Method, device and application of resting-state brain functional connectivity in biological typing of depression

Abstract

The application relates to a method, device and application of a depression biological typing method based on resting-state brain function connection, which can objectively type depression biological subtypes, establish a corresponding relationship between typing and treatment response, realize individual optimal treatment strategy recommendation for patients, improve the treatment response rate and reduce the clinical trial and error cost. The method comprises the following steps: (1) data acquisition and preprocessing; (2) definition of biological subtype clustering and characteristics; (3) subtype verification and standardization; and (4) rapid typing of new patients.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and more particularly to a method for biotyping depression based on resting-state brain functional connectivity, a device for biotyping depression based on resting-state brain functional connectivity, and its application. Background Technology

[0002] Major Depressive Disorder (MDD) is a highly heterogeneous mental disorder with a global prevalence of about 15%, but clinical treatment faces serious bottlenecks: only 30% of patients achieve complete remission with first-line antidepressants, nearly 50% develop treatment-resistant depression, and traditional symptom-based diagnostic classification cannot effectively guide treatment selection, resulting in high costs of trial and error in treatment and a heavy disease burden.

[0003] Current research on depression subtyping has the following key shortcomings: The classification is based on a single symptom scale or a single modality of data (such as genetic or local imaging features only), which fails to capture systemic abnormalities in the whole brain functional network. Poor repeatability: Subtypes derived from a single cohort are difficult to validate in independent populations and lack cross-center and cross-device universality; Insufficient clinical translation: The existing subtypes have not established a clear correspondence with various treatment methods (drugs, physical therapy, etc.), and cannot directly guide clinical decision-making; Low classification efficiency: Traditional clustering methods require a complete re-execution of the clustering process for new patients, which cannot achieve rapid individualized matching.

[0004] Therefore, there is an urgent need to develop a bio-subtype classification technology for depression that is based on whole-brain functional characteristics, is repeatable, and can directly guide the selection of multiple treatments, in order to solve the current clinical dilemma of "one-size-fits-all" treatment for depression. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the technical problem to be solved by this invention is to provide a resting-state brain functional connectivity biotyping method for depression, which can objectively classify biological subtypes of depression and establish a correspondence between "typing and treatment response" to achieve personalized optimal treatment strategy recommendations for patients, improve treatment response rate, and reduce clinical trial and error costs.

[0006] The technical solution of this invention is: a method for biotyping depression based on resting-state brain functional connectivity, comprising the following steps: (1) Data acquisition and preprocessing: Resting-state functional magnetic resonance imaging (rs-fMRI) data of subjects were collected, covering a multi-center cohort with different scanning equipment, including patients with depression and healthy controls; the rs-fMRI data were standardized and preprocessed, including head motion correction, time-level correction, spatial standardization, denoising and ComBat harmonization; the blood oxygen level dependent on BOLD signal time series of n brain regions of the whole brain was extracted, the Pearson correlation coefficient between brain regions was calculated, an n×n whole brain functional connectivity matrix was generated, and n(n-1) / 2 functional connectivity features of the upper triangular region were extracted; (2) Define biological subtype clusters and features: The improved non-negative matrix factorization (NMF) algorithm is used to perform bi-cluster analysis on the preprocessed functional connectivity features. The threshold is set for the non-smoothness constraint parameter of the improved NMF algorithm. The optimal number of clusters is determined to be 3 subtypes by the co-table correlation coefficient and the sum of squared residuals. Non-smoothness constraints are used to improve feature sparsity during the clustering process. The optimal number of clusters is determined by 10 times of 10-fold cross-validation. Define the core functional connectivity features of each subtype. (3) Subtype validation and standardization: The repeatability of subtypes is validated in an independent validation queue. A subtype classifier is constructed using a linear support vector machine (SVM). A subtype standardization template is established, and the core functional connection feature matrices of the three subtypes are saved as subtype reference templates. (4) Rapid subtyping of new patients: The rs-fMRI data of new patients are standardized and preprocessed, including head motion correction, time layer correction, spatial standardization, denoising and ComBat harmonization, to generate a whole brain functional connectivity matrix; the cosine similarity between the whole brain functional connectivity matrix of new patients and the standardized templates of the three subtypes is calculated, and the new patients are assigned to the subtype with the highest similarity, which is the one with the smallest cosine distance.

[0007] The beneficial technical effects of the present invention are as follows:

[0008] 1. Objectivity and repeatability of typing: Based on whole-brain functional connectivity features and improved NMF algorithm, stable typing is achieved across centers and devices, with an independent queue verification accuracy of ≥83%, solving the problems of strong subjectivity and poor repeatability of traditional typing.

[0009] 2. Improved targeted treatment: Establishing a clear correspondence between "subtype-treatment" significantly improves the response rate of drug treatment for subtype 1 patients and ensures precise matching of ECT treatment for subtype 2 patients, avoiding blind trial and error. It is expected that the overall treatment remission rate will be increased by more than 30%.

[0010] 3. High clinical applicability: New patients can complete the assessment through a quick process of "feature matching - subtype determination - treatment recommendation" (without the need for re-clustering), which is suitable for different populations (adults, adolescents, post-stroke depression) and can be directly integrated into the clinical imaging workstation.

