Individualized TMS target localization method and system for depressive disorders based on fMRI subtypes

Abstract

The present invention relates to an individualized TMS target positioning method for depressive disorder based on fMRI subtypes, and the steps are: using a group of resting-state fMRI data of healthy individuals to construct a normative model; constructing a depressive disorder deviation map based on the normative model and a group of resting-state fMRI data of depressive disorder subjects; using a cluster analysis method to classify depressive disorder subtypes based on the obtained depressive disorder deviation map, and obtaining a deviation map of the depressive disorder subtype; calculating the brain regions of interest of different depressive disorder subtypes based on the deviation map of the depressive disorder subtype; training a classifier that can distinguish different depressive disorder subtypes based on the deviation map of the depressive disorder subtype; using the depressive subtype classifier to classify the fMRI data of individuals with depressive disorder, obtaining the classification results of the depressive subtypes and the brain regions of interest corresponding to the subtypes, and calculating the individualized intervention targets of the depressive disorder subtypes. The present invention fully considers the heterogeneity of depressive disorders and the differences in resting-state fMRI indicators in different ages and genders, and can achieve individualized TMS target positioning.

Description

Technical Field

[0001] The present invention belongs to the technical field of TMS target positioning, and in particular relates to a TMS individualized target positioning method and system for depressive disorder based on fMRI subtypes. Background Art

[0002] TMS (Transcranial magnetic stimulation) stimulates the left DLPFC by applying a strong magnetic field. It was approved by the U.S. Food and Drug Administration (FDA) in 2008 for the treatment of refractory major depressive disorder. However, the efficacy of existing TMS in treating depression is not good. The heterogeneity of TMS treatment results may be due to the inconsistency of DLPFC stimulation targets. When treating depressive disorders, determining the optimal treatment site within the DLPFC remains an important goal of psychiatry in order to improve treatment outcomes.

[0003] More and more studies have shown that the clinical effect of TMS is closely related to the functional connectivity of the target. For example, the negative connectivity between the subgenual anterior cingulate cortex and the DLPFC is related to the efficacy, and the subgenual anterior cingulate cortex (sgACC) is used as a seed point, and its functional connectivity peak point in the DLPFC is calculated as the TMS target to treat depressive disorders. However, these methods rely on the selection of a single brain seed region, ignoring the biological differences between different subjects. The limitation of this method is that it assumes that the seed area shows consistent functional abnormalities in all subjects, without taking into account the challenges posed by the heterogeneity of depressive disorders.

[0004] Recent studies have pointed out that there are significant differences in the TMS treatment effects among subjects with different neuroimaging subtypes, which highlights the importance of determining TMS targets based on the neuroimaging characteristics of subtypes. However, most previous studies have ignored the significant changes that MRI indicators may undergo during the life cycle. Therefore, we constructed a normative model of the functional connectivity values ​​of the left DLPFC based on large-scale healthy population data and further defined the neurobiological subtypes of depressive disorders. This model takes into account the heterogeneity of depression and the changes in MRI indicators at different ages. By combining this standard model with the subtypes identified from the resting-state fMRI (functional magnetic resonance imaging) data of a group of depressive disorder subjects, we can implement individualized TMS target positioning for each subject, aiming to optimize the therapeutic effect of TMS based on the specific neurobiological characteristics of each subject. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art that the heterogeneity and individual differences of depressive disorders are not taken into account, and to provide a method for personalized TMS target positioning for depressive disorders based on fMRI subtypes. By considering the heterogeneity of depression and the changes in MRI indicators of different ages and genders, a normative model is constructed, and combined with the subtypes identified in the resting-state fMRI data of a group of depressive disorder subjects, personalized TMS target positioning is achieved.

[0006] The present invention solves the technical problem by the following technical solutions:

[0007] A method for individualized TMS target localization for depressive disorder based on fMRI subtypes, the method steps are as follows:

[0008] Step 1: Use resting-state fMRI data from a group of healthy individuals to build a normative model;

[0009] Step 2: construct a depressive disorder bias map based on the normative model of healthy individuals and the resting-state fMRI data of a group of depressive disorder subjects;

[0010] Step 3, based on the obtained depressive disorder deviation map, a cluster analysis method is used to classify depressive disorder subtypes to obtain a deviation map of depressive disorder subtypes;

[0011] Step 4, based on the deviation map of depressive disorder subtypes, calculate the brain regions of interest of different depressive disorder subtypes;

[0012] Step 5: Based on the deviation map of depressive disorder subtypes, train a classifier that can distinguish different depressive disorder subtypes;

[0013] Step 6: Use the depressive disorder subtype classifier to classify the fMRI data of individuals with depressive disorders, obtain the classification results of depressive disorder subtypes and the brain regions of interest corresponding to the subtypes, and calculate the individualized intervention targets for depressive disorder subtypes.

[0014] Moreover, in step 1, a set of resting fMRI data of healthy individuals is used to construct a normative model, and the specific method is:

[0015] 1.1. Collect resting-state fMRI data from a group of healthy individuals;

[0016] 1.2. Preprocessing of the collected resting-state fMRI data of healthy individuals, including removal of the initial time point, time layer correction, head motion correction, segmentation, regression covariates, registration to MNI space, and filtering;

[0017] 1.3. The pre-processed resting-state fMRI data of healthy individuals were used to extract BOLD signals of 116 brain regions according to the AAL atlas, of which one brain region BOLD signal was the BOLD signal of the left dorsolateral prefrontal cortex (L_DLPFC);

[0018] 1.4. Use DPABI software to calculate the correlation between the BOLD signal of the left dorsolateral prefrontal cortex and the BOLD signals of the other 115 brain regions, and perform Fisher-z transformation to obtain the functional connectivity value to form a data set of L_DLPFC functional connectivity values. Each healthy individual has 115 functional connectivity values ​​of the left dorsolateral prefrontal cortex;

[0019] 1.5. The Combat method was used to control the different site effects of the data set of functional connectivity values ​​of the left dorsolateral prefrontal cortex, reduce the interference of the site effect on modeling, and obtain 115 functional connectivity values ​​of the left dorsolateral prefrontal cortex in the common image space, that is, the functional connectivity values ​​of the left dorsolateral prefrontal cortex of the healthy group;

[0020] 1.6. Based on the 115 functional connectivity values ​​of the left dorsolateral prefrontal cortex in the common imaging space, a normative model of the 115 functional connectivity values ​​of the left dorsolateral prefrontal cortex associated with age and gender was constructed for all healthy individuals using Gaussian process regression (GPR);

[0021] 1.7. Use the PCN toolkit to evaluate the performance of the canonical model.

