Delta wave band fractal dimension individual heterogeneity-based schizophrenia typing method
By using frequency band filtering and fractal dimension analysis of EEG data, combined with standardized models and metabolic indicators, the problem of insufficient frequency band specificity and individual heterogeneity in the diagnosis of schizophrenia has been solved, enabling refined disease subtyping and progression monitoring, which has high clinical translational value.
Patent Information
- Application Number
- CN202511691868.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2025-11-11
- Filing Date
- 2025-11-18
- Publication Date
- 2026-02-24
AI Technical Summary
Existing technologies for the diagnosis of schizophrenia suffer from insufficient frequency band specificity, inadequate capture of individual heterogeneity, and insufficient clinical relevance. They are unable to accurately identify disease-specific abnormalities and heterogeneity among patients, and lack a multimodal indicator system to explain the disease mechanism.
By preprocessing and frequency band filtering EEG data, delta band signals are extracted. The Katz fractal dimension method is used to calculate complexity features. Individual bias analysis is performed by combining a normalized model and discriminant analysis. By using a Bayesian structural equation model combined with metabolic indicators, the classification and progression monitoring of schizophrenia can be achieved.
It achieves sensitive capture of disease abnormalities at the frequency band specific level, breaks through the limitations of traditional group mean comparison, accurately characterizes the degree of individual abnormality, and combines metabolic indicators to verify physiological significance, possessing refined subtyping and clinical assessment capabilities.
Smart Images
Figure CN121565431A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of neural information processing and auxiliary diagnosis technology of mental illness, specifically a method for the identification, classification and progression monitoring of schizophrenia based on the fractal dimension deviation characteristics of delta band of electroencephalogram (EEG) signals. Background Technology
[0002] Electroencephalogram (EEG), a non-invasive brain activity detection technique, can directly reflect the electrophysiological activity of neuronal populations with millisecond-level temporal resolution. Compared with functional magnetic resonance imaging (fMRI) based on blood oxygen level dependent effect (BOLD), EEG can not only more sensitively capture the dynamic oscillations of the brain in different frequency bands, but also provide multi-dimensional information including power spectrum characteristics, event-related potentials (ERPs), cross-regional functional connectivity, and signal complexity. Although its spatial resolution is limited by the skull and volume conduction effects, EEG has irreplaceable advantages in temporal resolution and directness of neural electrical activity, and has been widely used in the study of the mechanisms of neuropsychiatric diseases and in clinical applications.
[0003] Schizophrenia is a severe mental disorder characterized by impaired information processing and altered neurophysiological complexity, leading to cognitive, emotional, and perceptual disturbances. Numerous studies have shown that schizophrenia not only manifests as localized brain dysfunction but also involves impaired integration and coordination of large-scale neural networks. EEG, as a tool for directly recording brain electrical signals, offers advantages such as high temporal resolution, rich information, and ease of operation, and has therefore been widely used in mental illness research. Particularly in frequency band oscillation analysis, low-frequency delta band (2-4 Hz) activity is considered crucial for information transmission and integration, and its abnormalities are closely related to resting-state dysfunction in schizophrenia. Existing research has utilized EEG complexity indices to explore the neural activity characteristics of schizophrenia, revealing significant differences in brain signal complexity among patients. Complexity indices, by capturing the temporal diversity and unpredictability of signals, can more meticulously reflect the brain's information processing capacity. For example, fractal dimension, as an important method for measuring signal self-similarity and structural complexity, has been shown to distinguish schizophrenic patients from healthy controls and has demonstrated good robustness and reliability in EEG studies of mental illness.