[0011] 4. Standardization and Scalability: Provides unified classification templates and treatment recommendation rules to promote the transformation of depression classification from "experience-dependent" to "objective standards" and support multi-center clinical promotion.

[0012] 5. Significant cost-effectiveness: Reduces the waste of medical resources and the economic burden on patients caused by ineffective treatments, and lowers the incidence of treatment-resistant depression.

[0013] It also provides a biotyping device for depression based on resting-state brain functional connectivity, which includes: The data acquisition and preprocessing module is configured to collect resting-state functional magnetic resonance imaging (rs-fMRI) data from subjects, covering a multi-center cohort with different scanning equipment, including patients with depression and healthy controls; it performs standardized preprocessing on the rs-fMRI data, including head motion correction, temporal correction, spatial standardization, denoising, and ComBat harmonization; it extracts the time series of blood oxygenation level-dependent BOLD signals from n brain regions of the whole brain, calculates the Pearson correlation coefficient between brain regions, generates an n×n whole-brain functional connectivity matrix, and extracts n(n-1) / 2 functional connectivity features from the upper triangular region; A biological subtype clustering and feature module is defined, and its configuration employs an improved nonnegative matrix factorization (NMF) algorithm to perform bi-cluster analysis on preprocessed functional connectivity features. The non-smoothness constraint parameter of the improved NMF algorithm is set with a threshold. The optimal number of clusters (3 subtypes) is determined by the co-integer correlation coefficient and the sum of squared residuals. Non-smoothness constraints are used during clustering to improve feature sparsity. The optimal number of clusters is determined through 10 rounds of 10-fold cross-validation. The core functional connectivity features of each subtype are defined. The subtype validation and standardization module is configured to validate the reproducibility of subtypes in an independent validation queue. It constructs a subtype classifier using a linear support vector machine (SVM) and establishes a subtype standardization template, storing the core functional connection feature matrices of the three subtypes as a subtype reference template. The new patient rapid subtyping module is configured to perform standardized preprocessing on the rs-fMRI data of new patients, including head motion correction, temporal layer correction, spatial standardization, denoising, and ComBat harmonization, to generate a whole-brain functional connectivity matrix. It calculates the cosine similarity between the new patient's whole-brain functional connectivity matrix and the standardized templates of the three subtypes, and assigns the new patient to the subtype with the highest similarity, where the highest similarity is the smallest cosine distance.

[0014] It also provides the application of a resting-state brain functional connectivity biotyping method for depression, which can be used for the treatment of depression. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method for biotyping depression based on resting-state brain functional connectivity according to the present invention.

[0016] Figure 2 This is a core functional connectivity feature map (including the distribution of inter-network connection weights) for three biological subtypes of depression.

[0017] Figure 3 This is a schematic diagram of the treatment recommendation system interface, which includes an image input port, a subtype output module, and a treatment suggestion display module.

[0018] Figure 4 The nonnegative factor decomposition of the functional connectivity matrix of depressed subjects and the validation of subtypes on the ECT treatment dataset are shown.

[0019] Figure 5 The correlation coefficients and sums of squared residuals for different numbers of clusters are shown.

[0020] Figure 6 The top 100 functional connectivity patterns of the three subtypes of depression in the discovery set are shown.

[0021] Figure 7 The correlation between frontoparietal network connectivity subtypes and HAMA is shown.

[0022] Figure 8 The confusion matrix for the three subtypes of depression is shown.

[0023] Figure 9 The correlation coefficients and sum of squared residuals for the common representation of subtypes on the validation set are shown.

[0024] Figure 10 The top 100 functional connectivity patterns of the three subtypes of depression on the validation set are shown.

[0025] Figure 11 The spatial consistency of each subtype of the discovery set and the verification set is shown.

[0026] Figure 12 The correlation statistics between frontoparietal connectivity subtypes and anxiety scores on the validation set are shown.

[0027] Figure 13 The generalization classification accuracy on the validation and discovery sets is shown.

[0028] Figure 14 The study shows a comparison of the differences in antidepressant efficacy among different subtypes on two independent ECT treatment datasets. Detailed Implementation

[0029] like Figure 1 As shown, this method for biotyping depression based on resting-state brain functional connectivity includes the following steps: (1) Data acquisition and preprocessing: Resting-state functional magnetic resonance imaging (rs-fMRI) data of subjects were collected, covering a multi-center cohort with different scanning equipment, including patients with depression and healthy controls; the rs-fMRI data were standardized and preprocessed, including head motion correction, time-level correction, spatial standardization, denoising and ComBat harmonization; the blood oxygen level dependent on BOLD signal time series of n brain regions of the whole brain was extracted, the Pearson correlation coefficient between brain regions was calculated, an n×n whole brain functional connectivity matrix was generated, and n(n-1) / 2 functional connectivity features of the upper triangular region were extracted; (2) Define biological subtype clusters and features: The improved non-negative matrix factorization (NMF) algorithm is used to perform bi-cluster analysis on the preprocessed functional connectivity features. The threshold is set for the non-smoothness constraint parameter of the improved NMF algorithm. The optimal number of clusters is determined to be 3 subtypes by the co-table correlation coefficient and the sum of squared residuals. Non-smoothness constraints are used to improve feature sparsity during the clustering process. The optimal number of clusters is determined by 10 times of 10-fold cross-validation. Define the core functional connectivity features of each subtype. (3) Subtype validation and standardization: The repeatability of subtypes is validated in an independent validation queue. A subtype classifier is constructed using a linear support vector machine (SVM). A subtype standardization template is established, and the core functional connection feature matrices of the three subtypes are saved as subtype reference templates. (4) Rapid subtyping of new patients: The rs-fMRI data of new patients are standardized and preprocessed, including head motion correction, time layer correction, spatial standardization, denoising and ComBat harmonization, to generate a whole brain functional connectivity matrix; the cosine similarity between the whole brain functional connectivity matrix of new patients and the standardized templates of the three subtypes is calculated, and the new patients are assigned to the subtype with the highest similarity, which is the one with the smallest cosine distance.

