically involving a multimodal analysis method that integrates structural magnetic resonance imaging, functional magnetic resonance imaging, brain network connectivity, and behavioral phenotypes, for brain network typing, disease mechanism analysis, and treatment prognosis prediction in patients with generalized epilepsy. Background Technology

[0002] Generalized epilepsy is a neurological disorder characterized by widespread epileptic discharges, exhibiting significant heterogeneity in clinical manifestations, brain structural changes, functional abnormalities, and drug responses. Currently, the pathogenesis of generalized epilepsy is not fully understood; clinical classification primarily relies on seizure type and electroencephalogram (EEG), which struggles to explain individual differences and variations in treatment response. Magnetic resonance imaging (MRI) studies have shown that generalized epilepsy is not a structurally normal disease but rather involves widespread abnormalities in sensorimotor networks, visual networks, default mode networks, and subcortical structures, suggesting that research should focus on network-based approaches. Brain functional abnormalities in epilepsy patients are not limited to the resting state but also occur in primary sensorimotor and higher cognitive tasks. The integration of structural and functional characteristics has become a hot topic in epilepsy research in recent years.

[0003] Previous studies have largely employed single brain states (resting state or a specific task state) or single imaging modalities for population-level analysis, making it difficult to reveal the network propagation mechanisms of diseases and individual-level differences. On the other hand, the "network propagation hypothesis" and "node pressure hypothesis" of brain diseases positize that diseases do not occur solely at local lesions but spread through functional connections within the brain network, with certain key brain regions acting as disease hubs and playing a central role in disease occurrence and progression. However, current research largely remains at the population average level, lacking a systematic analytical framework at the individual level, encompassing multiple brain states, and applicable to classification and prognostic prediction. Furthermore, the response of epilepsy patients to anti-epileptic drugs varies significantly, and currently, reliable imaging or biological indicators are lacking for predicting treatment outcomes. Analysis based solely on structural or functional abnormalities is insufficient to explain these differences.

[0004] Therefore, it is necessary to propose a new analytical framework: at the individual level, integrate multi-brain state functional connectivity and structural abnormality information to quantify the criticality of different brain regions as disease transmission hubs; based on this, identify clinically consistent disease biological subtypes and reveal the spatial distribution patterns of different subtypes in brain network vulnerability; and further establish a correlation model between brain susceptibility characteristics and clinical outcomes to provide systematic technical support for the mechanism analysis, biological subtype identification, and treatment prognosis prediction of generalized epilepsy. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the present invention aims to propose a disease hub mapping and brain vulnerability modeling method based on multi-brain state functional connectivity, which analyzes the abnormal brain network patterns of generalized epilepsy at the individual level, constructs an interpretable disease classification system, and further enables the prediction of treatment efficacy.

[0006] This invention quantifies the degree of brain atrophy and its critical role as a disease transmission hub at the individual level by fusing structural and multi-state functional imaging, revealing the commonalities and specificities of brain network abnormalities in different functional states. Based on this, a dual clustering framework is constructed to identify clinically consistent disease biological subtypes, and the brain network damage patterns and susceptibility characteristics of different subtypes are systematically analyzed. Furthermore, a brain susceptibility quantification model is established to explain the disease occurrence and development mechanism from the perspective of network transmission, and a clinical efficacy prediction model is constructed based on susceptibility characteristics, realizing a comprehensive analysis from disease mechanism analysis to individualized prognostic assessment.

[0007] The technical solution of this invention is a biological subtype identification method based on brain injury mapping of multiple brain functional states, the method comprising the following steps:

[0008] Step 1: Obtain structural T1-weighted magnetic resonance (MRI) data and functional magnetic resonance (fMRI) data of the subject in resting state and various task states. Segment the T1-weighted MRI data to obtain the gray matter volume (GMV) map. Process the fMRI data, including removing the first 5 time points, time-layer correction, head movement correction, standardization and regression of head movement, cerebrospinal fluid, and whole brain mean.