[0004] While existing research has demonstrated the potential of EEG complexity analysis in schizophrenia, several limitations remain. First, most studies calculate complexity based on full-band signals, neglecting the functional differences in information transmission and cognitive processing across different bands, potentially masking specific abnormalities associated with disease progression. Second, current research largely focuses on group-level comparisons of differences, lacking individualized bias analysis methods and failing to accurately capture heterogeneity among patients. Furthermore, the clinical application of complexity indicators still lacks systematic integration with physiological indicators, failing to fully explain the neural and metabolic mechanisms of the disease. Therefore, an innovative method is urgently needed that combines frequency-specific analysis with individual bias modeling and clinical physiological information to improve the diagnostic accuracy and subtype identification capabilities of schizophrenia. Existing methods are mostly based on full-band signal complexity calculations, failing to fully consider the functional differences of different frequency bands in neural information transmission and cognitive processing, resulting in limited ability to identify disease specificity. Traditional analyses are mostly limited to statistical comparisons at the population level, i.e., case-control analyses, which are difficult to reflect the heterogeneity among individual patients and lack an analytical framework that can quantify individual deviations from the norm. The correlation mechanism between complexity indicators and clinical physiological or metabolic indicators is still unclear, making it difficult to form a multimodal indicator system that can explain disease mechanisms and assist in diagnosis. Summary of the Invention
[0005] This invention addresses the problems of insufficient frequency band specificity, insufficient capture of individual heterogeneity, and insufficient clinical relevance in existing technologies, and proposes a method for the identification and classification of schizophrenia.
[0006] This invention first preprocesses and filters EEG data to extract delta band signals, and then uses the Katz fractal dimension method to calculate the complexity features of EEG. Subsequently, it establishes the complexity development trajectory of a healthy control group using a standardized model, calculating the degree to which individual patients deviate from the standardized range. Furthermore, this invention uses a hierarchical heterogeneity analysis method based on discriminant analysis to subtype the patient population. Finally, this invention introduces a Bayesian structural equation model to combine EEG deviation features with metabolic indicators to explore the correlation between disease progression and physiological mechanisms. Therefore, the purpose of this invention is to propose a schizophrenia subtyping method based on individual heterogeneity of delta band fractal dimension, which includes: Step 1: Collect EEG data samples and filter the collected EEG data from each sample; extract delta band signals; sample data includes EEG data, age, gender, and disease diagnosis information; Step 2: Calculate the complexity features of the EEG signals for each sample; Step 3: Remove site effects and establish a normalized model to generate the complex developmental trajectory of the healthy control group, and calculate the individual bias value of the patients; Step 4: Based on individual bias characteristics, heterogeneity analysis based on discriminant analysis first distinguishes between first-episode and chronic patients, and further identifies potential subtypes in the chronic patient population.
[0007] Furthermore, the specific method of step 1 is as follows: First, artifact fragments were identified and removed; then the data were downsampled to 250Hz; then, independent component analysis was performed on each subject to remove eye movement artifacts; then, the data were bandpass filtered from 1 to 45Hz; finally, reference electrode normalization was used for rereference.
[0008] Furthermore, in step 2, the complex EEG features FD are calculated using a region-based approach; The region is divided into: frontal lobe, central area, temporal lobe, and occipital lobe; The calculation method for complex EEG features FD is as follows: ; Where L is the total length of the time series, a is the average step size, and d is the Euclidean distance between the first point in the time series and the point farthest from it. This represents the length of the time series.
[0009] Furthermore, the specific method for step 3 is as follows: Step 3.1: Use the ComBat coordination method to correct the batch effect of the fractal dimension features of the delta band, and remove the site effect; Step 3.2: In a healthy control group, a normalized trajectory of delta band complexity changing with age in different brain regions was established using a generalized additive model; Step 3.3: Calculate the individual deviation value of patients relative to the healthy group based on the normalized model. : ; in, This represents the predicted fractal dimension value of participant i in brain region j, which is obtained by inputting the patient's age and sex into a GAM normalized model trained on a healthy control group. This represents the variance of the predicted values. This represents the variance of the normalized distribution.