[0030] The beneficial technical effects of the present invention are as follows:

[0031] 1. Objectivity and repeatability of typing: Based on whole-brain functional connectivity features and improved NMF algorithm, stable typing is achieved across centers and devices, with an independent queue verification accuracy of ≥83%, solving the problems of strong subjectivity and poor repeatability of traditional typing.

[0032] 2. Improved targeted treatment: Establishing a clear correspondence between "subtype-treatment" significantly improves the response rate of drug treatment for subtype 1 patients and ensures precise matching of ECT treatment for subtype 2 patients, avoiding blind trial and error. It is expected that the overall treatment remission rate will be increased by more than 30%.

[0033] 3. High clinical applicability: New patients can complete the assessment through a quick process of "feature matching - subtype determination - treatment recommendation" (without the need for re-clustering), which is suitable for different populations (adults, adolescents, post-stroke depression) and can be directly integrated into the clinical imaging workstation.

[0034] 4. Standardization and Scalability: Provides unified classification templates and treatment recommendation rules to promote the transformation of depression classification from "experience-dependent" to "objective standards" and support multi-center clinical promotion.

[0035] 5. Significant cost-effectiveness: Reduces the waste of medical resources and the economic burden on patients caused by ineffective treatments, and lowers the incidence of treatment-resistant depression.

[0036] Preferably, in step (1), the exclusion criteria for head motion correction are: translation > 3mm or rotation > 3°; spatial normalization is based on the AAL template, the target space is the MNI space, and the voxel size is 3mm×3mm×3mm; denoising is to remove cerebrospinal fluid / vascular signal interference, and also includes the removal of white matter signals and linear trend regression; ComBat harmonization is to eliminate the center / device effect and retain the biological trends of age, sex, and diagnosis.

[0037] Preferably, in step (2), the threshold for the non-smoothness constraint parameter is set to sparseness=0.5.

[0038] Preferably, in step (2), the three subtypes include: Subtype 1: Prefrontal-parietal hyperconnection features, the core of which involves enhanced functional connectivity between the default mode network (DMN) and the ventral attention network (VAN) and the somatic motor network (SMN) and the frontoparietal network (FPN), and related semantic processing behaviors; Subtype 2: Low connectivity features between the striatum, occipital lobe, and prefrontal lobe, with the core being weakened functional connectivity between the frontoparietal network (FPN) and the visual network (VIS) and the subcortical network (SCN) and the visual network (VIS), resulting in abnormalities in associative memory retrieval and emotional processing. Subtype 3: Abnormal features in subcortical-occipital connectivity, mainly involving enhanced connectivity between the subcortical network SCN and the visual network VIS, weakened connectivity between the limbic network LIM and the somatic motor network SMN, and defects in associated reward processing.

[0039] Preferably, in step (4), the matching threshold for cosine similarity is 0.6. When the similarity is lower than the threshold, a prompt message for supplementing the collected data is output.

[0040] Preferably, the method further includes step (5), establishing a subtype-treatment response correspondence database based on independent treatment cohort data: subtype 1 uses drug therapy; subtype 2 uses electroconvulsive therapy (ECT); subtype 3 uses drug therapy combined with reward pathway regulation intervention, and the alternative strategy is traditional Chinese medicine combined with physical therapy.

[0041] Preferably, in step (5), all treatments are fine-tuned based on the patient's age and comorbidities to ensure clinical applicability.