[0009] Step 2: Based on the standard cortical partition template and subcortical partition template, the whole brain is divided into several brain regions; the GMV of each brain region is calculated, and the ComBat method is used to eliminate the site effect of multicenter data; with healthy controls as a reference, the Z score of the GMV of each brain region of the patient is calculated as an indicator of brain atrophy deviation.

[0010] Step 3: Using the same brain region template as in Step 2, calculate the functional connectivity matrix of healthy controls in the resting state and various task states, and use the group average as the standard functional connectivity. For each patient, perform spatial correlation analysis between their GMV deviation map and the standard functional connectivity pattern of each brain region, and quantify the key role of each brain region in the occurrence and development of the disease using the correlation coefficient. For each state, generate a whole-brain correlation coefficient map for each patient, which is the epicenter map.

[0011] Step 4: First, for any given brain state, a matrix is ​​formed by concatenating the key vectors of all patients' brain regions. Then, a cohesive hierarchical clustering method is used, based on the distance between samples and the average link criterion, to progressively merge similar clusters and construct a hierarchical clustering tree structure, obtaining the cluster labels for individual patients. Next, combining the cluster labels from multiple brain states, a multi-state clustering matrix for patients is constructed. Based on this matrix, a secondary cohesive hierarchical clustering is performed to obtain patient subtypes reflecting the consistency of multiple brain states, resulting in the final subtype labels.

[0012] Step 5: For any given subtype, first obtain the epicenter map of all patients in all states of that subtype, then calculate the single-sample T-test statistic for each state, and then average the single-sample T-test statistic for each state to obtain the final measure index characterizing the cross-state epicenter, and determine the brain region with a false detection rate of less than 0.05 as the epicenter region.

[0013] Step 6: For any brain region of the patient, construct a susceptibility model by combining whether it is atrophied and whether it is the epicenter, and divide the brain region into four categories with susceptibility from high to low: both atrophied and epicenter is an infectious region, only atrophied is a damaged region, not atrophied but is the epicenter is a susceptible region, and neither atrophied nor epicenter is a normal region.

[0014] Furthermore, the brain atrophy deviation index in step 2 is specifically calculated as follows:

[0015] ;

[0016] in, For the first The first patient's Deviation indicators of brain atrophy from healthy controls in individual brain regions For the first The first patient's GMV of each brain region For healthy controls at the 1st Mean GMV of each brain region For healthy controls, in the first... The standard deviation of GMV for each brain region; regions with a Z value less than -2 are defined as atrophic regions.

[0017] Furthermore, the correlation calculation method in step 3 is as follows:

[0018] ;

[0019] in: Represents the correlation coefficient. Indicates sample size. Indicates the first The rank difference of a sample after sorting by two variables. ,in For variables rank value, For variables The rank value.

[0020] Furthermore, the specific formula for calculating the single-sample T-test statistic in step 5 is as follows:

[0021] ;

[0022] in: This is the t-test statistic for a one-sample test. The sample mean. The population mean The standard deviation of the sample is 1. Here are the sample size and sample standard deviation. The calculation formula is: .

[0023] The beneficial effects of this invention are as follows: This invention, for the first time, constructs an epilepsy brain network vulnerability assessment system based on multi-state functional connectivity. It can systematically characterize abnormal brain network patterns across resting and various task states at the individual level, overcoming the limitation of traditional single-state analysis in capturing the dynamic propagation characteristics of the disease. By spatially coupling structural atrophy maps with multi-state functional connectivity templates, this invention can quantitatively identify epicenter regions that serve as disease propagation hubs under different functional states, achieving a multi-dimensional description of epilepsy network abnormalities. The proposed dual clustering framework can integrate common information from multiple brain states, identifying robust and clinically consistent epilepsy subtypes, providing technical support for revealing the differences in epicenter distribution between primary and higher cortices of different subtypes. Based on this, a brain susceptibility model is constructed, dividing brain region susceptibility levels through a dual dimension of atrophy and epicenter, elucidating the pathological mechanism of epilepsy from a network propagation perspective, and providing a quantitative analysis tool for verifying the network propagation hypothesis of the disease. Furthermore, the association model between brain susceptibility characteristics and clinical drug response established in this invention can effectively predict treatment efficacy based on the proportion of four susceptibility regions in an individual, providing imaging reference for early identification and individualized medication of refractory epilepsy. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the overall process of this invention.