[0010] Furthermore, the specific method for step 4 is as follows: Step 4.1: Use a linear discriminant model to learn the optimal direction to distinguish between first-episode and chronic patients; the optimization objective of the linear discriminant model is: ; Its constraints are: ; in, This is the weight vector, representing the direction of linear discrimination; The bias term controls the translation position of the hyperplane in the feature space; These are slack variables used to handle cases where the data cannot be linearly and completely partitioned. Represents the regularization parameter; This represents the label of the nth sample. This represents the feature vector of the i-th sample; Based on this, patients are divided into first-episode and chronic patients, and the chronic patient group is input into the next step for subtype identification; Step 4.2: Subtype identification based on a discriminative clustering framework using multi-hyperplane maximum margin classification; assuming patient characteristics are... Cluster label is The goal is to maximize the separation between the patient and the reference pattern by alternately optimizing k linear classifiers. ; Its constraints are: ; in, This represents the weight vector of the k-th hyperplane. This represents the bias term of the k-th hyperplane. Describing a hyperplane, Indicates the first The transpose of the weight vectors corresponding to each subtype This represents the feature vector of the i-th sample. Indicates the first The bias terms corresponding to each subtype Represents slack variables; By alternating the updating of cluster labels across multiple hyperplanes The system determines the direction of deviation and reassigns patients to the nearest hyperplane; this process is repeated until convergence, thereby achieving subtyping within each patient; ultimately, K interpretable deviation patterns and their corresponding patient subtypes are obtained.
[0011] Furthermore, a Bayesian structural equation model was used to quantitatively estimate and assess the uncertainty of the relationship between "EEG deviation - disease course - metabolism - medication". Metabolic indicators include: total cholesterol, triglycerides, high-density lipoprotein and low-density lipoprotein; the model is used to simultaneously estimate the following four effects: (1) the effect of drugs on metabolic levels; (2) the direct effect of drugs on fractal dimension bias; (3) the mediating effect of drugs on bias indirectly through metabolic levels; (4) the direct effect of disease course on bias; Given the nth observation of a patient, the Bayesian structural equation model consists of two structural equations: ; ; Here, Med indicates medication use, and Dur indicates disease course. This represents the overall score of metabolic indicators. This represents the error term in the metabolic equation. Represents the i-th individual, , These represent the intercept term and the pathway coefficient from drug to metabolism in the metabolic equation, respectively. The intercept term represents the complexity of the deviation equation. The path coefficient represents the deviation from disease course to complexity. The path coefficient representing the deviation from metabolism to complexity. The path coefficient representing the deviation from drug to complexity. Indicating complexity bias, The error term represents the complexity of the deviation equation; In each iteration, the effects of each path are calculated simultaneously to form the complete posterior distribution of these effects and extract quantifiable metrics: use Quantifying the direct effects of the drug Med on metabolic Met; use Quantifying the direct effect of the drug Med on the complexity bias Z; use Quantifying the indirect effects of drug Med on bias through the metabolism of Met; use The direct effect of quantifying the duration of disease (Dur) on the deviation (Z).
[0012] Furthermore, the specific method of step 3.2 is as follows: For the i-th healthy subject, given brain region j, its delta band complexity is defined as Y. ij First, construct an age-based smoothing function. Based on thin-plate regression splines, the maximum basis complexity is k=3, and the smoothing function... Represented as: ; in, For the basis functions of thin plate splines, Let K be the spline coefficients to be estimated, and K be the maximum dimension of the basis function, which is K=3. Let i represent the age of the i-th individual; The model estimates the parameters by minimizing the penalized likelihood: ; in, Indicates the penalty squared error; X represents the response variable vector, i.e., the delta band complexity; X is the linear term design matrix; Z represents the linear coefficient vector; Z is the basis function matrix of the smoothing term; Represents the coefficient vector of the smoothing function; The smoothing parameter is represented by P; the penalty matrix is also represented by P. Automatic selection using restricted maximum likelihood method That is, to simultaneously optimize within the framework of the restricted maximum likelihood method: ; Then, the generalized additable model is constructed as follows: ; in, This represents the delta band complexity of the i-th healthy subject given brain region j. For age-based smoothing terms, The intercept is... Linear effect for gender; For the error term, Indicates the gender of the i-th individual.
