Teenager depression cognitive impairment subtype classification and prognosis prediction method

Abstract

A juvenile depression cognitive impairment subtype classification and prognosis prediction method relates to the technical field of medical treatment, and mainly comprises the following steps: performing clinical evaluation and therapeutic response evaluation on a subject, performing MRI and magnetoencephalogram data acquisition, constructing a whole brain MSN of the subject, identifying MSN abnormal characteristics, obtaining functional connection change of a frequency band when magnetoencephalogram is abnormal, and determining the cognitive impairment subtype classification and prognosis prediction of the cognitive impairment subtype of the subject. A subtype classification model is established by fusing the MSN and cognitive function evaluation data, and a prognosis prediction model is established by analyzing MSN abnormal features, functional connection changes of frequency bands during abnormality, multi-dimensional treatment reactions and high-risk behaviors. According to the method, different levels of fusion measurement are carried out on the juvenile depression with cognitive function impairment brain mechanism through multi-modal brain images, a subtype classification model with diagnosis and treatment values is established, and a prognosis prediction model with clinical transformation potential is constructed; therefore, a theoretical basis and a technical means are provided for individualized precise diagnosis and treatment of the cognitive impairment of the juvenile depression.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and in particular to a method for classifying subtypes of cognitive impairment in adolescent depression and predicting prognosis. Background Art

[0002] Adolescent depression is a psychological disorder that occurs in adolescents, mostly caused by personal character factors and heavy academic pressure. Adolescent depression is a serious mental health problem that can lead to persistent sadness and loss of interest in activities. It affects adolescents' thinking, feelings and behavior, and may cause emotional, functional and physical problems.

[0003] Currently, some brain network technologies have been used in the study of brain structure and cognitive impairment in depression, such as the morphometric similarity network (MSN). MSN was first proposed by Professor Bullmore's team at the University of Cambridge to characterize the covariation pattern of whole-brain structure in children and predict cognitive differences between individuals by calculating the MSN node degree. Specifically, MSN characterizes the structural covariation across brain regions at the individual level by calculating the correlation between any pair of brain regions in multimodal morphological characteristics (including gray matter morphological indicators such as cortical thickness and cortical surface area; white matter morphological indicators such as anisotropy fraction and mean diffusivity) to reflect the similarity of cytoarchitecture, genetic expression similarity, and white matter connectivity between brain regions.

[0004] Another example is atomic magnetometer magnetoencephalography (MEG), a technique that complements the high spatial resolution of MSNs. As one of the most cutting-edge noninvasive, safe, and noiseless functional imaging technologies, MEG can present real-time images of neuronal activity in the brain with millisecond-level temporal resolution. It has been widely used in fundamental cognitive science fields such as language processing, auditory processing, and visual perception. Combined with analytical techniques such as frequency-tagged stimulation paradigm design and phase-amplitude coupling, MEG can capture the brain's dynamic processing mechanisms of information. Currently, MEG has been gradually applied to the study of cognitive impairment in depression. For example, in facial expression recognition tasks, depressed patients exhibit abnormal spatiotemporal interactions in the beta-band cortical-limbic-striatal neural circuit. However, previous MEG studies of depression have used relatively simple cognitive testing tasks and simple stimulation paradigms, failing to reflect the complex and multidimensional changes in cognitive function.

[0005] Currently, the subtype classification of adolescent depression is mainly based on cognitive symptoms or imaging data, and prognosis prediction also mainly relies on regression analysis of simple linear correlation between clinical symptoms and imaging indicators.

[0006] The shortcomings of existing technologies include: first, subtype classification based solely on clinical symptoms. Since the same symptoms can stem from different biological causes, and vice versa, the same biological cause can lead to different symptoms. Furthermore, it is difficult to predict disease outcomes based on symptoms, leading to significant treatment heterogeneity. Second, prognostic models that primarily rely on simple linear regression of clinical symptoms and imaging indicators suffer from nonlinear relationship inaccuracies, interaction blind spots, neglect of temporal dynamics, inadequate high-dimensional data processing, confounding by competing risks, amplified measurement errors, and masking of patient heterogeneity. Summary of the Invention

[0007] Based on the above problems, the purpose of the present invention is to provide a method for classifying subtypes of cognitive impairment in adolescents with depression and predicting prognosis. Through multimodal brain imaging, the brain mechanisms of adolescents with depression accompanied by cognitive impairment are measured at different levels of fusion, a subtype classification model with diagnostic and therapeutic value is established, and a prognosis prediction model with clinical translation potential is constructed, thereby providing a theoretical basis and technical means for the individualized and precise diagnosis and treatment of cognitive impairment in adolescents with depression.

