Construction method and system of schizophrenia symptom severity prediction model
By constructing a symptom severity prediction model for schizophrenia based on EEG data and graph theory algorithms, the problem of reliance on subjective judgment in existing technologies has been solved, and more accurate symptom assessment and disease diagnosis have been achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-26
- Publication Date
- 2026-04-03
AI Technical Summary
In current technology, the assessment of the severity of schizophrenia symptoms relies on the doctor's subjective judgment, which lacks objectivity and consistency, and makes it difficult to provide an accurate assessment of the condition before symptoms are obvious.
By acquiring historical EEG data from patients' visuospatial memory tasks, a weighted undirected connectivity matrix is constructed. Key graph theory indicators are calculated using graph theory algorithms, and a multiple linear regression model is used to predict symptom severity, reducing subjective judgment.
It provides a more accurate assessment of the severity of schizophrenia symptoms, reduces the time cost of manual operation and subjective judgment, and provides a more reliable basis for clinical diagnosis and treatment.
Smart Images
Figure CN121789956A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of disease auxiliary diagnosis technology, and in particular to a method and system for constructing a predictive model for the severity of schizophrenia symptoms. Background Technology
[0002] Schizophrenia is a serious mental illness that has a profound impact on patients' social functioning and quality of life.
[0003] Currently, mental illness is mainly assessed by using the Positive and Negative Syndrome Scale (PANSS) to assist doctors in evaluating patients' behavior or clinical manifestations. However, this assessment method has the following limitations: First, the assessment results heavily rely on the doctor's clinical experience and subjective judgment. In addition, most patients lack insight, and the exposure of symptoms and the different levels and perspectives of different assessors on the exposure affect the consistency and objectivity of the assessment. Furthermore, sufficient exposure of symptom phenotypes is required for identification and attention. Before patients develop obvious clinical symptoms, it is difficult to provide accurate judgment on their potential condition or changes in condition (worsening or clinical effects). Summary of the Invention
[0004] This invention provides a method for constructing a predictive model for the severity of schizophrenia symptoms, addressing the aforementioned problems in the prior art, namely, how to provide accurate criteria for judging the severity of schizophrenia symptoms. This invention provides a method for constructing a predictive model for the severity of schizophrenia symptoms, the method comprising: Acquire historical EEG data of patients during the exploration phase of a visuospatial memory task, and construct a weighted undirected connectivity matrix based on the historical EEG data to reflect the connection strength between different brain regions; Based on a weighted undirected connectivity matrix and combined with graph theory algorithms, several key graph theory metrics are obtained. These metrics include clustering coefficients to characterize the local information processing capability of the network, path lengths to characterize the average distance of information transmission in the network, global efficiency to characterize the overall parallel information processing capability of the network, and local efficiency to characterize the fault tolerance capability of the network when local nodes are damaged. Multiple key graph theory indicators were obtained as predictors. The predictors and the dependent variable PANSS score were input into the constructed multiple linear regression model to determine the linear relationship between the multiple key graph theory indicators and the dependent variable PANSS score. Based on the linear relationship between the multiple key graph theory indicators and the PANSS score, a schizophrenia symptom severity prediction model was obtained to predict the severity of symptoms in schizophrenia patients.
[0005] Optionally, the clustering coefficient can be obtained by traversing each node in the weighted undirected connection matrix and determining the ratio of the actual number of connections between each node's neighboring nodes to the maximum number of connections for each node.
[0006] Optionally, the shortest path length between all node pairs in the distance matrix is determined by converting the weighted undirected connection matrix into a distance matrix, and the average of the shortest path lengths is taken as the path length.
[0007] Optionally, a shortest path algorithm is used to determine the shortest path length between all pairs of nodes in the distance matrix; wherein, the shortest path algorithm is Dijkstra's algorithm.
[0008] Optionally, the shortest path length between all pairs of nodes in the weighted undirected connection matrix is obtained, and the average of the reciprocals of the shortest path length is taken as the global efficiency; for each removed node, the global efficiency of the subgraph formed by its neighboring nodes is obtained and taken as the local efficiency of the removed node.
[0009] Optionally, the historical EEG data can be obtained through wavelet transform denoising and independent component analysis (ICA).
[0010] Optionally, the multiple key graph theory metrics can be Z-score standardized.