[0042] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium. When executed, the program includes the steps of the methods of the above embodiments. The storage medium can be ROM / RAM, magnetic disk, optical disk, memory card, etc. Therefore, corresponding to the method of the present invention, the present invention also includes a resting-state brain functional connectivity biotyping device for depression. This device is typically represented in the form of functional modules corresponding to the steps of the method. The device includes: The data acquisition and preprocessing module is configured to collect resting-state functional magnetic resonance imaging (rs-fMRI) data from subjects, covering a multi-center cohort with different scanning equipment, including patients with depression and healthy controls; it performs standardized preprocessing on the rs-fMRI data, including head motion correction, temporal correction, spatial standardization, denoising, and ComBat harmonization; it extracts the time series of blood oxygenation level-dependent BOLD signals from n brain regions of the whole brain, calculates the Pearson correlation coefficient between brain regions, generates an n×n whole-brain functional connectivity matrix, and extracts n(n-1) / 2 functional connectivity features from the upper triangular region; A biological subtype clustering and feature module is defined, and its configuration employs an improved nonnegative matrix factorization (NMF) algorithm to perform bi-cluster analysis on preprocessed functional connectivity features. The non-smoothness constraint parameter of the improved NMF algorithm is set with a threshold. The optimal number of clusters (3 subtypes) is determined by the co-integer correlation coefficient and the sum of squared residuals. Non-smoothness constraints are used during clustering to improve feature sparsity. The optimal number of clusters is determined through 10 rounds of 10-fold cross-validation. The core functional connectivity features of each subtype are defined. The subtype validation and standardization module is configured to validate the reproducibility of subtypes in an independent validation queue. It constructs a subtype classifier using a linear support vector machine (SVM) and establishes a subtype standardization template, storing the core functional connection feature matrices of the three subtypes as a subtype reference template. The new patient rapid subtyping module is configured to perform standardized preprocessing on the rs-fMRI data of new patients, including head motion correction, temporal layer correction, spatial standardization, denoising, and ComBat harmonization, to generate a whole-brain functional connectivity matrix. It calculates the cosine similarity between the new patient's whole-brain functional connectivity matrix and the standardized templates of the three subtypes, and assigns the new patient to the subtype with the highest similarity, where the highest similarity is the smallest cosine distance.

[0043] It also provides the application of a resting-state brain functional connectivity biotyping method for depression, which can be used for the treatment of depression.

[0044] The present invention will now be described in more detail.

[0045] This study primarily utilizes the resting-state functional magnetic resonance imaging (fMRI) dataset from REST-meta-MDD (http: / / rfmri.org / REST-meta-MDD). Participants included 925 patients with depression from the first batch of data in the REST-meta-MDD dataset, and 120 patients with depression treated with ECT collected by international collaborating teams (UNM, UCLA). Following the quality control guidelines of REST-meta-MDD et al., participants with incomplete features or fewer than 15 depression samples were removed. The final dataset included 851 participants from 16 sites in the REST-meta-MDD dataset, 76 participants from the UNM site, and 44 participants from the UCLA site. Participants from 15 sites in the REST-meta-MDD dataset were used as the discovery dataset, and participants from the remaining site were used as the validation set. The two ECT datasets were used to validate the clinical response rate to antidepressant treatment. Clinical information for all participants included age, sex, disease duration, HDRS, HAMA, whether it was first episode (FED), and basic information on medication use. Symptom severity was assessed using the HDRS scale, and anxiety symptoms were assessed using the Hamilton Anxiety Rating Scale (HAMA). HAMA assessment was not performed at the UNM and UCLA sites. This study was supported by the relevant ethics committee, and all participants signed informed consent forms before participating in the experiment.

[0046] Data acquisition and preprocessing

[0047] 1) REST-meta-MDD

[0048] Data preprocessing for fMRI from 16 sites in the REST-meta-MDD dataset was performed using Data Processing Assistant for Resting-State fMRI (DPARSF, http: / / rfmri.org / DPARSF) software according to a standard preprocessing workflow, including removing data from the first 10 time points, inter-slice correction, head motion correction, spatial normalization, 0.01–0.1 Hz bandpass filtering, and 4 mm half-width Gaussian kernel smoothing. Furthermore, 24 head motion parameters, white matter signal, cerebrospinal fluid signal, and the mean whole-brain signal were removed from the raw signal as covariates for each subject. To reduce the influence of head motion, subjects with a mean frame-by-frame displacement greater than 0.2 mm were excluded. Detailed acquisition parameters and preprocessing steps for each site are described in detail in the article by Yan et al.

[0049] 2) ECT treatment dataset

[0050] A total of 120 participants underwent ECT treatment for depression. Participants were required to not have other mental illnesses or neurodegenerative diseases, not have alcohol or drug dependence, and not have a history of head trauma.

[0051] Preprocessing of rsfMRI data from both sites was performed using SPM12, including slice correction, head motion correction, spatial normalization, 0.01–0.15 Hz bandpass filtering, 6 mm half-width Gaussian kernel smoothing, and covariate regression. Six head motion parameters, white matter signal, cerebrospinal fluid signal, and the mean of whole-brain signal were removed from the raw signals as covariates.

[0052] The ECT treatment process and parameters for depression are summarized as follows: In the initial stage of ECT treatment for depression, a right-sided unilateral electrode was placed with an ultra-short pulse width (0.3 ms), followed by 6 × ST intervals. Then, based on efficacy response assessment, electrodes for unresponsive patients were switched to bitemporal electrodes. During treatment, sufficient seizure duration (EEG seizure duration >25 seconds) was maintained, with treatment performed three times a week. If the subject consistently did not respond to treatment, treatment was discontinued. During treatment, patients were oxygenated using disposable bags and masks, and appropriate induction (methoxycital or etomidate) and relaxation (succinylcholine) were administered, while blood pressure, pulse, and blood oxygen saturation were monitored in real time. Throughout the ECT treatment process, clinical psychiatrists assessed the subjects' symptoms and clinical efficacy using the HDRS scale. All subjects underwent HDRS assessments before starting ECT treatment and within one week after completing ECT treatment. After ECT treatment, an HDRS score less than 25% of the pre-ECT score was considered a partial ECT response, less than 50% was considered a full ECT response, and an HDRS score <7 after treatment was considered ECT remission. The clinical symptom response and remission statistics before and after treatment at both sites are shown in Table 1.