[0008] The technical solution adopted by the present invention to achieve its invention object is a method for classifying subtypes of cognitive impairment in adolescent depression and predicting prognosis, comprising the following steps:

[0009] S1. Conduct clinical assessments on subjects at baseline (i.e., enrollment) and during follow-up, including assessment of high-risk behavioral characteristics;

[0010] S2. Evaluate the treatment response of the subjects during follow-up to obtain multi-dimensional treatment response;

[0011] S3. MRI data were collected at the time of enrollment and follow-up to obtain morphological indicators of the cortical and subcortical regions;

[0012] S4. Performing magnetoencephalography (MEG) data collection and data preprocessing on subjects during enrollment and follow-up. The MEG data collection includes collecting dynamic changes in brain activity signals under different cognitive test tasks.

[0013] S5. Construct and analyze individual whole-brain MSNs to obtain abnormal MSN characteristics related to multi-dimensional cognitive functions;

[0014] S6. Analyze magnetoencephalography data to obtain functional connectivity changes in abnormal time-frequency bands;

[0015] S7. Subtype classification is performed using abnormal MSN features related to the subjects' multidimensional cognitive functions and functional connectivity in abnormal time-frequency bands;

[0016] S8. Construct a prognostic prediction model, including:

[0017] The subjects' abnormal MSN features related to multidimensional cognitive function, functional connectivity changes in abnormal time-frequency bands, multidimensional treatment responses, and high-risk behaviors were subjected to rCCA regularized canonical correlation analysis to obtain the brain network features and their expression levels corresponding to different treatment responses and high-risk behaviors, and screen out the most clinically significant brain network features.

[0018] The brain network features screened out from the subject training set are added to obtain a brain network feature sum array, and a prognostic prediction model of the brain network feature sum array for treatment response and high-risk behavior is established based on multiple linear regression;

[0019] S9. Calculate the sum array of brain network features for the subject to be predicted, and then input the sum array into the prognosis prediction model to obtain the predicted values ​​of treatment response and high-risk behavior.

[0020] Furthermore, the high-risk behaviors include self-harm, suicidal, impulsive, and violent behaviors;

[0021] The therapeutic response includes ineffective treatment, partially effective treatment, effective treatment, and clinical remission;

[0022] The treatment response assessment described in step S2 is specifically determined by calculating the HAMD score reduction rate = [(post-treatment score - pre-treatment score) / pre-treatment score] * 100%. A HAMD score reduction rate <25% indicates ineffective treatment, 25% ≤ HAMD score reduction rate <50% indicates partially effective treatment, 50% ≤ HAMD score reduction rate <75% indicates effective treatment, and a HAMD score reduction rate >75% indicates clinical remission.

[0023] Further, the step S3 is specifically as follows:

[0024] S31. Perform high-resolution 3D-T1WI structural images and multi-shell diffusion-weighted imaging (DWI). Use multi-shell DWI for diffusion tensor imaging (DTI) and neurite direction dispersion and density imaging diffusion model analysis.

[0025] S32. Perform data preprocessing. Preprocessing of high-resolution 3D-T1WI structural images includes motion correction, skull stripping, Talairach coordinate transformation, grayscale homogenization, white matter and gray matter segmentation, and topological correction. Preprocessing of multi-slice diffusion-weighted imaging (DWI) data includes eddy current removal and motion correction, brain region segmentation, and establishment of a voxel-based diffusion tensor model.

[0026] S33. For the cortical region, the following morphological features were extracted based on the Glasser atlas: cortical thickness CT, cortical surface area SA, cortical volume CV, local folding index LGI, anisotropy fraction FA, mean diffusivity MD, radial diffusivity RD, axial diffusivity AD, neuronal density index NDI, directional dispersion index ODI, and isotropic volume fraction ISOVF;

[0027] S34. For the subcortical regions, the following morphological features were extracted from 17 subcortical nuclei - bilateral hippocampus, amygdala, thalamus, globus pallidus, nucleus accumbens, putamen, caudate nucleus, hypothalamus and the entire brainstem - based on the Aseg coordinates: volume, fractional anisotropy FA, ​​mean diffusivity MD, radial diffusivity RD, axial diffusivity AD, neuronal density index NDI, directional dispersion index ODI and isotropic volume fraction ISOVF.