[0022] Furthermore, in step 2, a depressive disorder deviation map is constructed based on the normative model of healthy individuals and the resting-state fMRI data of a group of depressive disorder subjects, and the specific method is:

[0023] 2.1. Collect resting-state fMRI data of a group of depressive disorder subjects;

[0024] 2.2. Preprocessing of the resting-state fMRI data of the subjects with depressive disorders, including removal of the initial time point, time layer correction, head motion correction, segmentation, regression covariates, registration to MNI space, filtering, etc.

[0025] 2.3. BOLD signals of 116 brain regions were extracted from the preprocessed fMRI data of the depressive disorder subjects according to the AAL atlas, one of which was the BOLD signal of the left dorsolateral prefrontal cortex;

[0026] 2.4. Use DPABI software to calculate the correlation between the BOLD signal of the left dorsolateral prefrontal cortex and the BOLD signals of the other 115 brain regions, and perform Fisher-z transformation to obtain the functional connectivity value, which constitutes a data set of the functional connectivity value of the left dorsolateral prefrontal cortex. Each depressive disorder subject has 115 functional connectivity values ​​of the left dorsolateral prefrontal cortex;

[0027] 2.5. Combat method was used to control the different site effects of the left dorsolateral prefrontal cortex functional connectivity data set of the above-mentioned depressive disorder subjects, and the functional connectivity values ​​of the left dorsolateral prefrontal cortex of the healthy group collected previously were used as a common image space to obtain the functional connectivity data of the left dorsolateral prefrontal cortex of depressive disorder in the common image space with the healthy group;

[0028] 2.6. Based on the functional connectivity data of the left dorsolateral prefrontal cortex of depression disorder in the common image space with the healthy group, the left dorsolateral prefrontal cortex functional connectivity value of each depression disorder subject was compared with the L_DLPFC connectivity value of the brain region in the healthy normative model, and the Z value was used to quantify the degree of deviation:

[0029] is the observed functional connectivity value of the left dorsolateral prefrontal cortex,

[0030] is the predicted functional connectivity value of the left dorsolateral prefrontal cortex,

[0031] is the uncertainty of the prediction,

[0032] is the variance obtained from the canonical distribution n;

[0033] The Z-score provides a statistical estimate of the degree of deviation of the functional connectivity value of the left dorsolateral prefrontal cortex of each depressive disorder subject relative to the healthy population, resulting in a depressive disorder deviation map.

[0034] Furthermore, the step 3 uses a cluster analysis method to classify depressive disorder subtypes based on the obtained depressive disorder deviation map to obtain a deviation map of depressive subtypes, and the specific method is:

[0035] 3.1. Using the depressive disorder deviation map as a clustering feature;

[0036] 3.2. Use the MATLAB k-mean algorithm to explore the clustering performance of different numbers of clusters, namely the silhouette coefficient and the Kalinsky-Harabas index, and determine the number of clusters with the best clustering performance;

[0037] 3.3. Using the optimal number of clusters and based on the deviation map of depressive disorders, we applied the k-means clustering algorithm to explore depressive disorder subtypes with different deviation patterns and obtained the deviation map of depressive subtypes.

[0038] Furthermore, in step 4, based on the deviation map of the depressive disorder subtypes, the brain regions of interest of different depressive disorder subtypes are calculated, and the specific method is:

[0039] 4.1, based on the deviation map of depressive disorder subtypes, observe the deviation pattern characteristics of different subtypes: the percentage of deviations in different brain regions for each subtype;

[0040] 4.2, the five brain regions with the largest percentage of deviations in the deviation maps of different depressive disorder subtypes are respectively selected as the brain regions of interest of the depressive disorder subtype.

[0041] Furthermore, the step 5 trains a classifier capable of distinguishing different depressive disorder subtypes based on the deviation map of the depressive disorder subtypes, and the specific method is:

[0042] 5.1. Using the deviation map of depressive disorder subtypes as the feature set;

[0043] 5.2. The data of the depressive disorder subjects were divided into training set and test set by 10-fold cross validation method;

[0044] 5.3. Use the Python library scikit-learn to train the support vector machine SVM classifier;

[0045] 5.4. Draw the receiver operating characteristic (ROC) curve for depressive disorder and calculate the AUC value to test the performance of the classifier, and train a classifier that can distinguish different subtypes of depressive disorder.

[0046] Furthermore, in step 6, the fMRI data of individuals with depressive disorders are classified using a depression subtype classifier, the classification results of depression subtypes and the brain regions of interest corresponding to the subtypes are obtained, and individualized intervention targets for depression subtypes are calculated, and the specific method is:

[0047] 6.1. Collect resting-state fMRI data from individual depressive disorder subjects;

[0048] 6.2. Preprocessing of the collected resting-state fMRI data, including removal of the initial time point, temporal layer correction, head motion correction, segmentation, regression covariates, registration to MNI space, and filtering;

[0049] 6.3. BOLD signals of 116 brain regions were extracted from the preprocessed data of individual depressive disorder subjects according to the AAL atlas, one of which was the BOLD signal of the left dorsolateral prefrontal cortex;

[0050] 6.4. Use DPABI software to calculate the correlation between the BOLD signals of the left dorsolateral prefrontal cortex and the other 115 brain regions, and perform Fisher-z transformation to obtain the functional connectivity value, forming a data set of the functional connectivity value of the left dorsolateral prefrontal cortex;

[0051] 6.5. Use the Combat method to match the left dorsolateral prefrontal cortex functional connectivity data of the individual depressive disorder to the common image space of the healthy group and the depression group collected previously, and obtain the left dorsolateral prefrontal cortex functional connectivity data of the individual depressive disorder in the common image space of the healthy group and the depression group;

[0052] 6.6. Compare the functional connectivity values ​​of the left dorsolateral prefrontal cortex and each brain region with the healthy normative model, use the Z value to quantify the degree of deviation, and obtain the deviation map of the functional connectivity of the left dorsolateral prefrontal cortex;

[0053] 6.7. Pass the deviation map through the SVM classifier to determine the subtype classification to which it belongs, and use the brain region of interest of the classification as the seed point;

[0054] 6.8. Based on the seed points, the functional connectivity at the voxel level in the left dorsolateral prefrontal cortex was calculated, and the peak point of the functional connectivity was located as the individualized TMS target for transcranial magnetic stimulation.