[0053] Table 1

[0054] Brain functional connectivity features

[0055] For all participants, this study divided the whole-brain fMRI signal into n brain regions based on the AAL template. Pearson correlation was used to calculate the correlation between the average time-series signals of any two brain regions, resulting in n(n-1) / 2 functional connectivity features for each participant. Fisher-Z transform was then applied to the functional connectivity features, and a linear regression model was further used to remove the influence of age, sex, location, and average frame-by-frame displacement covariates.

[0056] Cluster analysis

[0057] To identify subtypes of depression, this study employed nonnegative factorization (NMF) in R to decompose the functional connectivity matrix of the participants, enabling simultaneous clustering of features and participants. This study used a non-smooth NMF decomposition method to enhance the sparsity of the clustering results. Figure 4As shown, on the discovery dataset, the functional connectivity matrix Y (6670 functional connectivity features * number of participants) was subjected to nonnegative matrix factorization, Y = W * H, where the basis matrix W (6670 functional connectivity features * r) clusters the functional connectivity into r classes, corresponding to different functional connectivity patterns; the coefficient weight matrix H (r * number of participants) represents the clustering results for all participants, i.e., the weight of each participant on the r functional connectivity patterns. Based on the coefficient weight matrix, each participant is assigned to the subtype that best matches the r subtypes. The optimal number of clusters is determined based on the cophenetic correlation coefficient and the residual sum of squares (RSS) index, which characterize the stability of the clustering results. To find the optimal number of clusters, this study sets the number of clusters to 2 to 7, repeats the NMF decomposition 30 times for each cluster number, calculates the cophenetic correlation coefficient and the residual sum of squares, and selects the optimal number of clusters. After determining the optimal number of clusters, the NMF decomposition is repeated 50 times to obtain the final clustering results. After obtaining the subtype category for each subject, the specific connections of each subtype are characterized by the first 100 features in the basis matrix.

[0058] Clinical statistical analysis and subtype classification studies

[0059] After obtaining the clustering results for each subtype, the study further analyzed the differences and correlations in clinical symptoms among the subtypes, controlling for age, gender, and site covariates. To further test whether the functional connectivity matrix could classify and identify the three subtypes, the study used linear support vector machines and 10-fold cross-validation to classify the three subtypes on the discovery and validation sets respectively.

[0060] Validation on independent datasets

[0061] To verify the reliability and stability of subtype identification, this study conducted repeated experiments on an independent validation set of 651 participants with depression, obtaining the best clustering results for depression on the validation set. Pearson correlation was used to measure the spatial consistency of functional connectivity patterns of corresponding subtypes on the discovery and validation sets. To further verify the consistency of subtype identification across the two independent datasets, this study performed cross-dataset classification, directly applying the classification diagnostic model from the discovery set to the validation set, and vice versa.

[0062] Efficacy validation on ECT treatment dataset

[0063] To investigate the differences in response rates of different connectivity patterns in antidepressant therapy, this study investigated the differences in response rates among different subtypes on an ECT antidepressant treatment dataset targeting the temporal lobe cortex. After obtaining the representations of the three subtypes on the discovery and validation sets respectively, the average functional connectivity representation of each subtype on the two datasets was calculated, and the top 100 functional connections with the largest contributions were used as the final specific representations of each subtype. Figure 4 As shown, ECT treatment subjects at two independent sites were connected to the cosine distance of each subject's function to the subtype-specific representation, and then matched to the most similar subtype. Finally, the efficacy response of each subtype in the two ECT datasets was statistically analyzed, including partial response rate (HDRS reduction of 25-50%) and complete response rate (HDRS reduction of >50%).

[0064] Depression Subtype Identification Based on Nonnegative Matrix Factorization

[0065] like Figure 5 As shown, the study statistically analyzed the co-representational correlation coefficients and residual sums of squares for subtype identification with cluster numbers ranging from 2 to 7. Based on the optimal cluster number criterion, the first point where the co-representational correlation coefficient decreased the most with the change in the number of clusters was selected, i.e., the optimal cluster number was 3; and the last point where the residual sum of squares decreased the fastest was selected, i.e., the optimal cluster number was 3. Therefore, the optimal cluster number for the depression subtype was finally determined to be 3.

[0066] The basis matrix and coefficient matrix obtained from NMF decomposition represent the functional connectivity representation patterns of each subtype and the classification results of depressed subjects, respectively. Subtype classification was based on the highest coefficients for each subject across the three representation patterns. Figure 6 The top 100 most contributing connections for each subtype were shown. The results showed that the functional connections of subtype 1 were mainly located in the frontoparietal network, with key nodes including the frontofrontal cortex, medial superior frontal gyrus, angular gyrus, and preclavicular lobe; the functional connections of subtype 2 were mainly located in the frontoparietal network and the subcortical nucleus-parietal lobe connection, with key nodes including the dorsolateral prefrontal lobe, middle frontal gyrus, medial superior frontal gyrus, superior occipital gyrus, and clavicular lobe; and the functional connections of subtype 3 were mainly located in the contralateral loss of connectivity of each network, mainly including the cerebellum, subcortical nuclei, and occipital cortex.