[0028] Furthermore, in step S4, the cognitive test tasks include facial expression recognition, Go / No-go experiment, N-back experiment, and visual motion experiment, and the data preprocessing includes spatial registration correction, artifact removal, filtering, power spectral density analysis, bad conductor removal, epoch segmentation, and baseline correction.

[0029] Furthermore, the step S5 is specifically as follows:

[0030] S51, constructing a whole-brain morphometric similarity network (MSN) based on the morphological indices of the cortex and subcortical regions obtained in step S3, and identifying the morphological indices with the greatest contribution;

[0031] S52. All subjects' MSNs were divided into modules. Differences in the inner and outer edges of different modules of MSNs between subjects and healthy controls were compared and analyzed. Functional annotation analysis was performed on the edges with significant differences. Changes in the inner and outer edges of different modules of MSNs of subjects were longitudinally followed up before and after treatment, and the correlation between baseline MSNs and follow-up MSNs was analyzed.

[0032] S53. The LMM linear mixed-effects model was used to analyze the relationship between abnormal MSN characteristics and clinical symptoms, and the rCCA regularized canonical correlation analysis was used to determine the relationship between abnormal MSN characteristics and multidimensional cognitive functions.

[0033] Furthermore, the specific method of constructing the whole-brain morphometric similarity network MSN in step S51 is:

[0034] All morphological indices were z-value standardized, and cortical MSNs, subcortical MSNs, and cortical-subcortical MSNs were constructed based on 360 brain regions bilaterally in the Glasser atlas and 17 subcortical nuclei in the Aseg coordinates;

[0035] The specific method of identifying the morphological index with the greatest contribution in step S51 is:

[0036] Importance feature analysis was performed on cortical MSNs, subcortical MSNs, and cortical-subcortical MSNs, respectively. That is, one indicator was deleted from all morphological indicators each time, the MSNs were recalculated, and the Pearson correlation coefficient between the MSNs with all indicators and the MSNs after the indicators were deleted was calculated to identify the morphological indicator with the greatest contribution.

[0037] Furthermore, the module division in step S52 is specifically as follows:

[0038] The cortical MSNs were divided into seven modules according to the Yeo functional network atlas, including the visual network, sensorimotor network, dorsolateral attention network, ventral attention network, limbic network, frontoparietal network, and default network;

[0039] Subcortical MSNs were divided into K-means clusters based on the morphological correlation of subcortical nuclei;

[0040] The cortical-subcortical MSNs were divided by K-means clustering based on the morphological correlation between subcortical nuclei and cortical modules.

[0041] Further, the step S6 is specifically as follows:

[0042] S61. Under the stimulation of different cognitive test tasks, the time-frequency analysis method was used to extract spectral features. The time-frequency energy values ​​at the sensor level were compared between groups using the cluster permutation test. The time-frequency segments with significant differences between the groups were selected as regions of interest.

[0043] S62. By comparing the brain area energy values ​​of the subjects and healthy controls at the source level under the region of interest, the brain areas involved in the abnormal time-frequency signals are obtained, and the abnormal functional connectivity changes between brain areas in the maximum functional connectivity window are calculated.

[0044] Further, the step S7 is specifically as follows:

[0045] Abnormal MSN features related to the subjects' multidimensional cognitive functions and functional connectivity changes in abnormal time-frequency bands were merged and vectorized, centered on the mean, and PCA unsupervised dimensionality reduction was performed using singular value decomposition to extract the principal component values ​​of the number of principal components with an overall model explanation rate greater than or equal to 80%;

[0046] The principal component values ​​were clustered using the HYDRA heterogeneity discriminant analysis method. The adjusted Rand index (ARI) was calculated using 10-fold cross-validation to select the optimal number of clusters. The clinical and brain imaging features of each subtype were compared and clarified to verify the accuracy of the classification. The XGBoost model with Bayesian parameter optimization was used to distinguish the imaging features of each subtype.