[0055] A positioning system for the fMRI subtype-based individualized TMS target positioning method for depression, specifically comprising: a computer and a data acquisition module, a data preprocessing module, a resting fMRI index calculation module, a data standardization module, a healthy subject standard model construction module, a depression subject deviation degree calculation module, a depression subject subtype construction module, a depression subject subtype interest brain region calculation module, a depression subject subtype classifier module, and an individualized target acquisition module.

[0056] The data acquisition module is used to acquire resting-state fMRI data of a group of healthy individuals, a group of depressive disorder subjects, and an individual depressive disorder subject;

[0057] The data preprocessing module is responsible for preprocessing the resting-state fMRI data of a group of healthy individuals, a group of depressive disorder subjects, and individual depressive disorder subjects.

[0058] The resting-state fMRI index calculation module calculates the functional connectivity values ​​of the left dorsolateral prefrontal cortex and other brain regions at the voxel level based on the resting-state fMRI of a group of healthy individuals, a group of depressive disorder subjects and individual depressive disorder subjects after preprocessing;

[0059] The data normalization module uses the Combat method to control the site effect between different data sets and make the functional connectivity values ​​of the left dorsolateral prefrontal cortex and other brain regions of a group of healthy individuals, a group of depressive disorder subjects, and individual depressive disorder subjects all in the common imaging space;

[0060] A module for building a normative model for healthy subjects, using GPR to build a normative model of functional connectivity between the left dorsolateral prefrontal cortex and 115 brain regions related to age and gender for all healthy individuals, and using the PCN toolkit to estimate and evaluate the GPR model;

[0061] The module for calculating the degree of deviation of depressive disorder subjects compares the functional connectivity value of the left dorsolateral prefrontal cortex of each depressive disorder subject with the healthy normative model and quantifies the degree of deviation using the Z value;

[0062] The subtype construction module of depressive disorder subjects was built, and the k-means clustering algorithm was applied to explore the subtypes of depressive disorder with different deviation patterns, and the silhouette coefficient and Karlinsky-Harabas index were used to determine the optimal number of clusters;

[0063] The calculation module of the brain regions of interest of the subtypes of depressive disorder subjects is based on the deviation map of the subtypes of depressive disorder, and observes the deviation pattern characteristics of different subtypes: the percentage of deviations in different brain regions for each subtype, and the five brain regions with the largest percentage of deviations are taken as the brain regions of interest;

[0064] The depressive disorder subject subtype classifier module uses the deviation map based on the depressive disorder subtypes and uses the Python library scikit-learn to train the SVM classifier, draw the ROC curve and calculate the AUC value to evaluate the performance of the classifier;

[0065] The individualized target acquisition module passes the deviation map of individual depressive disorder subjects through the SVM classifier to determine the subtype classification to which they belong, and uses the brain region of interest of the classification as the seed point; based on the seed point, the functional connectivity of the individual depressive disorder subjects at the voxel level in the left dorsolateral prefrontal cortex is calculated, and the peak point of the functional connectivity is located as the individualized TMS target of transcranial magnetic stimulation.

[0066] The advantages and beneficial effects of the present invention are:

[0067] The fMRI subtype-based personalized TMS target localization method for depressive disorder of the present invention utilizes resting-state fMRI data of a group of healthy people and depressive disorder subjects, aiming to construct a personalized TMS target localization method. Compared with traditional methods, this method does not rely on strong assumptions about a single deep brain region, and fully considers the significant variations of imaging indicators in different age groups.

[0068] The fMRI subtype-based individualized TMS target localization method for depression disorders of the present invention fully considers the differences in imaging indicators under different ages and genders. The method uses a standard model to detect neurobiological differences in individuals with depression disorders and observes the degree of deviation from the common range of normal people. Compared with traditional localization methods, this individualized target localization method takes individual differences into account, thereby more effectively performing individualized target localization calculations. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 The figure is a calculation flow chart of the method for individualized TMS target localization for depressive disorders based on fMRI subtypes of the present invention;

[0070] Figure 2 An example of a normative model constructed using resting-state fMRI data from a group of healthy individuals (the figure shows the normative model of the functional connectivity between the left dorsolateral prefrontal cortex and the right supplementary motor area changing with age in the normal female population. The blue solid line is the normative trajectory of the functional connectivity between the left dorsolateral prefrontal cortex and the right supplementary motor area changing with age. The green dotted line is the 95% confidence interval, and the black dotted line is the 99% confidence interval);

[0071] Figure 3 An example of a canonical model constructed using resting-state fMRI data from a group of healthy individuals (the figure shows a canonical model of the functional connectivity between the left dorsolateral prefrontal cortex and the right cerebellum_7b subregion changing with age in a normal female population. The blue solid line is the canonical trajectory of the functional connectivity between the left dorsolateral prefrontal cortex and the right cerebellum_7b subregion changing with age. The green dotted line is the 95% confidence interval, and the black dotted line is the 99% confidence interval);

[0072] Figure 4 Depression disorder graph deviation graph for subjects with depression disorder;

[0073] Figure 5 Based on the deviation diagram of depressive disorder subtypes, cluster analysis method was used to classify depressive disorder subtypes;

[0074] Figure 6 To classify the fMRI data of individuals with depressive disorders according to the depressive disorder subtype classifier, obtain the classification results of depressive subtypes, and calculate the individualized intervention targets for depressive disorder subtypes (coronal view);

[0075] Figure 7 To classify the fMRI data of individuals with depressive disorders according to the depressive disorder subtype classifier, obtain the classification results of depressive subtypes, and calculate the individualized intervention targets (horizontal position diagram) of depressive disorder subtypes;

[0076] Figure 8 It is a structural diagram of the positioning system of the present invention. DETAILED DESCRIPTION

[0077] The present invention is further described in detail below through specific examples. The following examples are only illustrative and not restrictive, and the protection scope of the present invention cannot be limited thereto.