[0013] This invention presents a method for identifying and classifying schizophrenia based on individual heterogeneity in the delta band fractal dimension. Firstly, it innovates in frequency band specificity analysis, enabling feature extraction from key information processing frequency bands for schizophrenia, thus more sensitively capturing disease abnormalities. Secondly, by introducing the ComBat method, it effectively coordinates the site effect of cross-center data collection, improving the stability of modeling and the reliability of results. Thirdly, this invention utilizes multi-center large-sample training to obtain a standardized model and further conducts individualized bias modeling, accurately characterizing the degree of abnormality of individual patients relative to the healthy population, overcoming the limitations of traditional group mean comparison. Subsequently, this invention combines a hierarchical heterogeneity analysis method based on discriminant analysis to achieve disease stage identification and potential subtype classification, thereby possessing refined classification capabilities. Furthermore, this invention combines EEG deviation features with metabolic indicators such as blood glucose and blood lipids, using a Bayesian structural equation model to verify the physiological significance of complex EEG abnormalities. Finally, this method can be widely applied to the auxiliary diagnosis, disease classification, progression monitoring, and clinical assessment of schizophrenia, possessing high clinical translational and application value. Attached Figure Description
[0014] Figure 1 This is an example of the framework of the present invention.
[0015] Figure 2 This is an example result of the present invention - hierarchical subtype identification.
[0016] Figure 3The results of this invention are shown in the example – their correlation with disease course. The correlation was calculated using the Spearman correlation method.
[0017] Figure 4 This invention presents an example of the results—the association with metabolic characteristics. Bayesian structural equation modeling was used to explore the association between disease progression and metabolic characteristics.
[0018] Figure 5 For the sensitivity analysis of the results of this invention, 20% of the participants were randomly removed, and the current analysis was repeated. The results show that the two-subtype division is optimal.
[0019] Figure 6 Sensitivity analysis, a bipartite split verification, is presented as an example of the results of this invention. The results show that the two-subtype division is optimal. Detailed Implementation
[0020] This example uses a real dataset for testing, collecting data from four sites, including data from 726 healthy controls (mean age: 42.36 ± 14.93 years), 99 patients with first-episode schizophrenia (mean age: 30.29 ± 12.94 years), and 64 patients with chronic schizophrenia (mean age: 22.67 ± 4.41 years). EEG data were collected from all participants, with additional physiological and biochemical data collected from 52 patients with schizophrenia, including chlorpromazine equivalent, glucose, total cholesterol, triglycerides, high-density lipoprotein, and low-density lipoprotein.
[0021] Step 1: Preliminary preprocessing of EEG data to obtain a clean signal filtered to the delta band.
[0022] Preprocessing was performed using the EEGLAB toolbox in MATLAB. First, the `clean_rawdata` function was called to automatically identify and remove artifact fragments. Then, the data was downsampled to 250 Hz. Next, Independent Component Analysis (ICA) was performed on each subject to remove artifacts such as eye movement artifacts. Afterward, the data was bandpass filtered from 1 to 45 Hz. Finally, Rereference Electrode Normalization (REST) was used for rereference. Because some data acquisitions incorporated fMRI scans, the processing steps were modified. Specifically, template averaging was first used to remove MRI artifacts caused by gradient switching. Then, the `clean_rawdata` function was used to detect and remove artifact fragments, and the data was downsampled to 250 Hz. Optimal Basis Set was used to correct for cardiac impact artifacts. Then, ICA was used to remove eye movement and residual artifacts. The data was then bandpass filtered from 1 to 45 Hz. Finally, REST was used for rereference.
[0023] Step 2: Calculate the complexity of EEG using the Katz fractal dimension method.