[0047] Furthermore, the expression of the prognosis prediction model in step S8 is:

[0048] Y=β0+β1X1+β2X2+…+β k X k +∈, where Y is the dependent variable including treatment response score or high-risk behavior risk value, X1, X2, …, X k is the input variable of the brain network feature, that is, the feature in the sum array of the brain network features, β0 is the intercept term, β1, β2, ..., β k is the regression coefficient, ∈ is the error term

[0049] The beneficial effects of the present invention are:

[0050] (1) Using multiple morphological features such as cortical-subcortical regions, we constructed the whole-brain MSNs of adolescent patients with depression and identified abnormal features of MSNs. We clarified the changes in brain activity in the abnormal time-frequency bands of magnetoencephalograms in adolescent patients with depression. We integrated the two with cognitive function assessment data and established a subtype classification model with diagnostic and therapeutic value.

[0051] (2) By analyzing the predictive effects of abnormal MSN characteristics and abnormal time-frequency brain activity on multidimensional cognitive impairment, treatment response, and high-risk behaviors, a prognostic prediction model with clinical translational potential was established.

[0052] (3) It provides theoretical basis and technical means for the individualized and precise diagnosis and treatment of cognitive impairment in adolescent depression. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is the overall technical roadmap for the embodiments of the present invention;

[0054] Figure 2 Schematic diagram of the individual whole-brain MSN construction and analysis process according to an embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0056] Figures 1-2 A specific embodiment of the method for classifying subtypes of cognitive impairment in adolescent depression and predicting prognosis of the present invention is shown, and the steps include:

[0057] S1. All subjects underwent clinical assessment at baseline (i.e., enrollment), and at 1, 6, and 12 months of follow-up. The clinical assessment included assessment of high-risk behavior characteristics. High-risk behaviors included self-harm, suicidal, impulsive, and violent behaviors, which were assessed using the Deliberate Self-Injury Scale, the Suicidal Attitudes Assessment Scale, the Barrett Impulsiveness Scale, and the Violent Behavior Scale (including the psychological violence and physical violence subscales).

[0058] S2. Treatment response assessments were performed on the subjects at 1, 6, and 12 months of follow-up to obtain multi-dimensional treatment responses. The present invention did not interfere with the clinical treatment of the subjects. The treatment plan for the subjects was formulated and adjusted by specialists from the Mental Health Center of West China Hospital, Sichuan University, and included a full course of treatment and adequate dosage of antidepressants, as well as cognitive behavioral therapy or repetitive transcranial magnetic stimulation (rTMS).

[0059] The therapeutic response includes ineffective treatment, partially effective treatment, effective treatment, and clinical remission;

[0060] Follow-up evaluation of the patient's treatment response was conducted by calculating the HAMD score reduction rate = [(post-treatment score - pre-treatment score) / pre-treatment score] * 100%. A HAMD score reduction rate < 25% was considered ineffective, 25% ≤ HAMD score reduction rate < 50% was considered partially effective, 50% ≤ HAMD score reduction rate < 75% was considered effective, and a HAMD score reduction rate > 75% was considered clinical remission.

[0061] S3. Perform MRI data acquisition and data preprocessing at enrollment, 1 month, 6 months, and 12 months of follow-up to obtain morphological indicators of the cortical and subcortical regions. The specific steps are as follows:

[0062] S31. MRI data were acquired using a Siemens Magnetom Prisma 3.0T MRI scanner and a 32-channel phased head coil at the West China Magnetic Resonance Research Center. High-resolution 3D-T1WI structural images and multi-shell diffusion-weighted imaging (DWI) were acquired. Multi-shell DWI was used for diffusion tensor imaging (DTI) and neurite orientation dispersion and density imaging diffusion model analysis.

[0063] High-resolution 3D-T1WI: magnetization prepared rapid gradient echo (MPRAGE) sequence, TR / TE = 1900 / 2.26 ms, flip angle = 9°, matrix = 256 × 256, FOV = 256 × 256 mm 2 , number of layers = 176, layer thickness = 1 mm;

[0064] Multi-slice DWI: spin-echo echo-planarimaging (SE-EPI) sequence, TR / TE = 4000 / 85 ms, flip angle = 90°, SENSE acceleration factor P = 2.5, Multi-Band acceleration factor = 2, FOV = 220 × 220 mm 2 , matrix = 110 × 110, slice thickness = 2 mm, number of slices = 54, b value = 0, 1000, 2000 s / mm 2 , there are 32 diffusion-sensitive gradient directions at each b value;

[0065] S32. Data preprocessing was performed. FreeSurfer software was used for preprocessing high-resolution 3D-T1WI structural images. FreeSurfer's "recon-all" preprocessing process included motion correction, skull stripping, Talairach coordinate transformation, grayscale homogenization, white matter and gray matter segmentation, and topological correction. Multi-shell diffusion-weighted imaging (DWI) data were preprocessed using FSL software. The preprocessing process included eddy current removal and motion correction, brain region segmentation, and the establishment of a voxel-based diffusion tensor model. Poor-quality data were excluded during the preprocessing process to avoid affecting subsequent data analysis.