[0078] A method for individualized TMS target localization for depressive disorder based on fMRI subtypes, the method steps are as follows:

[0079] Step 1: Use the resting-state fMRI data of a group of healthy individuals to build a normative model. The specific method is:

[0080] 1.1. Collect resting-state fMRI data from a group of healthy individuals;

[0081] 1.2. Preprocessing of the collected resting-state fMRI data of healthy individuals, including removal of the initial time point, time layer correction, head motion correction, segmentation, regression covariates, registration to MNI space, and filtering;

[0082] 1.3. The pre-processed resting-state fMRI data of healthy individuals were used to extract BOLD signals of 116 brain regions according to the AAL atlas, of which one brain region BOLD signal was the BOLD signal of the left dorsolateral prefrontal cortex (L_DLPFC);

[0083] 1.4. Use DPABI software to calculate the correlation between the BOLD signal of the left dorsolateral prefrontal cortex and the BOLD signals of the other 115 brain regions, and perform Fisher-z transformation to obtain the functional connectivity value to form a data set of L_DLPFC functional connectivity values. Each healthy individual has 115 functional connectivity values ​​of the left dorsolateral prefrontal cortex;

[0084] 1.5. The Combat method was used to control the different site effects of the data set of functional connectivity values ​​of the left dorsolateral prefrontal cortex, reduce the interference of the site effect on modeling, and obtain 115 functional connectivity values ​​of the left dorsolateral prefrontal cortex in the common image space, that is, the functional connectivity values ​​of the left dorsolateral prefrontal cortex of the healthy group;

[0085] 1.6. Based on the 115 functional connectivity values ​​of the left dorsolateral prefrontal cortex in the common imaging space, a normative model of the 115 functional connectivity values ​​of the left dorsolateral prefrontal cortex associated with age and gender was constructed for all healthy individuals using Gaussian process regression (GPR);

[0086] 1.7. Use the PCN toolkit to evaluate the performance of the canonical model.

[0087] The REST-meta-MDD database was constructed by the Depression Imaging Research Alliance, which consists of 25 research groups in China. These research groups agreed to share the resting-state fMRI data of 729 normal controls aged 16-50 years. These data were uniformly preprocessed by each research group using DPABI according to the default process of DPABI, including removing the initial time point, time layer correction, head motion correction, segmentation, regression covariates, alignment to MNI space, filtering, etc., and provided time series data of AAL atlas.

[0088] The DPAIBINet tool in the DPABI software was used to construct the network, and a 116*116 functional connection matrix was obtained. Since the data came from multiple sites, we used the Harmonization tool in the DPABI software and the Combat algorithm to adjust the site effect, with age and gender as covariates of interest. Subsequent analysis was performed based on the functional connection data processed by the Combat algorithm. We selected the left middle frontal gyrus of the AAL atlas (which has been found to be the location of the left DLPFC in some articles) as the L_DLPFC, and obtained 115 functional connection values ​​of the L_DLPFC.

[0089] Two-fold cross validation was used. The functional connectivity values ​​of L_DLPFC were used as the response variable (y), and age and gender were used as the predictor variables (x), with the minimum age being 16 years and the maximum age being 50 years. A standard model of the functional connectivity values ​​of 115 L_DLPFC related to age for all healthy participants was constructed using GPR, and the GPR model was estimated and evaluated using the PCN toolkit, and deviations from the standard model were defined. Figure 2 The two examples shown in the figure are examples of normative models constructed based on 729 normal participants aged 16-50 in the REST-meta-MDD database. The blue solid line is the normative trajectory of the functional connectivity value of the left dorsolateral prefrontal cortex changing with age, the green dotted line is the 95% confidence interval, and the black dotted line is the 99% confidence interval.

[0090] In the figure, abbreviations are: L_DLPFC: left dorsolateral prefrontal cortex; R_SMA: right supplementary motor area; R_Cereb_7b: right cerebellum_7b subregion; FC: functional connectivity.

[0091] Figure 2 is the normalized model of the functional connectivity between the left dorsolateral prefrontal cortex and the right supplementary motor area with age in normal female population; the blue solid line is the normalized trajectory of the functional connectivity between the left dorsolateral prefrontal cortex and the right supplementary motor area with age, the green dotted line is the 95% confidence interval, and the black dotted line is the 99% confidence interval.

[0092] Figure 3is the normalized model of the functional connectivity between the left dorsolateral prefrontal cortex and the right cerebellum_7b subregion changing with age in normal female population. The blue solid line is the normalized trajectory of the functional connectivity between the left dorsolateral prefrontal cortex and the right cerebellum_7b subregion changing with age, the green dotted line is the 95% confidence interval, and the black dotted line is the 99% confidence interval.

[0093] Step 2: Based on the normative model of healthy people and the resting-state fMRI data of a group of depressive disorder subjects, a depressive disorder deviation map is constructed. The specific method is:

[0094] 2.1. Collect resting-state fMRI data of a group of depressive disorder subjects;

[0095] 2.2. Preprocessing of the resting-state fMRI data of the subjects with depressive disorders, including removal of the initial time point, time layer correction, head motion correction, segmentation, regression covariates, registration to MNI space, filtering, etc.