[0024] The preprocessed EEG signal was filtered to the delta band (2–4 Hz), and the Katz fractal dimension (FD) algorithm was used to calculate the complexity features of the EEG. The Katz fractal dimension algorithm formula is as follows:
[0025] ;
[0026] Where L is the total length of the time series, a is the average step size, and d is the Euclidean distance between the first point in the time series and the point farthest from it.
[0027] To further reduce the impact of noise and the differences in EEG caps at different data collection locations, we limited the analysis to electrodes common to all sites. Ultimately, this scheme performed regional feature extraction based on the frontal lobe, central region, temporal lobe, and occipital lobe. The electrode divisions were as follows: frontal lobe (F1, F2, Fz), central region (C1, C2, Cz), temporal lobe (T7, T8, TP7, TP8), and occipital lobe (O1, O2, Oz).
[0028] Step 3: Remove site effects and establish a standardized model to generate the complex development trajectory of the healthy control group and calculate the individual bias value of the patients.
[0029] To address potential systematic differences arising from different data collection sites, the ComBat coordination method was employed to correct for batch effects in the delta-band fractal dimension features, preserving biological differences (age, sex, and disease diagnosis information) while removing site effects. Subsequently, a normalized trajectory of delta-band complexity across different brain regions with age was established in a healthy control group using a generalized additive model (GAM). Finally, based on this normalized trajectory, z-scores were used to calculate the degree of deviation of patients from the normalized range, thereby obtaining their individual deviation values relative to the healthy group. The z-score calculation formula is as follows:
[0030] ;
[0031] in, This represents the predicted fractal dimension value of participant i in brain region j, which is obtained by inputting the patient's age and sex into a GAM normalized model trained on a healthy control group. This represents the variance of the predicted values. This represents the variance of the normalized distribution. The calculation integrates three types of information: the error between observed and predicted values, the predicted variance at the test points, and the overall variance of the normalized data. Considering that predictions in sparse or high-variance regions of the input data space are associated with increased uncertainty, this method takes into account the data distribution. This is particularly crucial in clinical applications because it ensures consistent handling of uncertainty across different prediction scenarios. Therefore, when a prediction occurs in a region where the normalized model exhibits high variability, the model automatically reduces the confidence level of the prediction.
[0032] Step 4: Based on individual bias characteristics, patient subtypes are classified using a hierarchical heterogeneity analysis method based on discriminant analysis.
[0033] By using discriminant analysis-based hierarchical heterogeneity analysis and clustering based on individual patient bias characteristics, we first distinguish between first-episode and chronic patients, and then identify potential subtypes in the chronic patient population, thereby revealing the heterogeneous characteristics of the disease.
[0034] Step 5: Combining Bayesian structural equation modeling with individual bias characteristics and metabolic indicators, we explore the relationship between disease progression and physiological mechanisms.
[0035] By introducing a Bayesian structural equation model, we combined the patient's EEG deviation characteristics with the course of the disease, metabolic indicators (such as total cholesterol, triglycerides, high-density lipoprotein and low-density lipoprotein) and medication use to explore the relationship between the progression of schizophrenia and the systemic physiological mechanisms.
[0036] Step 6, Sensitivity Analysis. To verify the stability of the heterogeneity analysis results based on discriminant analysis, two strategies were used to conduct repeated experiments: (1) randomly selecting 80% of the samples and repeating them 1000 times, and (2) randomly dividing the samples in half and repeating them independently 1000 times each.