[0066] S33. For the cortical region, the following morphological features were extracted based on the Glasser atlas: cortical thickness CT, cortical surface area SA, cortical volume CV, local folding index LGI, anisotropy fraction FA, mean diffusivity MD, radial diffusivity RD, axial diffusivity AD, neuronal density index NDI, directional dispersion index ODI, and isotropic volume fraction ISOVF;

[0067] S34. For subcortical regions, the following morphological features were extracted from 17 subcortical nuclei (bilateral hippocampus, amygdala, thalamus, globus pallidus, nucleus accumbens, putamen, caudate nucleus, hypothalamus, and the entire brainstem) based on Aseg coordinates: volume, fractional anisotropy (FA), mean diffusivity (MD), radial diffusivity (RD), axial diffusivity (AD), neuronal density index (NDI), directional dispersion index (ODI), and isotropic volume fraction (ISOVF);

[0068] S4. Magnetoencephalography (MEG) data were collected and preprocessed at enrollment, 1 month, 6 months, and 12 months of follow-up. The MEG data were collected using the Kunmai OPM-MEG device at a frequency of 1000 Hz to collect dynamic changes in brain activity signals under different cognitive test tasks. The cognitive test tasks included facial expression recognition, Go / No-go experiment, N-back experiment, and visual motion experiment. The tasks were projected onto a screen 1.5 m away from the subjects and further refracted onto a mirror approximately 25 cm away from the eyes to achieve a clear visual stimulation effect. All subjects pressed keys to complete the tasks. Each experiment was divided into two parts, each including 100 stimuli, each lasting 3 s, with an interval of 2 s. Reaction time and accuracy were recorded. The average signal-to-noise ratio parameter of the OPM-MEG was 15 fT / Hz, the temporal resolution was 1 ms, and the spatial resolution was 3-5 mm. Participants were required to lie supine on the scanning bed with their heads placed in a 64-channel detector array and to keep their heads as still as possible during the scanning process.

[0069] Brainstorm software was used for data preprocessing, including spatial registration correction, artifact removal, filtering, power spectral density analysis, bad conductor removal, epoch segmentation, and baseline correction.

[0070] S5. This example refers to the prior art method published by Seidlitz (Neuron, 2018) to construct and analyze individual whole-brain MSNs to obtain abnormal MSN features related to multi-dimensional cognitive functions. The specific steps are as follows:

[0071] S51. Constructing a whole-brain morphometric similarity network (MSN) based on the morphological indices of the cortical and subcortical regions obtained in step S3. The specific method is as follows: performing z-value normalization on all morphological indices, and constructing cortical MSNs (360*360), subcortical MSNs (17*17), and cortical-subcortical MSNs (377*377) based on 360 brain regions on both sides of the Glasser atlas and 17 subcortical nuclei with ASEG coordinates.

[0072] Re-identify the morphological index with the greatest contribution. The specific method is: perform importance feature analysis on cortical MSNs, subcortical MSNs, and cortical-subcortical MSNs respectively. That is, delete one index from all morphological indexes each time, recalculate the MSN, and calculate the Pearson correlation coefficient between the MSN with all indexes and the MSN with the index deleted to identify the morphological index with the greatest contribution.

[0073] S52. Modularity Analysis and Longitudinal Follow-up: All subjects' MSNs were divided into modules. Differences in the inner and outer edges of different modules of MSNs between subjects and healthy controls were compared and analyzed. Significantly different edges were functionally annotated using Brain Annotation Toolbox. Pre- and post-treatment changes in the inner and outer edges of different modules of MSNs of subjects were longitudinally followed up, and the correlation between baseline MSNs and follow-up MSNs was analyzed.