[0096] 2.3. BOLD signals of 116 brain regions were extracted from the preprocessed fMRI data of the depressive disorder subjects according to the AAL atlas, one of which was the BOLD signal of the left dorsolateral prefrontal cortex;

[0097] 2.4. Use DPABI software to calculate the correlation between the BOLD signal of the left dorsolateral prefrontal cortex and the BOLD signals of the other 115 brain regions, and perform Fisher-z transformation to obtain the functional connectivity value, which constitutes a data set of the functional connectivity value of the left dorsolateral prefrontal cortex. Each depressive disorder subject has 115 functional connectivity values ​​of the left dorsolateral prefrontal cortex;

[0098] 2.5. Combat method was used to control the different site effects of the left dorsolateral prefrontal cortex functional connectivity data set of the above-mentioned depressive disorder subjects, and the functional connectivity values ​​of the left dorsolateral prefrontal cortex of the healthy group collected previously were used as a common image space to obtain the functional connectivity data of the left dorsolateral prefrontal cortex of depressive disorder in the common image space with the healthy group;

[0099] 2.6. Based on the functional connectivity data of the left dorsolateral prefrontal cortex of depression disorder in the common image space with the healthy group, the left dorsolateral prefrontal cortex functional connectivity value of each depression disorder subject was compared with the L_DLPFC connectivity value of the brain region in the healthy normative model, and the Z value was used to quantify the degree of deviation:

[0100] is the observed functional connectivity value of the left dorsolateral prefrontal cortex,

[0101] is the predicted functional connectivity value of the left dorsolateral prefrontal cortex,

[0102] is the uncertainty of the prediction,

[0103] is the variance obtained from the canonical distribution n;

[0104] The Z-score provides a statistical estimate of the degree of deviation of the functional connectivity value of the left dorsolateral prefrontal cortex of each depressive disorder subject relative to the healthy population, resulting in a depressive disorder deviation map.

[0105] The REST-meta-MDD database provides time series data of AAL atlas brain regions of 849 depressed subjects. We collected 31 subjects with depressive disorders aged 16-50 years in Tianjin Anding Hospital. A total of 880 depressed subjects, these two datasets were uniformly preprocessed using the DPABI default process, including removing the initial time point, time layer correction, head motion correction, segmentation, regression covariates, alignment to MNI space, filtering and other steps.

[0106] We used the DPAIBINet tool in the DPABI software to construct the network and obtained a 116*116 functional connection matrix. Since the data came from multiple sites, we used the Combat algorithm to adjust for site effects, with age and gender as covariates of interest. Subsequent analysis was performed based on the functional connection data processed by the Combat algorithm to extract the functional connection of the L_DLPFC and obtain 115 functional connection values ​​of the L_DLPFC.

[0107] The functional connectivity values ​​of L_DLPFC of each depressive disorder subject were compared with the healthy normative model, and the degree of deviation was quantified using the Z-score:

[0108] is the observed DC value,

[0109] is the predicted DC value,

[0110] is the uncertainty of the prediction,

[0111] is the variance obtained from the canonical distribution n.

[0112] The Z-score provides a statistical estimate of the degree to which each subject's DC value deviates from the healthy population in each brain region. Thus, an individual deviation map was obtained for each subject.

[0113] like Figure 4 As shown, based on the above, a deviation map of depressive disorder subjects can be constructed.

[0114] Step 3, based on the obtained depressive disorder deviation map, a cluster analysis method is used to classify depressive disorder subtypes to obtain a deviation map of depressive subtypes;

[0115] The specific method is:

[0116] The deviation map of depression subtypes is obtained by classification. The specific method is:

[0117] 3.1. Using the depressive disorder deviation map as a clustering feature;

[0118] 3.2. Use the MATLAB k-mean algorithm to explore the clustering performance of different numbers of clusters, namely the silhouette coefficient and the Kalinsky-Harabas index, and determine the number of clusters with the best clustering performance;

[0119] 3.3. Using the optimal number of clusters and based on the deviation map of depressive disorders, we applied the k-means clustering algorithm to explore depressive disorder subtypes with different deviation patterns and obtained the deviation map of depressive subtypes.

[0120] We used the subjects' deviation maps as clustering features and applied the k-means clustering algorithm to explore depressive disorder subtypes with different deviation patterns. The silhouette coefficient and the Karlinsky-Harabas index were used to determine the optimal number of clusters.

[0121] like Figure 5 As shown, the optimal clustering of the depressive disorder deviation map shown in the example is 2 categories.

[0122] Step 4: Based on the deviation map of depression subtypes, the brain regions of interest of different depressive disorder subtypes are calculated. The specific method is:

[0123] 4.1. Based on the deviation map of depressive disorder subtypes, observe the deviation pattern characteristics of different subtypes: the percentage of deviations in different brain regions for each subtype;

[0124] 4.2. The five brain regions with the largest percentage of deviations in the deviation maps of different depressive disorder subtypes are selected as the brain regions of interest of the depressive disorder subtype.

[0125] Step 5: Based on the deviation map of depressive disorder subtypes, train a classifier that can distinguish different depressive disorder subtypes. The specific method is:

[0126] 5.1. Using the deviation map of depressive disorder subtypes as the feature set;

[0127] 5.2. The data of the depressive disorder subjects were divided into training set and test set by 10-fold cross validation method;

[0128] 5.3. Use the Python library scikit-learn to train the support vector machine SVM classifier;

[0129] 5.4. Draw the receiver operating characteristic (ROC) curve for depressive disorder and calculate the AUC value to test the performance of the classifier, and train a classifier that can distinguish different subtypes of depressive disorder.

[0130] Using the deviation map of the subtype as the feature set, the data of the depressive disorder subjects were divided into training and test sets by 10-fold cross-validation method. The Python library scikit-learn was used to train the SVM classifier; the ROC curve was drawn, and the AUC value was calculated to evaluate the performance of the classifier.

[0131] Step 6: Use the depression subtype classifier to classify the fMRI data of individuals with depressive disorders, obtain the classification results of depression subtypes and the brain regions of interest corresponding to the subtypes, and calculate the individualized intervention targets for the depression subtypes. The specific method is:

[0132] 6.1. Collect resting-state fMRI data from individual depressive disorder subjects;

[0133] 6.2. Preprocessing of the collected resting-state fMRI data, including removal of the initial time point, temporal layer correction, head motion correction, segmentation, regression covariates, registration to MNI space, and filtering;

[0134] 6.3. BOLD signals of 116 brain regions were extracted from the preprocessed data of individual depressive disorder subjects according to the AAL atlas, one of which was the BOLD signal of the left dorsolateral prefrontal cortex;

[0135] 6.4. Use DPABI software to calculate the correlation between the BOLD signals of the left dorsolateral prefrontal cortex and the other 115 brain regions, and perform Fisher-z transformation to obtain the functional connectivity value, forming a data set of the functional connectivity value of the left dorsolateral prefrontal cortex;