[0011] This invention provides a system for constructing a predictive model for the severity of schizophrenia symptoms, comprising: The acquisition module is used to acquire the patient's historical EEG data during the exploration phase of the visuospatial memory task, and to construct a weighted undirected connectivity matrix based on the historical EEG data to reflect the connection strength between different brain regions. The determination module is used to obtain multiple key graph theory indicators based on a weighted undirected connection matrix and combined with graph theory algorithms. Among them, the multiple key graph theory indicators include clustering coefficients to characterize the local information processing capability of the network, path lengths to characterize the average distance of information transmission in the network, global efficiency to characterize the overall parallel information processing capability of the network, and local efficiency to characterize the fault tolerance capability of the network when local nodes are damaged. The module is used to take multiple key graph theory indicators as predictors, input the predictors and the dependent variable PANSS score into the constructed multiple linear regression model to determine the linear relationship between the multiple key graph theory indicators and the dependent variable PANSS score; based on the linear relationship between the multiple key graph theory indicators and the PANSS score, a schizophrenia symptom severity prediction model is obtained to predict the severity of schizophrenia symptoms in patients.
[0012] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention provides a method for constructing a predictive model for the severity of schizophrenia symptoms. This method collects EEG data from patients during the visual-spatial memory probing phase and calculates four key graph theory indicators using graph theory algorithms. Then, using the graph theory indicators as core predictive factors, the predicted PANSS score of the patient can be calculated, reducing the time cost of manual operation and subjective judgment. This provides clinicians with more accurate assessment criteria to assist in the diagnosis and treatment of mental illnesses. Attached Figure Description
[0013] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.
[0014] Figure 1 A flowchart illustrating a method for constructing a schizophrenia symptom severity prediction model provided in an embodiment of the present invention; Figure 2 A residual-fit value graph provided in an embodiment of the present invention; Figure 3 This is a comparison chart of predicted and actual values provided in an embodiment of the present invention. Detailed Implementation
[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0016] The technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present invention will now be described with reference to the accompanying drawings.
[0017] Figure 1 This is a flowchart illustrating a method for constructing a schizophrenia symptom severity prediction model according to an embodiment of the present invention. Figure 1 As shown in this embodiment, a method for constructing a predictive model for the severity of schizophrenia symptoms includes: S1: Obtain the patient's historical EEG data during the exploration phase of the visuospatial memory task, and construct a weighted undirected connectivity matrix based on the historical EEG data to reflect the connection strength between different brain regions.
[0018] Generally, the visuospatial memory task is a cognitive task used to assess a patient's cognitive function, particularly spatial memory and recognition abilities. This task is performed using a delayed-matching sample paradigm, comprising an encoding phase and a probing phase. In the encoding phase, the patient is shown spatial location stimuli; in the probing phase, the patient needs to determine whether the target location matches the previously seen stimulus. Graph theory algorithms are a class of mathematical methods used to analyze and process graph-structured data. Graphs consist of nodes (or vertices) and edges connecting the nodes, used to represent relationships between objects or network structures.
[0019] For example, EEG data were acquired using a 64-lead EEG recording system (such as the Brain Products GmbH system) at a sampling rate of 500 Hz. The visuospatial memory task employed a delayed matching sample paradigm, comprising an encoding phase (displaying spatial location stimuli for 2 seconds) and a probing phase (requiring the patient to judge whether the target location matches for 1.5 seconds), with a total task duration of approximately 30 minutes. Data collection was conducted in a quiet environment, and all participants were patients with schizophrenia meeting the DSM-5 diagnostic criteria.
[0020] For example, preprocessing was performed using MATLAB toolkits such as EEGLAB. Wavelet transform denoising was performed using the Daubechies 4 wavelet, with a 5-level decomposition, and noise was removed using soft thresholding and universal thresholding rules. Independent component analysis (ICA) was performed using the FastICA algorithm to calculate each independent component, and components related to artifacts such as eye movements and electromyography were removed by visual inspection. Finally, historical EEG signals from 0.1 to 40 Hz were extracted for subsequent analysis.
[0021] S2: Based on the weighted undirected connection matrix and combined with graph theory algorithms, obtain multiple key graph theory indicators; among which, the multiple key graph theory indicators include clustering coefficients for characterizing the local information processing capability of the network, path lengths for characterizing the average distance of information transmission in the network, global efficiency for characterizing the overall parallel information processing capability of the network, and local efficiency for characterizing the fault tolerance capability of the network when local nodes are damaged.
[0022] For example, the process of determining multiple graph theory indicators based on preprocessed EEG signals may specifically include: Clustering coefficient calculation: Traverse each node in the network, calculate the ratio of the actual connection strength between its neighboring nodes to the maximum possible connection strength, and finally report the arithmetic mean of the ratios of all nodes.
[0023] Path length calculation: First, the functional connectivity matrix (where larger values indicate stronger connections) is converted into a distance matrix (where larger values indicate greater distances and higher resistance to information transmission). The conversion rule is to subtract the original connectivity strength value from 1 and set the main diagonal of the matrix to zero. Then, a classic shortest path algorithm (such as Dijkstra's algorithm) is used to calculate the shortest path length between all pairs of nodes in the distance matrix. Finally, the arithmetic mean of the shortest path lengths for all pairs of nodes is reported.