[0074] The module division is specifically as follows: cortical MSNs are divided into 7 modules according to Yeo functional network atlas, including visual network, sensorimotor network, dorsolateral attention network, ventral attention network, limbic network, frontoparietal network, and default network;

[0075] Subcortical MSNs were divided into K-means clusters based on the morphological correlation of subcortical nuclei;

[0076] The cortical-subcortical MSNs were divided by K-means clustering based on the morphological correlation between subcortical nuclei and cortical modules;

[0077] S53. Clinical correlation analysis: The relationship between abnormal MSN characteristics and clinical symptoms was analyzed using the LMM linear mixed-effects model. Regularized canonical correlation analysis (rCCA) was used to determine the relationship between abnormal MSN characteristics and multidimensional cognitive function.

[0078] S6. Perform OPM-MEG magnetoencephalography data analysis to obtain functional connectivity changes in abnormal time-frequency bands. The specific steps are as follows:

[0079] S61. Under the stimulation of different cognitive test tasks, the time-frequency analysis method was used to extract spectral features. The time-frequency energy values ​​at the sensor level were compared between groups using the cluster permutation test. The time-frequency segments with significant differences between the groups were selected as regions of interest.

[0080] S62. Comparing the brain region energy values ​​of the subject and the healthy control at the source level under the region of interest, obtaining the brain region involved in the abnormal time-frequency signal, and calculating the abnormal functional connectivity changes between brain regions in the maximum functional connectivity window;

[0081] S7. Perform subtype classification, including:

[0082] Abnormal MSN features related to the subjects' multidimensional cognitive functions and functional connectivity changes in abnormal time-frequency bands were merged and vectorized, centered on the mean, and PCA unsupervised dimensionality reduction was performed using singular value decomposition to extract the principal component values ​​of the number of principal components with an overall model explanation rate greater than or equal to 80%;

[0083] Cluster analysis of the principal component values ​​was performed using the HYDRA heterogeneity discriminant analysis method. The adjusted rand index (ARI) was calculated using 10-fold cross-validation to select the optimal number of clusters. The clinical and brain imaging features of each subtype were compared and clarified to verify the accuracy of the classification. The XGBoost model with Bayesian parameter optimization was used to distinguish the imaging features of each subtype.

[0084] This example also uses an existing multi-center adolescent depression dataset to repeat the above process to verify the reliability and generalizability of the classification model in other datasets;

[0085] S8. Construct a prognostic prediction model, including:

[0086] The subjects' abnormal MSN characteristics related to multidimensional cognitive function, functional connectivity changes in abnormal time-frequency bands, multidimensional treatment response (treatment ineffective, partially effective, effective, clinical remission), and high-risk behaviors (self-harm, suicidal, impulsive, and violent) were analyzed using rCCA regularized canonical correlation analysis to obtain brain network characteristics and their expression levels corresponding to different treatment responses and high-risk behaviors, and to screen out the most clinically significant brain network characteristics.

[0087] The brain network features screened out from the subject training set are added together to obtain a brain network feature sum array, and a prognostic prediction model of the brain network feature sum array for treatment response and high-risk behavior is established based on multiple linear regression, expressed as: Y = β0 + β1X1 + β2X2 + ... + β k X k +∈, where Y is the dependent variable including treatment response score or high-risk behavior risk value, X1, X2, …, X k is the input variable of the brain network feature, that is, the feature in the sum array of the brain network features, β0 is the intercept term, β1, β2, ..., β k is the regression coefficient, ∈ is the error term;

[0088] S9. For the subject to be predicted, the method in step S8 is used to calculate the brain network feature sum array, and then the sum array is input into the prognosis prediction model obtained in step S8 to obtain the predicted values ​​of treatment response and high-risk behavior.

[0089] This example also performed model validation: the data collected by the project were included as an internal dataset, and the existing multi-center adolescent depression dataset was integrated as an external dataset, and intra- and inter-dataset cross-validation was performed to verify the accuracy and generalizability of the prognostic prediction model.

[0090] The above embodiments of the present invention are merely examples for illustrating the present invention and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations and modifications can be made based on the above description. It is not possible to enumerate all embodiments here. Any obvious variations or modifications arising from the technical solution of the present invention remain within the scope of protection of the present invention.