[0136] 6.5. Use the Combat method to match the left dorsolateral prefrontal cortex functional connectivity data of the individual depressive disorder to the common image space of the healthy group and the depression group collected previously, and obtain the left dorsolateral prefrontal cortex functional connectivity data of the individual depressive disorder in the common image space of the healthy group and the depression group;

[0137] 6.6. Compare the functional connectivity values ​​of the left dorsolateral prefrontal cortex and each brain region with the healthy normative model, use the Z value to quantify the degree of deviation, and obtain the deviation map of the functional connectivity of the left dorsolateral prefrontal cortex;

[0138] 6.7. Pass the deviation map through the SVM classifier to determine the subtype classification to which it belongs, and use the brain region of interest of the classification as the seed point;

[0139] 6.8. Based on the seed points, the functional connectivity at the voxel level in the left dorsolateral prefrontal cortex was calculated, and the peak point of the functional connectivity was located as the individualized TMS target for transcranial magnetic stimulation.

[0140] Collect resting-state fMRI data of individual depressive disorder subjects; preprocess the collected resting-state fMRI data of individual depressive disorder subjects, including removing the initial time point, time layer correction, head motion correction, segmentation, regression covariates, alignment to MNI space, filtering and other steps; use the AAL atlas to divide the brain into 116 brain regions; use DPABI software to calculate the functional connectivity between L_DLPFC and the remaining 115 brain regions; apply the Combat method to control the site effect between different data sets; compare the functional connectivity values ​​of L_DLPFC and each brain region of the resting-state fMRI data of individual depressive disorder subjects with the healthy normative model, use the Z value to quantify the degree of deviation, and obtain the deviation map of the functional connectivity of L_DLPFC in the resting-state fMRI data of individual depressive disorder subjects; pass the deviation map of the new subject through the SVM classifier to determine the subtype classification to which it belongs, and use the brain region of interest of the classification as the seed point; calculate its functional connectivity at the voxel level of the left dorsolateral prefrontal cortex based on the seed point, and locate the peak point of the functional connectivity as the individual TMS target of transcranial magnetic stimulation.

[0141] like Figure 5 As shown in the figure, the functional connectivity of the resting-state fMRI data of individual depressive disorder subjects at the L_DLPFC voxel level is calculated based on the seed point, and the peak point of the functional connectivity is used as the target area of ​​transcranial magnetic stimulation. That is, using the seed point of the subtype, the peak point of the functional connectivity in the L_DLPFC is calculated, and the peak point is the individualized TMS target. Figure 6 , Figure 7 Coronal and lateral views of the individualized TMS target for one subject. Figure 4 , Figure 5 In the figure, abbreviations are: L_DLPFC: left dorsolateral prefrontal cortex; ROIs: regions of interest; FC: functional connectivity.

[0142] A positioning system for the fMRI subtype-based individualized TMS target positioning method for depression, specifically comprising: a computer and a data acquisition module, a data preprocessing module, a resting fMRI index calculation module, a data standardization module, a healthy subject standard model construction module, a depression subject deviation degree calculation module, a depression subject subtype construction module, a depression subject subtype interest brain region calculation module, a depression subject subtype classifier module, and an individualized target acquisition module.

[0143] The data acquisition module is used to acquire resting-state fMRI data of a group of healthy individuals, a group of depressive disorder subjects, and an individual depressive disorder subject;

[0144] The data preprocessing module is responsible for preprocessing the resting-state fMRI data of a group of healthy individuals, a group of depressive disorder subjects, and individual depressive disorder subjects.

[0145] The resting-state fMRI index calculation module calculates the functional connectivity values ​​of the left dorsolateral prefrontal cortex and other brain regions at the voxel level based on the resting-state fMRI of a group of healthy individuals, a group of depressive disorder subjects and individual depressive disorder subjects after preprocessing;

[0146] The data normalization module uses the Combat method to control the site effect between different data sets and make the functional connectivity values ​​of the left dorsolateral prefrontal cortex and other brain regions of a group of healthy individuals, a group of depressive disorder subjects, and individual depressive disorder subjects all in the common imaging space;

[0147] A module for building a normative model for healthy subjects, using GPR to build a normative model of functional connectivity between the left dorsolateral prefrontal cortex and 115 brain regions related to age and gender for all healthy individuals, and using the PCN toolkit to estimate and evaluate the GPR model;

[0148] The module for calculating the degree of deviation of depressive disorder subjects compares the functional connectivity value of the left dorsolateral prefrontal cortex of each depressive disorder subject with the healthy normative model and quantifies the degree of deviation using the Z value;

[0149] The subtype construction module of depressive disorder subjects was built, and the k-means clustering algorithm was applied to explore the subtypes of depressive disorder with different deviation patterns, and the silhouette coefficient and Karlinsky-Harabas index were used to determine the optimal number of clusters;

[0150] The calculation module of the brain regions of interest of the subtypes of depressive disorder subjects is based on the deviation map of the subtypes of depressive disorder, and observes the deviation pattern characteristics of different subtypes: the percentage of deviations in different brain regions for each subtype, and the five brain regions with the largest percentage of deviations are taken as the brain regions of interest;

[0151] The depressive disorder subject subtype classifier module uses the deviation map based on the depressive disorder subtypes and uses the Python library scikit-learn to train the SVM classifier, draw the ROC curve and calculate the AUC value to evaluate the performance of the classifier;

[0152] The individualized target acquisition module passes the deviation map of individual depressive disorder subjects through the SVM classifier to determine the subtype classification to which they belong, and uses the brain region of interest of the classification as the seed point; based on the seed point, the functional connectivity of the individual depressive disorder subjects at the voxel level in the left dorsolateral prefrontal cortex is calculated, and the peak point of the functional connectivity is located as the individualized TMS target of transcranial magnetic stimulation.

[0153] The present invention uses resting-state fMRI data from a group of healthy people and subjects with depressive disorders to construct an individualized TMS target localization method. Compared with traditional methods, this method does not rely on strong assumptions about a single deep brain region and fully considers the significant variation of imaging indicators in different ages and genders.