[0037] The validation results on the real experimental set are shown below. Figure 2(a) Schematic diagram of the heterogeneity analysis method based on discriminant analysis. (b) HYDRA subtyping based on all patients; ARI indicates that the two subtypes are optimal solutions. The inset shows that the two subtype solutions have an accuracy of 67% in distinguishing between first-episode and chronic patients. (c) Applying discriminant analysis-based heterogeneity analysis again to chronic patients to identify potential subtypes; ARI indicates that the two subtypes are optimal solutions. The inset shows that there is a significant difference between the two subtypes in terms of disease duration (t=-2.221, p=0.030). The bar chart represents the mean, and the error bars represent the standard error of the mean. (d) t-SNE visualization of the bias characteristics of chronic patients (for illustration only), with labels showing the subtypes identified in this subgroup. Abbreviations: HYDRA, Heterogeneity through Discriminative Analysis; ARI, Adjusted Rand Index; t-SNE, t-Distributed Stochastic Neighbor Embedding. The analytical framework proposed in this invention can effectively distinguish between first-episode patients and chronic patients, and stably identify two potential disease subtypes within the chronic patient population. The results of the association analysis of clinical characteristics are shown below. Figure 3 and Figure 4 The results showed that individual patient complexity bias was significantly correlated with multiple clinical characteristics, and this correlation was not affected by medication factors. This suggests that complexity bias can reflect disease-specific characteristics and validates the robustness of the results. Sensitivity analysis results are shown below. Figure 5 and Figure 6 This further demonstrates that the subtype classification results are not driven by a small number of samples, but rather reflect a stable and reliable population pattern.
Claims
1. A schizophrenia classification method based on individual heterogeneity of delta band fractal dimension, the method comprising: Step 1: Collect EEG data samples and filter the collected EEG data samples; Delta band signals were extracted; sample data included EEG data, age, gender, and disease diagnosis information. Step 2: Calculate the complexity features of the EEG signals for each sample; Step 3: Remove site effects and establish a normalized model to generate the complex developmental trajectory of the healthy control group, and calculate the individual bias value of the patients; Step 4: Based on individual bias characteristics, heterogeneity analysis based on discriminant analysis first distinguishes between first-episode and chronic patients, and further identifies potential subtypes in the chronic patient population.
2. The schizophrenia classification method based on individual heterogeneity of delta band fractal dimension as described in claim 1, characterized in that the specific method of step 1 is as follows: First, artifact fragments were identified and removed; then the data were downsampled to 250Hz; then, independent component analysis was performed on each subject to remove eye movement artifacts; then, the data were bandpass filtered from 1 to 45Hz; finally, reference electrode normalization was used for rereference.
3. The schizophrenia classification method based on individual heterogeneity of delta band fractal dimension as described in claim 1, characterized in that, in step 2, the EEG complexity feature FD is calculated by dividing the region; The region is divided into: frontal lobe, central area, temporal lobe, and occipital lobe; The calculation method for complex EEG features FD is as follows: ； in, L is the total length of the time series, a is the average step size, and d is the Euclidean distance between the first point in the time series and the point farthest from it. This represents the length of the time series.
4. The schizophrenia classification method based on individual heterogeneity of delta band fractal dimension as described in claim 1, characterized in that the specific method of step 3 is as follows: Step 3.1: Use the ComBat coordination method to correct the batch effect of the fractal dimension features of the delta band, and remove the site effect; Step 3.2: In a healthy control group, a normalized trajectory of delta band complexity changing with age in different brain regions was established using a generalized additive model; Step 3.3: Calculate the individual deviation value of patients relative to the healthy group based on the normalized model. : ； in, This represents the predicted fractal dimension value of participant i in brain region j, which is obtained by inputting the patient's age and sex into a GAM normalized model trained on a healthy control group. This represents the variance of the predicted values. This represents the variance of the normalized distribution.