[0025] Figure 2 This refers to the cross-state epicenter region for different biological subtypes of epilepsy patients.

[0026] Figure 3 This study assesses the classification effect of the brain susceptibility model on the clinical drug treatment response in epilepsy patients. Detailed Implementation

[0027] The technical solution of the present invention will be further described below with reference to embodiments. The present invention proposes a method for analyzing epilepsy biological subtypes based on brain injury mapping and brain vulnerability analysis across multiple brain functional states. Through multi-center, multimodal imaging and behavioral data fusion, it performs disease injury mapping analysis combining resting state and multiple task states to identify disease biological subtypes, construct a brain susceptibility model, and reveal the responses of different biological subtypes to clinical treatment. Specifically, it includes the following steps:

[0028] Step 1: This invention first acquires multimodal brain imaging and behavioral data (including structural images, resting-state and task-oriented fMRI, and clinical data such as drug treatment response) from independent datasets of epilepsy patients and healthy controls. All brain imaging data are from 3T MRI, and the total sample includes 348 patients with generalized epilepsy. All patients with generalized epilepsy were diagnosed by neurologists according to the International League Against Epilepsy criteria. Healthy controls had no history of mental illness, neurological disease, or drug abuse. All participants signed informed consent.

[0029] Step 2: This invention employs a unified preprocessing workflow for MRI data from different sites. The main steps include deleting the first five time points, time correction, head motion correction, structural image segmentation and spatial standardization, filtering, covariate regression, and spatial smoothing. To eliminate site effects, this invention uses the ComBat algorithm to correct all structural functional features.

[0030] Step 3: After data preprocessing, this invention first obtains the GMV (Gross Mercury Volume) of the whole brain regions of interest for patients and healthy controls based on T1 structural image segmentation and using a priori brain partitioning template. Subsequently, the patient's GMV is Z-scored using healthy controls as a standard to obtain the degree to which each brain region deviates from the healthy population. This deviation map is the basis for subsequent mapping analysis.

[0031] Step 4: In order to characterize the key role of different brain regions in the occurrence and development of the disease, this invention first uses the same whole-brain regional division as in Step 3 in healthy individuals, and calculates the functional connectivity between regions of interest in both resting and task states; then, spatial correlation analysis is performed between the patient's GMV deviation map and the whole-brain functional connectivity map of each region of interest in the healthy individuals, and the magnitude of the correlation is used to quantify the probability of a certain region as the epicenter.

[0032] Step 5: In order to fully consider the impact of different brain states on the identification of biological subtypes, this invention adopts a dual clustering method: First, within the same state, patients are hierarchically clustered based on the whole-brain epicenter probability map; then, hierarchical clustering is performed again based on the clustering labels of patients in different states, thus identifying two robust cross-state biological subtypes.

[0033] Step 6: To obtain the specific epicenter region for each biological subtype, this invention performs a one-sample t-test within each state on the epicenter maps of patients of the same subtype under all states, and obtains the statistical map of the corresponding epicenter probability by averaging across states. The final epicenter region is determined with an FDR less than 0.05 as the threshold. The results show that subtype 1 has the primary cortex, such as sensorimotor cortex, as the key epicenter region, while subtype 2 has the subcortical and frontoparietal higher cortex as the epicenter region.

[0034] Step 7: To clarify the clinical significance and value of the epicenter, this invention proposes a brain susceptibility model. Based on the characteristics of atrophy and epicenter, the whole brain region is divided into four categories of susceptibility from high to low. The proportion of each of the four categories is calculated. It is found that the proportion of brain susceptibility regions in subtype 1 is less than that in subtype 2, and the clinical drug efficacy is significantly higher in subtype 1 than in subtype 2. In addition, the proportion of the four categories of regions successfully classifies the clinical drug response, providing support for individualized precision treatment in clinical practice.