[0154] Although the embodiments and drawings of the present invention are disclosed for illustrative purposes, those skilled in the art will appreciate that various substitutions, changes and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the scope of the present invention is not limited to the contents disclosed in the embodiments and drawings.

Claims

1. A method for personalized TMS target localization for depressive disorders based on fMRI subtypes, characterized by: The method steps are: Step 1: Use resting-state fMRI data from a group of healthy individuals to build a normative model; Step 2: construct a depressive disorder bias map based on the normative model of healthy individuals and the resting-state fMRI data of a group of depressive disorder subjects; Step 3, based on the obtained depressive disorder deviation map, a cluster analysis method is used to classify depressive disorder subtypes to obtain a deviation map of depressive disorder subtypes; Step 4, based on the deviation map of depressive disorder subtypes, calculate the brain regions of interest of different depressive disorder subtypes; Step 5: Based on the deviation map of depressive disorder subtypes, train a classifier that can distinguish different depressive disorder subtypes; Step 6: Use the depressive disorder subtype classifier to classify the fMRI data of individuals with depressive disorders, obtain the classification results of depressive disorder subtypes and the brain regions of interest corresponding to the subtypes, and calculate the individualized intervention targets for depressive disorder subtypes; In step 4, based on the deviation map of the depressive disorder subtypes, the brain regions of interest of different depressive disorder subtypes are calculated, and the specific method is: 4.1, based on the deviation map of depressive disorder subtypes, observe the deviation pattern characteristics of different subtypes: the percentage of deviations in different brain regions for each subtype; 4.2, the five brain regions with the largest percentage of deviations in the deviation maps of different depressive disorder subtypes are respectively selected as the brain regions of interest of the depressive disorder subtype.

2. The method for individualized TMS target localization for depressive disorder based on fMRI subtypes according to claim 1, characterized in that: In step 1, a set of resting fMRI data of healthy individuals is used to construct a normative model, and the specific method is: 1.

1. Collect resting-state fMRI data from a group of healthy individuals; 1.

2. Preprocessing of the collected resting-state fMRI data of healthy individuals, including removal of the initial time point, time layer correction, head motion correction, segmentation, regression covariates, registration to MNI space, and filtering; 1.

3. The pre-processed resting-state fMRI data of healthy individuals were used to extract BOLD signals of 116 brain regions according to the AAL atlas, of which one brain region BOLD signal was the BOLD signal of the left dorsolateral prefrontal cortex; 1.

4. Use DPABI software to calculate the correlation between the BOLD signal of the left dorsolateral prefrontal cortex and the BOLD signals of the other 115 brain regions, and perform Fisher-z transformation to obtain the functional connectivity value to form a data set of functional connectivity values. Each healthy individual has 115 functional connectivity values ​​of the left dorsolateral prefrontal cortex; 1.

5. The Combat method was used to control the different site effects of the data set of functional connectivity values ​​of the left dorsolateral prefrontal cortex, reduce the interference of the site effect on modeling, and obtain 115 functional connectivity values ​​of the left dorsolateral prefrontal cortex in the common image space, that is, the functional connectivity values ​​of the left dorsolateral prefrontal cortex of the healthy group; 1.

6. Based on the 115 functional connectivity values ​​of the left dorsolateral prefrontal cortex in the common image space, a normative model of the 115 functional connectivity values ​​of the left dorsolateral prefrontal cortex related to age and gender was constructed for all healthy individuals using Gaussian process regression; 1.

7. Use the PCN toolkit to evaluate the performance of the canonical model.

3. The method for individualized TMS target localization for depressive disorder based on fMRI subtypes according to claim 1, characterized in that: In step 2, based on the normative model of healthy individuals and the resting-state fMRI data of a group of depressive disorder subjects, a depressive disorder deviation map is constructed. The specific method is: 2.

1. Collect resting-state fMRI data of a group of depressive disorder subjects; 2.

2. Preprocessing of the resting-state fMRI data of the subjects with depressive disorders, including removal of the initial time point, time layer correction, head motion correction, segmentation, regression covariates, registration to MNI space, filtering, etc. 2.

3. BOLD signals of 116 brain regions were extracted from the preprocessed fMRI data of the depressive disorder subjects according to the AAL atlas, one of which was the BOLD signal of the left dorsolateral prefrontal cortex; 2.

4. Use DPABI software to calculate the correlation between the BOLD signal of the left dorsolateral prefrontal cortex and the BOLD signals of the other 115 brain regions, and perform Fisher-z transformation to obtain the functional connectivity value, which constitutes a data set of the functional connectivity value of the left dorsolateral prefrontal cortex. Each depressive disorder subject has 115 functional connectivity values ​​of the left dorsolateral prefrontal cortex; 2.

5. Combat method was used to control the different site effects of the left dorsolateral prefrontal cortex functional connectivity data set of the above-mentioned depressive disorder subjects, and the functional connectivity values ​​of the left dorsolateral prefrontal cortex of the previous healthy group were used as a common image space to obtain the functional connectivity data of the left dorsolateral prefrontal cortex of depressive disorder in the common image space with the healthy group; 2.

6. Based on the functional connectivity data of the left dorsolateral prefrontal cortex of depression disorder in the common image space with the healthy group, the left dorsolateral prefrontal cortex functional connectivity value of each depression disorder subject was compared with the L_DLPFC connectivity value of the brain region in the healthy normative model, and the Z value was used to quantify the degree of deviation: ； is the observed functional connectivity value of the left dorsolateral prefrontal cortex, is the predicted functional connectivity value of the left dorsolateral prefrontal cortex, is the uncertainty of the prediction, is the variance obtained from the canonical distribution n; The Z-score provides a statistical estimate of the degree of deviation of the functional connectivity value of the left dorsolateral prefrontal cortex of each depressive disorder subject relative to the healthy population, resulting in a depressive disorder deviation map.

4. The method for individualized TMS target localization for depressive disorder based on fMRI subtypes according to claim 1, characterized in that: The step 3 uses a cluster analysis method to classify depression subtypes based on the obtained depressive disorder deviation map to obtain a deviation map of depression subtypes, and the specific method is: 3.