[0067] Clinical correlation analysis and subtype classification results

[0068] The study analyzed the differences in gender, age, disease duration, HDRS, HAMA, and medication use among the three subtypes using ANOVA (Table 2). The results showed significant gender differences among the three subtypes of depression. p = 0.0076), and no significant differences were found in other clinical characteristics.

[0069] Table 2

[0070] The experiment further analyzed the correlation between each subtype and HDRS and HAMA, and the results showed that the frontoparietal network connectivity subtype was significantly correlated with HAMA score. r = 0.20, p = 0.025, such as Figure 7 ).

[0071] Based on functional connectivity features, a linear support vector machine was used to perform 10-fold cross-validation classification on the three subtypes. The results showed that the classification accuracy of the three subtypes was 94.0%, and the confusion matrix of the three subtypes was as follows. Figure 8 As shown.

[0072] Reproducible studies of depression subtypes

[0073] To verify the reproducibility of the depression subtype clustering results, this study repeated the same analysis procedure on the independent validation set of depression data from 16 sites. Based on the co-occurrence correlation coefficient and the sum of squared residuals, the number of the three subtypes (e.g., Figure 9 ).

[0074] Similarly, after NMF decomposition on the validation set, three subtypes were obtained. The top 100 functional connections with the highest contribution in each subtype are as follows: Figure 10 As shown.

[0075] The study used Pearson correlation coefficient to further analyze the spatial correlation of corresponding subtypes on the validation and discovery sets. The results showed that the spatial connectivity patterns of corresponding subtypes were significantly correlated on both datasets (frontoparietal subtype: r = 0.98, p <1×10 -5 Forehead-occipital connection subtype: r = 0.95, p <1×10 -5 Contralateral disconnection subtype: r = 0.84, p <1×10 -5 , Figure 11 By comparing the repetitive functional connections in the specific characterization connections of the corresponding subtypes on the discovery and validation sets, it was found that the main functional connections in subtype 1 were located in the frontal-parietal lobe connection, followed by the limbic system-frontal / parietal lobe connection; the connections in subtype 2 mainly included the frontal-occipital lobe connection and the subcortical nucleus-parietal / occipital lobe connection; the connections in subtype 3 mainly included contralateral loss of connectivity between the cerebellum, occipital lobe, frontal lobe, and subcortical nuclei. Figure 11 ).

[0076] As shown in Table 3, no significant differences were observed among the subtypes in terms of age, disease duration, HDRS, HAMA, and the medication used on the validation set.

[0077] Table 3

[0078] Consistent with the discovery set, frontoparietal connectivity subtypes and HAMA scores were significantly correlated on the validation set. r = 0.20, p = 0.014, Figure 12 ).

[0079] In the subtype classification study on the validation set, the results showed that the classification accuracy of the three subtypes was 94.2%. To verify the generalization ability of cross-dataset subtype recognition, cross-dataset classification recognition was performed on the discovery set and validation set respectively. The results showed that the accuracy of generalization from the discovery set to the validation set was 91.2%, and the accuracy of generalization from the validation set to the discovery set was 87.2% (e.g., ...). Figure 13 ).

[0080] Studies on the efficacy of ECT in treating depression by subtypes

[0081] Predicting the efficacy of antidepressants is an important way to validate effective biomarkers for mental illnesses. Previous studies have shown that neuroimaging biomarkers can be applied to the antidepressant efficacy in treating depression. Although ECT is considered a relatively effective treatment for major depressive disorder, the cure rate is still only 50-60%. To examine whether the depression subtypes identified in this study show different clinical efficacy with ECT treatment, the study used the top 100 most contributing specific connections of the three subtypes found on the discovery and validation sets. Based on cosine similarity, each participant in the two independent ECT treatment datasets (UNM and UCLA) was mapped to its best-matching subtype, and the efficacy findings for each subtype were statistically analyzed (e.g., Figure 14 The different subtypes showed different clinical efficacy. The partial response rate (HAMD decrease of 25%-50%) in the frontoparietal connectivity subtype was 100%, and the complete response rate was 75%. The partial response rate / complete response rate in the frontoparietal connectivity subtype and the contralateral disconnection subtype were 44% / 33% and 58% / 32%, respectively. The partial response rate / complete response rate in the frontoparietal connectivity subtype and the contralateral disconnection subtype were 83% / 67% and 80% / 70%, respectively.

[0082] This study, for the first time, explored and validated different subtypes of depression based on whole-brain functional connectivity data from 971 participants with depression, through cross-site cross-validation and comparison of ECT clinical efficacy. The study identified three reproducible, cross-dataset subtypes of depression: frontoparietal network connectivity subtype, frontoparietal network connectivity subtype, and contralateral disconnection subtype. The frontoparietal network connectivity subtype was significantly associated with anxiety symptoms, and the frontoparietal network connectivity subtype showed a near 100% response rate to ECT.