Claims

1. A method for classifying subtypes of cognitive impairment in adolescent depression and predicting prognosis, characterized in that: The steps include: S1. Conduct clinical assessments on subjects at baseline (i.e., enrollment) and during follow-up, including assessment of high-risk behavioral characteristics; S2. Evaluate the treatment response of the subjects during follow-up to obtain multi-dimensional treatment response; S3. MRI data were collected at the time of enrollment and follow-up to obtain morphological indicators of the cortical and subcortical regions; S4. Performing magnetoencephalography (MEG) data collection and data preprocessing on subjects during enrollment and follow-up. The MEG data collection includes collecting dynamic changes in brain activity signals under different cognitive test tasks. S5. Construct and analyze individual whole-brain MSNs to obtain abnormal MSN characteristics related to multi-dimensional cognitive functions; S6. Analyze magnetoencephalography data to obtain functional connectivity changes in abnormal time-frequency bands; S7. Subtype classification is performed using abnormal MSN features related to the subjects' multidimensional cognitive functions and functional connectivity in abnormal time-frequency bands; S8. Construct a prognostic prediction model, including: The subjects' abnormal MSN features related to multidimensional cognitive function, functional connectivity changes in abnormal time-frequency bands, multidimensional treatment responses, and high-risk behaviors were subjected to rCCA regularized canonical correlation analysis to obtain the brain network features and their expression levels corresponding to different treatment responses and high-risk behaviors, and screen out the most clinically significant brain network features. The brain network features screened out from the subject training set are added to obtain a brain network feature sum array, and a prognostic prediction model of the brain network feature sum array for treatment response and high-risk behavior is established based on multiple linear regression; S9. Calculate the sum array of brain network features for the subject to be predicted, and then input the sum array into the prognosis prediction model to obtain the predicted values ​​of treatment response and high-risk behavior.

2. The method for classifying subtypes of cognitive impairment in adolescent depression and predicting prognosis according to claim 1, characterized in that: The high-risk behaviors include self-harm, suicidal, impulsive, and violent behaviors; The therapeutic response includes ineffective treatment, partially effective treatment, effective treatment, and clinical remission; The treatment response assessment described in step S2 is specifically determined by calculating the HAMD score reduction rate = [(post-treatment score - pre-treatment score) / pre-treatment score] * 100%. A HAMD score reduction rate <25% indicates ineffective treatment, 25% ≤ HAMD score reduction rate <50% indicates partially effective treatment, 50% ≤ HAMD score reduction rate <75% indicates effective treatment, and a HAMD score reduction rate >75% indicates clinical remission.

3. The method for classifying subtypes of cognitive impairment in adolescent depression and predicting prognosis according to claim 1, characterized in that: The step S3 is specifically as follows: S31. Perform high-resolution 3D-T1WI structural images and multi-shell diffusion-weighted imaging (DWI). Use multi-shell DWI for diffusion tensor imaging (DTI) and neurite direction dispersion and density imaging diffusion model analysis. S32. Perform data preprocessing. Preprocessing of high-resolution 3D-T1WI structural images includes motion correction, skull stripping, Talairach coordinate transformation, grayscale homogenization, white matter and gray matter segmentation, and topological correction. Preprocessing of multi-slice diffusion-weighted imaging (DWI) data includes eddy current removal and motion correction, brain region segmentation, and establishment of a voxel-based diffusion tensor model. S33. For the cortical region, the following morphological features were extracted based on the Glasser atlas: cortical thickness CT, cortical surface area SA, cortical volume CV, local folding index LGI, anisotropy fraction FA, mean diffusivity MD, radial diffusivity RD, axial diffusivity AD, neuronal density index NDI, directional dispersion index ODI, and isotropic volume fraction ISOVF; S34. For the subcortical regions, the following morphological features were extracted from 17 subcortical nuclei - bilateral hippocampus, amygdala, thalamus, globus pallidus, nucleus accumbens, putamen, caudate nucleus, hypothalamus and the entire brainstem - based on the Aseg coordinates: volume, fractional anisotropy FA, ​​mean diffusivity MD, radial diffusivity RD, axial diffusivity AD, neuronal density index NDI, directional dispersion index ODI and isotropic volume fraction ISOVF.

4. The method for classifying subtypes of cognitive impairment in adolescent depression and predicting prognosis according to claim 1, characterized in that: In step S4, the cognitive test tasks include facial expression recognition, Go / No-go experiment, N-back experiment, and visual motion experiment, and the data preprocessing includes spatial registration correction, artifact removal, filtering, power spectral density analysis, bad conductor removal, epoch segmentation, and baseline correction.