1. Using the depressive disorder deviation map as a clustering feature; 3.

2. Use the k-mean algorithm of Matlab to explore the clustering performance of different numbers of clusters, namely the silhouette coefficient and the Kalinsky-Harabas index, and determine the number of clusters with the best clustering performance; 3.

3. Using the optimal number of clusters and based on the deviation map of depressive disorders, we applied the k-means clustering algorithm to explore depressive disorder subtypes with different deviation patterns and obtained the deviation map of depressive subtypes.

5. The method for individualized TMS target localization for depressive disorder based on fMRI subtypes according to claim 1, characterized in that: The step 5 trains a classifier capable of distinguishing different depressive disorder subtypes based on the deviation map of depressive disorder subtypes, and the specific method is: 5.

1. Using the deviation map of depressive disorder subtypes as the feature set; 5.

2. The data of the depressive disorder subjects were divided into training set and test set by 10-fold cross validation method; 5.

3. Use the Python library scikit-learn to train the support vector machine SVM classifier; 5.

4. Draw the receiver operating characteristic (ROC) curve for depressive disorder and calculate the AUC value to test the performance of the classifier, and train a classifier that can distinguish different subtypes of depressive disorder.

6. The method for individualized TMS target localization for depressive disorder based on fMRI subtypes according to claim 1, characterized in that: In step 6, the fMRI data of individuals with depressive disorders are classified using a depression subtype classifier, the classification results of depression subtypes and the brain regions of interest corresponding to the subtypes are obtained, and the individualized intervention targets of the depression subtypes are calculated. The specific method is: 6.

1. Collect resting-state fMRI data from individual depressive disorder subjects; 6.

2. Preprocessing of the collected resting-state fMRI data, including removal of the initial time point, temporal layer correction, head motion correction, segmentation, regression covariates, registration to MNI space, and filtering; 6.

3. BOLD signals of 116 brain regions were extracted from the preprocessed data of individual depressive disorder subjects based on the AAL atlas, one of which was the BOLD signal of the left dorsolateral prefrontal cortex; 6.

4. Use DPABI software to calculate the correlation between the BOLD signals of the left dorsolateral prefrontal cortex and the other 115 brain regions, and perform Fisher-z transformation to obtain the functional connectivity value, forming a data set of the functional connectivity value of the left dorsolateral prefrontal cortex; 6.

5. Use the Combat method to match the left dorsolateral prefrontal cortex functional connectivity data of the individual depressive disorder to the common image space of the healthy group and the depression group collected previously, and obtain the left dorsolateral prefrontal cortex functional connectivity data of the individual depressive disorder in the common image space of the healthy group and the depression group; 6.

6. Compare the functional connectivity values ​​of the left dorsolateral prefrontal cortex and each brain region with the healthy normative model, use the Z value to quantify the degree of deviation, and obtain the deviation map of the functional connectivity of the left dorsolateral prefrontal cortex; 6.

7. Pass the deviation map through the SVM classifier to determine the subtype classification to which it belongs, and use the brain region of interest of the classification as the seed point; 6.

8. Based on the seed points, the functional connectivity at the voxel level in the left dorsolateral prefrontal cortex was calculated, and the peak point of the functional connectivity was located as the individualized TMS target for transcranial magnetic stimulation.

7. A positioning system developed according to the method for individualized TMS target positioning for depressive disorders based on fMRI subtypes according to any one of claims 1 to 6, specifically comprising: A computer and a data acquisition module, a data preprocessing module, a resting fMRI index calculation module, a data standardization module, a healthy subject standard model construction module, a depressive disorder subject deviation degree calculation module, a depressive disorder subject subtype construction module, a depressive disorder subject subtype interest brain region calculation module, a depressive disorder subject subtype classifier module, and an individualized target acquisition module running in the computer; The data acquisition module is used to acquire resting-state fMRI data of a group of healthy individuals, a group of depressive disorder subjects, and an individual depressive disorder subject; The data preprocessing module is responsible for preprocessing the resting-state fMRI data of a group of healthy individuals, a group of depressive disorder subjects, and individual depressive disorder subjects. The resting-state fMRI index calculation module calculates the functional connectivity values ​​of the left dorsolateral prefrontal cortex and other brain regions at the voxel level based on the resting-state fMRI of a group of healthy individuals, a group of depressive disorder subjects and individual depressive disorder subjects after preprocessing; The data normalization module uses the Combat method to control the site effect between different data sets and make the functional connectivity values ​​of the left dorsolateral prefrontal cortex and other brain regions of a group of healthy individuals, a group of depressive disorder subjects, and individual depressive disorder subjects all in the common imaging space; A module for building a normative model for healthy subjects, using GPR to build a normative model of functional connectivity between the left dorsolateral prefrontal cortex and 115 brain regions related to age and gender for all healthy individuals, and using the PCN toolkit to estimate and evaluate the GPR model; The module for calculating the degree of deviation of depressive disorder subjects compares the functional connectivity value of the left dorsolateral prefrontal cortex of each depressive disorder subject with the healthy normative model and quantifies the degree of deviation using the Z value; The subtype construction module of depressive disorder subjects was built, and the k-means clustering algorithm was applied to explore the subtypes of depressive disorder with different deviation patterns, and the silhouette coefficient and Karlinsky-Harabas index were used to determine the optimal number of clusters; The calculation module of the brain regions of interest of the subtypes of depressive disorder subjects is based on the deviation map of the subtypes of depressive disorder, and observes the deviation pattern characteristics of different subtypes: the percentage of deviations in different brain regions for each subtype, and the five brain regions with the largest percentage of deviations are taken as the brain regions of interest; The depressive disorder subject subtype classifier module uses the deviation map based on the depressive disorder subtypes and uses the Python library scikit-learn to train the SVM classifier, draw the ROC curve and calculate the AUC value to evaluate the performance of the classifier; The individualized target acquisition module passes the deviation map of individual depressive disorder subjects through the SVM classifier to determine the subtype classification to which they belong, and uses the brain region of interest of the classification as the seed point; based on the seed point, the functional connectivity of the individual depressive disorder subjects at the voxel level in the left dorsolateral prefrontal cortex is calculated, and the peak point of the functional connectivity is located as the individualized TMS target of transcranial magnetic stimulation.

Images