[0083] Three subtypes of depression were consistently identified on two independent datasets, with significant correlations in functional connectivity patterns among the corresponding subtypes, indicating the reproducibility of the functional connectivity subtype definitions. Subtype 1 of depression is primarily characterized by frontoparietal connectivity, followed by frontolimbic and parietal-limbic connectivity. Frontoparietal connectivity is mainly related to cognitive control and emotion management, and previous studies have found abnormalities in the frontoparietal network in depression. Subtype 2 of depression is primarily characterized by fronto-occipital connectivity and subcortical nuclei-parietal / occipital / frontal connectivity. Subcortical nuclei mainly include the thalamus and globus pallidus. Previous studies have shown that the thalamus is a relay station for information transmission between subcortical nuclei and the cortex, and abnormal connectivity between the thalamus and cortex networks may be related to mood disorders in depression. Abnormal connectivity between the globus pallidus and the cortex may be related to reward regulation. Subtype 3 is mainly characterized by contralateral disconnection, including the cerebellum, occipital lobe, and frontal lobe. In processing emotional images, participants with depression showed increased activation in the frontal, temporal, and parietal lobes, and decreased activation in the occipital lobe. This suggests that decreased occipital lobe activation may be an initiating factor in emotional cognition in depression. Furthermore, a review of cerebellar function revealed that the cerebellum plays a crucial role in emotion management and mood disorders, in addition to motor control.

[0084] On the discovery and validation sets, the three subtypes of depression did not show consistent differences in age, sex, duration of illness, HAMA, HDRS, and medication use. However, the frontoparietal network connectivity subtype was correlated with HAMA scores on both datasets. Anxiety and depression are two distinct clinical symptoms, but they often co-occur. Clinically, depression diagnostic scales include both anxiety and depressive symptoms, making it impossible to distinguish between subjects with different symptoms of depression based solely on the final scale score. In 2017, Drysdale et al. discovered subtypes of depression associated with anhedonia and anxiety through functional connectivity, indicating that subjects with anxiety and anhedonia corresponded to different brain network connectivity patterns. This study found that frontoparietal network connectivity patterns can characterize the intensity of anxiety symptoms in subjects with depression, while the frontoparietal and contralateral disconnection subtypes were not correlated, further validating that different connectivity patterns correspond to different clinical symptoms.

[0085] Because depression is highly heterogeneous, clinical diagnosis may classify subjects with different etiologies and clinical manifestations as having the same type of depression. However, subjects with different etiologies may require different treatments. Currently, the efficacy of ECT (Electroconvulsive Therapy) for depression is approximately 50-60%. This study found that three subtypes of depression with different functional connectivity patterns showed different response rates to ECT. In particular, the fronto-occipital network connectivity subtype showed a significant reduction in symptoms for all subjects after ECT treatment, with a partial response rate of 100%, which was validated on two independent ECT treatment datasets. Consistent with these findings, Chang et al. found that changes in frontal and occipital fALFF (frontal lobe anaphylaxis) and HDRS (high-frequency epithelial symptom) in atypical mental illnesses, regardless of medication use, were significantly correlated with changes in depressive symptoms. Both studies indicate that abnormal changes in the frontal and occipital lobes are significantly associated with changes in depressive symptoms.

[0086] This invention, based on a large dataset of depression, is the first to employ an unbiased data-driven algorithm to explore the heterogeneity of whole-brain functional connectivity in depression, seeking stable and reproducible subtypes of depression and their differences in clinical ECT (electroconvulsive therapy) efficacy for antidepressant treatment. The results identified three subtypes that are distinguishable within the dataset and generalize across data: frontoparietal network connectivity, frontoparietal network connectivity, and contralateral disconnection. The frontoparietal network connectivity subtype is associated with anxiety symptoms. More importantly, the frontoparietal network connectivity subtype showed a near 100% response rate in ECT treatment for depression, validated on two independent datasets. Therefore, this study demonstrates that different connectivity patterns in depression may exhibit different clinical characteristics, potentially offering potential benefits for promoting personalized clinical interventions and diagnosis.

[0087] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A biotyping method for depression based on resting-state brain functional connectivity, characterized by: It includes the following steps: (1) Data acquisition and preprocessing: Resting-state functional magnetic resonance imaging (rs-fMRI) data of subjects were collected, covering a multi-center cohort with different scanning equipment, including patients with depression and healthy controls; the rs-fMRI data were standardized and preprocessed, including head motion correction, time-level correction, spatial standardization, denoising and ComBat harmonization; based on the brain template, the whole brain was divided into n brain regions, and the blood oxygen level dependent on BOLD signal time series of the whole brain regions was extracted, the Pearson correlation coefficient between brain regions was calculated, an n×n whole brain functional connectivity matrix was generated, and n(n-1) / 2 functional connectivity features of the upper triangular region were extracted; (2) Define biological subtype clusters and features: The improved non-negative matrix factorization (NMF) algorithm is used to perform bi-cluster analysis on the preprocessed functional connectivity features. The threshold is set for the non-smoothness constraint parameter of the improved NMF algorithm. The optimal number of clusters is determined to be 3 subtypes by the co-table correlation coefficient and the sum of squared residuals. Non-smoothness constraints are used to improve feature sparsity during the clustering process. The optimal number of clusters is determined by 10 times of 10-fold cross-validation. Define the core functional connectivity features of each subtype. (3) Subtype validation and standardization: The repeatability of subtypes is validated in an independent validation queue. A subtype classifier is constructed using a linear support vector machine (SVM). A subtype standardization template is established, and the core functional connection feature matrices of the three subtypes are saved as subtype reference templates. (4) Rapid subtyping of new patients: The rs-fMRI data of new patients are standardized and preprocessed, including head motion correction, time layer correction, spatial standardization, denoising and ComBat harmonization, to generate a whole brain functional connectivity matrix; the cosine similarity between the whole brain functional connectivity matrix of new patients and the standardized templates of the three subtypes is calculated, and the new patients are assigned to the subtype with the highest similarity, which is the one with the smallest cosine distance.