5. The method for classifying subtypes of cognitive impairment in adolescent depression and predicting prognosis according to claim 3, characterized in that: The step S5 is specifically as follows: S51, constructing a whole-brain morphometric similarity network (MSN) based on the morphological indices of the cortex and subcortical regions obtained in step S3, and identifying the morphological indices with the greatest contribution; S52. All subjects' MSNs were divided into modules. Differences in the inner and outer edges of different modules of MSNs between subjects and healthy controls were compared and analyzed. Functional annotation analysis was performed on the edges with significant differences. Changes in the inner and outer edges of different modules of MSNs of subjects were longitudinally followed up before and after treatment, and the correlation between baseline MSNs and follow-up MSNs was analyzed. S53. The LMM linear mixed-effects model was used to analyze the relationship between abnormal MSN characteristics and clinical symptoms, and the rCCA regularized canonical correlation analysis was used to determine the relationship between abnormal MSN characteristics and multidimensional cognitive functions.

6. The method for classifying subtypes of cognitive impairment in adolescent depression and predicting prognosis according to claim 5, characterized in that: The specific method of constructing the whole-brain morphometric similarity network MSN in step S51 is: All morphological indices were z-value standardized, and cortical MSNs, subcortical MSNs, and cortical-subcortical MSNs were constructed based on 360 brain regions bilaterally in the Glasser atlas and 17 subcortical nuclei in the Aseg coordinates; The specific method of identifying the morphological index with the greatest contribution in step S51 is: Importance feature analysis was performed on cortical MSNs, subcortical MSNs, and cortical-subcortical MSNs, respectively. That is, one indicator was deleted from all morphological indicators each time, the MSNs were recalculated, and the Pearson correlation coefficient between the MSNs with all indicators and the MSNs after the indicators were deleted was calculated to identify the morphological indicator with the greatest contribution.

7. The method for classifying subtypes of cognitive impairment in adolescent depression and predicting prognosis according to claim 6, characterized in that: The module division in step S52 is specifically as follows: The cortical MSNs were divided into seven modules according to the Yeo functional network atlas, including the visual network, sensorimotor network, dorsolateral attention network, ventral attention network, limbic network, frontoparietal network, and default network; Subcortical MSNs were divided into K-means clusters based on the morphological correlation of subcortical nuclei; The cortical-subcortical MSNs were divided by K-means clustering based on the morphological correlation between subcortical nuclei and cortical modules.

8. The method for classifying subtypes of cognitive impairment in adolescent depression and predicting prognosis according to claim 1, characterized in that: The step S6 is specifically as follows: S61. Under the stimulation of different cognitive test tasks, the time-frequency analysis method was used to extract spectral features. The time-frequency energy values ​​at the sensor level were compared between groups using the cluster permutation test. The time-frequency segments with significant differences between the groups were selected as regions of interest. S62. By comparing the brain area energy values ​​of the subjects and healthy controls at the source level under the region of interest, the brain areas involved in the abnormal time-frequency signals are obtained, and the abnormal functional connectivity changes between brain areas in the maximum functional connectivity window are calculated.

9. The method for classifying subtypes of cognitive impairment in adolescent depression and predicting prognosis according to claim 1, characterized in that: The step S7 is specifically as follows: Abnormal MSN features related to the subjects' multidimensional cognitive functions and functional connectivity changes in abnormal time-frequency bands were merged and vectorized, centered on the mean, and PCA unsupervised dimensionality reduction was performed using singular value decomposition to extract the principal component values ​​of the number of principal components with an overall model explanation rate greater than or equal to 80%; The principal component values ​​were clustered using the HYDRA heterogeneity discriminant analysis method. The adjusted Rand index (ARI) was calculated using 10-fold cross-validation to select the optimal number of clusters. The clinical and brain imaging features of each subtype were compared and clarified to verify the accuracy of the classification. The XGBoost model with Bayesian parameter optimization was used to distinguish the imaging features of each subtype.

10. The method for classifying subtypes of cognitive impairment in adolescent depression and predicting prognosis according to claim 1, characterized in that: The expression of the prognosis prediction model in step S8 is: Y=β0+β1X1+β2X2+…+β k X k +∈, where Y is the dependent variable including treatment response score or high-risk behavior risk value, X1, X2, …, X k is the input variable of the brain network feature, that is, the feature in the sum array of the brain network features, β0 is the intercept term, β1, β2, ..., β k is the regression coefficient, and ∈ is the error term.

Images