[0035] In this embodiment, the brain injury mapping analysis framework combining multiple brain states constructed by the present invention can stably identify biological subtypes of epilepsy, revealing the close correlation between different epicenter distribution patterns and clinical efficacy. The brain susceptibility model constructed by the present invention provides a foundation for clinical efficacy prediction and individualized precision treatment, and provides possible directions for future clinical intervention strategies.

[0036] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A method for identifying biological subtypes based on brain injury mapping of multiple brain functional states, the method comprising the following steps: Step 1: Obtain structural T1-weighted magnetic resonance (MRI) data and functional magnetic resonance (fMRI) data of the subject in resting state and various task states. Segment the T1-weighted MRI data to obtain the gray matter volume (GMV) map. Process the fMRI data, including removing the first 5 time points, time-layer correction, head movement correction, standardization and regression of head movement, cerebrospinal fluid, and whole brain mean. Step 2: Based on the standard cortical partition template and subcortical partition template, the whole brain is divided into several brain regions; the GMV of each brain region is calculated, and the ComBat method is used to eliminate the site effect of multicenter data; Using healthy controls as a reference, the Z-score of GMV for each brain region of the patients was calculated as an indicator of brain atrophy deviation; Step 3: Using the same brain region template as in Step 2, calculate the functional connectivity matrix of healthy controls in the resting state and various task states, and use the group average as the standard functional connectivity. For each patient, perform spatial correlation analysis between their GMV deviation map and the standard functional connectivity pattern of each brain region, and quantify the key role of each brain region in the occurrence and development of the disease using the correlation coefficient. For each state, generate a whole-brain correlation coefficient map for each patient, which is the epicenter map. Step 4: First, for any given brain state, the matrix formed by concatenating the key vectors of all patients' brain regions is used to construct a hierarchical clustering tree structure by merging similar clusters step by step based on the distance between samples and the average link criterion, thereby obtaining the clustering labels of individual patients. Then, by combining clustering labels of multiple brain states, a multi-state clustering matrix for patients is constructed; Based on this matrix, secondary cohesive hierarchical clustering is performed to obtain patient subtypes that reflect the consistency of multiple brain states, and the final subtype labels are obtained. Step 5: For any given subtype, first obtain the epicenter map of all patients in all states of that subtype, then calculate the single-sample T-test statistic for each state, and then average the single-sample T-test statistic for each state to obtain the final measure index characterizing the cross-state epicenter, and determine the brain region with a false detection rate of less than 0.05 as the epicenter region. Step 6: For any brain region of the patient, construct a susceptibility model by combining whether it is atrophied and whether it is the epicenter, and divide the brain region into four categories with susceptibility from high to low: both atrophied and epicenter is an infectious region, only atrophied is a damaged region, not atrophied but is the epicenter is a susceptible region, and neither atrophied nor epicenter is a normal region.

2. The biological subtype identification method based on brain injury mapping of multiple brain functional states as described in claim 1, characterized in that, The brain atrophy deviation index in step 2 is specifically calculated as follows: ； in, For the first The first patient's Deviation indicators of brain atrophy from healthy controls in individual brain regions For the first The first patient's GMV of each brain region For healthy controls at the 1st Mean GMV of each brain region For healthy controls, in the first... The standard deviation of GMV for each brain region; regions with a Z value less than -2 are defined as atrophic regions.

3. The biological subtype identification method based on brain injury mapping of multiple brain functional states as described in claim 1, characterized in that, The correlation calculation method in step 3 is as follows: ； in: Represents the correlation coefficient. Indicates sample size. Indicates the first The rank difference of a sample after sorting by two variables. ,in For variables rank value, For variables The rank value.

4. The biological subtype identification method based on brain injury mapping of multiple brain functional states as described in claim 1, characterized in that, The specific formula for calculating the single-sample T-test statistic in step 5 is as follows: ； in: This is the t-test statistic for a one-sample test. The sample mean. The population mean The standard deviation of the sample is 1. Here are the sample size and sample standard deviation. The calculation formula is: .

Images