2. The method for biotyping depression based on resting-state brain functional connectivity according to claim 1, characterized in that: In step (1), the exclusion criteria for head motion correction are: translation > 3mm or rotation > 3°; spatial normalization is based on the AAL template, the target space is the MNI space, and the voxel size is 3mm×3mm×3mm; denoising is to remove cerebrospinal fluid / vascular signal interference, and also includes the removal of white matter signals and linear trend regression; ComBat harmonization is to eliminate the center / device effect and retain the biological trends of age, sex, and diagnosis.

3. The method for biotyping depression based on resting-state brain functional connectivity according to claim 2, characterized in that: In step (2), the threshold for the non-smooth constraint parameter is set to sparseness=0.

5.

4. The method for biotyping depression based on resting-state brain functional connectivity according to claim 3, characterized in that: In step (2), the three subtypes include: Subtype 1: Prefrontal-parietal hyperconnection features, the core of which involves enhanced functional connectivity between the default mode network (DMN) and the ventral attention network (VAN) and the somatic motor network (SMN) and the frontoparietal network (FPN), and related semantic processing behaviors; Subtype 2: Low connectivity features between the striatum, occipital lobe, and prefrontal lobe, with the core being weakened functional connectivity between the frontoparietal network (FPN) and the visual network (VIS) and the subcortical network (SCN) and the visual network (VIS), resulting in abnormalities in associative memory retrieval and emotional processing. Subtype 3: Abnormal features in subcortical-occipital connectivity, mainly involving enhanced connectivity between the subcortical network SCN and the visual network VIS, weakened connectivity between the limbic network LIM and the somatic motor network SMN, and defects in associated reward processing.

5. The method for biotyping depression based on resting-state brain functional connectivity according to claim 4, characterized in that: In step (4), the matching threshold for cosine similarity is 0.

6. When the similarity is lower than the threshold, a prompt message for supplementing the collected data is output.

6. The method for biotyping depression based on resting-state brain functional connectivity according to claim 5, characterized in that: The method also includes step (5), which establishes a subtype-treatment response correspondence database based on independent treatment cohort data: subtype 1 is treated with drugs; subtype 2 is treated with electroconvulsive therapy (ECT); and subtype 3 is treated with drugs combined with reward pathway regulation intervention.

7. The method for biotyping depression based on resting-state brain functional connectivity according to claim 6, characterized in that: In step (5), all treatments are fine-tuned based on the patient's age and comorbidities to ensure clinical applicability.

8. A biotyping device for depression based on resting-state brain functional connectivity, characterized in that: It includes: The data acquisition and preprocessing module is configured to collect resting-state functional magnetic resonance imaging (rs-fMRI) data from subjects, covering a multi-center cohort with different scanning equipment, including patients with depression and healthy controls; it performs standardized preprocessing on the rs-fMRI data, including head motion correction, temporal correction, spatial standardization, denoising, and ComBat harmonization; it extracts the time series of blood oxygenation level-dependent BOLD signals from n brain regions of the whole brain, calculates the Pearson correlation coefficient between brain regions, generates an n×n whole-brain functional connectivity matrix, and extracts n(n-1) / 2 functional connectivity features from the upper triangular region; A biological subtype clustering and feature module is defined, and its configuration employs an improved nonnegative matrix factorization (NMF) algorithm to perform bi-cluster analysis on preprocessed functional connectivity features. The non-smoothness constraint parameter of the improved NMF algorithm is set with a threshold. The optimal number of clusters (3 subtypes) is determined by the co-integer correlation coefficient and the sum of squared residuals. Non-smoothness constraints are used during clustering to improve feature sparsity. The optimal number of clusters is determined through 10 rounds of 10-fold cross-validation. The core functional connectivity features of each subtype are defined. The subtype validation and standardization module is configured to validate the reproducibility of subtypes in an independent validation queue. It constructs a subtype classifier using a linear support vector machine (SVM) and establishes a subtype standardization template, storing the core functional connection feature matrices of the three subtypes as a subtype reference template. The new patient rapid subtyping module is configured to perform standardized preprocessing on the rs-fMRI data of new patients, including head motion correction, temporal layer correction, spatial standardization, denoising, and ComBat harmonization, to generate a whole-brain functional connectivity matrix. It calculates the cosine similarity between the new patient's whole-brain functional connectivity matrix and the standardized templates of the three subtypes, and assigns the new patient to the subtype with the highest similarity, where the highest similarity is the smallest cosine distance.

9. The application of the resting-state brain functional connectivity biotyping method for depression according to claim 1, characterized in that: It is used to treat depression.

Images