5. The schizophrenia classification method based on individual heterogeneity of delta band fractal dimension as described in claim 1, characterized in that, The specific method for step 4 is as follows: Step 4.1: Use a linear discriminant model to learn the optimal direction to distinguish between first-episode and chronic patients; the optimization objective of the linear discriminant model is: ； Its constraints are: ； in, This is the weight vector, representing the direction of linear discrimination; The bias term controls the translation position of the hyperplane in the feature space; These are slack variables used to handle cases where the data cannot be linearly and completely partitioned. Represents the regularization parameter; This represents the label of the nth sample. This represents the feature vector of the i-th sample; Based on this, patients are divided into first-episode and chronic patients, and the chronic patient group is input into the next step for subtype identification; Step 4.2: Subtype identification based on a discriminative clustering framework using multi-hyperplane maximum margin classification; assuming patient characteristics are... Cluster label is The goal is to maximize the separation between the patient and the reference pattern by alternately optimizing k linear classifiers. ； Its constraints are: ； in, This represents the weight vector of the k-th hyperplane. This represents the bias term of the k-th hyperplane. Describing a hyperplane, Indicates the first The transpose of the weight vectors corresponding to each subtype This represents the feature vector of the i-th sample. Indicates the first The bias terms corresponding to each subtype Represents slack variables; By alternating the updating of cluster labels across multiple hyperplanes The system determines the direction of deviation and reassigns patients to the nearest hyperplane; this process is repeated until convergence, thereby achieving subtyping within each patient; ultimately, K interpretable deviation patterns and their corresponding patient subtypes are obtained.
6. The schizophrenia classification method based on individual heterogeneity of delta band fractal dimension as described in claim 1, characterized in that, A Bayesian structural equation model was used to quantitatively estimate and assess the uncertainty of the relationship between "EEG deviation, disease course, metabolism, and medication". Metabolic indicators include: total cholesterol, triglycerides, high-density lipoprotein and low-density lipoprotein; the model is used to simultaneously estimate the following four effects: (1) the effect of drugs on metabolic levels; (2) the direct effect of drugs on fractal dimension bias; (3) the mediating effect of drugs on bias indirectly through metabolic levels; (4) the direct effect of disease course on bias; Given the nth observation of a patient, the Bayesian structural equation model consists of two structural equations: ； ； Here, Med indicates medication use, and Dur indicates disease course. This represents the overall score of metabolic indicators. This represents the error term in the metabolic equation. Represents the i-th individual, , These represent the intercept term and the pathway coefficient from drug to metabolism in the metabolic equation, respectively. The intercept term represents the complexity of the deviation equation. The path coefficient represents the deviation from disease course to complexity. The path coefficient representing the deviation from metabolism to complexity. The path coefficient representing the deviation from drug to complexity. Indicating complexity bias, The error term represents the complexity of the deviation equation; In each iteration, the effects of each path are calculated simultaneously to form the complete posterior distribution of these effects and extract quantifiable metrics: use Quantifying the direct effects of the drug Med on metabolic Met; use Quantifying the direct effect of the drug Med on the complexity bias Z; use Quantifying the indirect effects of drug Med on bias through the metabolism of Met; use The direct effect of quantifying the duration of disease (Dur) on the deviation (Z).
7. The schizophrenia classification method based on individual heterogeneity of delta band fractal dimension as described in claim 1, characterized in that, The specific method for step 3.2 is as follows: For the i-th healthy subject, given brain region j, its delta band complexity is defined as Y. ij First, construct an age-based smoothing function. Based on thin-plate regression splines, the maximum basis complexity is k=3, and the smoothing function... Represented as: ； in, For the basis functions of thin plate splines, Let K be the spline coefficients to be estimated, and K be the maximum dimension of the basis function, which is K=3. Let i represent the age of the i-th individual; The model estimates the parameters by minimizing the penalized likelihood: ； in, Indicates the penalty squared error; X represents the response variable vector, i.e., the delta band complexity; X is the linear term design matrix; Z represents the linear coefficient vector; Z is the basis function matrix of the smoothing term; Represents the coefficient vector of the smoothing function; The smoothing parameter is represented by P; the penalty matrix is also represented by P. Automatic selection using restricted maximum likelihood method That is, to simultaneously optimize within the framework of the restricted maximum likelihood method: ； Then, the generalized additable model is constructed as follows: ； in, This represents the delta band complexity of the i-th healthy subject given brain region j. For age-based smoothing terms, The intercept is... Linear effect for gender; For the error term, Indicates the gender of the i-th individual.