[0024] Global efficiency calculation: This calculates the arithmetic mean of the efficiencies between all pairs of nodes in the network. The efficiency between any two nodes is defined as the reciprocal of their shortest path length. This calculation is directly based on the original function connectivity matrix.
[0025] Local efficiency calculation: For each node in the network, after temporarily removing the node, calculate the global efficiency of the subgraph formed by all its neighboring nodes. This is the local efficiency of that node. Finally, report the arithmetic mean of the local efficiencies of all nodes.
[0026] Finally, the calculated clustering coefficients, path lengths, global efficiency, and local efficiency are Z-score standardized to eliminate the influence of dimensions and serve as input features for subsequent modeling.
[0027] S3: Using the acquired key graph theory indicators as predictors, input the predictors and the dependent variable PANSS score into the constructed multiple linear regression model to determine the linear relationship between the key graph theory indicators and the dependent variable PANSS score; based on the linear relationship between the key graph theory indicators and the PANSS score, obtain a schizophrenia symptom severity prediction model for predicting the severity of symptoms in schizophrenia patients.
[0028] For example, this invention selects a multiple linear regression model as the core algorithm, mainly based on the following characteristics and technical advantages: (1) Clear Explanation of Variable Relationships. The model can intuitively reflect the linear relationship between each predictor (clustering coefficient, path length, global efficiency, and local efficiency) and the target variable (PANSS score) through regression coefficients (β values). For example, the significant contributions of the clustering coefficient (β=82.7, P=0.048) and local efficiency (β=-96.92, P=0.027) indicate that they have clear directional and magnitude characteristics in predicting symptom severity (see Appendix Table 1 for details). This interpretability provides clinicians with a reference for personalized diagnosis and treatment based on biomarkers.
[0029] (2) Adaptability to small sample data. The sample size of this study was 18 cases, which falls into the category of small sample size. The multiple linear regression model effectively eliminated the difference in dimensions through standardization (Z-score transformation), reducing its dependence on large sample sizes. Compared with complex models that require a large amount of data for training (such as neural networks), this model can still maintain stable performance in limited data scenarios.
[0030] (3) Model robustness and generalization ability. Through 5-fold cross-validation, the model showed good consistency between the training and test sets: the mean absolute error (MAE) was 0.81, the residual standard deviation (RSE) was 0.67, and the correlation coefficient between the predicted and actual values was (r=0.76, P<0.001). The residual plot showed that the residuals were randomly distributed and had no trend, verifying the goodness of fit of the model. These indicators together show that the model has reliable generalization performance while avoiding overfitting.
[0031] (4) Capturing the synergistic effect of multiple indicators. Compared with a single graph theory indicator used as a predictor, incorporating multiple graph theory indicators into the model significantly improved the overall explanatory power (R²=66%). This indicates that the combination of multiple indicators can more comprehensively capture the complex features of brain network changes in the visuospatial working memory task, showing a significant advantage over single indicator models (such as the clustering coefficient with R²=21%). The regression analysis results of different predictor combinations on PANSS scores are shown in Tables 1 to 5.
[0032] Table 1. Regression analysis results of each predictor on PANSS score Table 2. Regression analysis results of clustering coefficients on PANSS scores during the exploration phase. Table 3. Regression analysis results of path length during the exploration phase on PANSS score. Table 4. Regression analysis results of global efficiency in the detection phase on PANSS score. Table 5. Regression analysis results of local efficiency during the detection phase on the PANSS score. For example, the predicted PANSS score output by a model predicting the severity of schizophrenia symptoms can be provided as a reference for clinicians. Doctors can then combine this information with the patient's specific situation to develop a more accurate treatment plan, such as... Figure 3 The image shows a comparison between the model's predicted values and the actual values.
[0033] For example, this embodiment can evaluate the performance of a schizophrenia symptom severity prediction model by employing 5-fold cross-validation. 5-fold cross-validation is a commonly used machine learning model evaluation method. This method can not only evaluate the model's generalization ability but also prevent the model from overfitting. When validating the model, this validation method can divide the training set into 5 subsets. In each iteration, 4 datasets are used as the training set, and the remaining subset is used as the test set, for a total of 5 iterations.
[0034] The results showed that the predictive model for the severity of schizophrenia symptoms was valid (F=6.27, P=0.005). After 5 iterations, the model's explanatory power was approximately 55%, with a mean absolute error (MAE) of 0.81, a mean absolute error of 0.14 (MAESD), and a residual standard deviation (RSE) of 0.67.
[0035] At the same time, such as Figure 2 As shown, the residuals exhibit an approximately normal distribution with no significant trend, indicating that the model satisfies the linear regression assumption. The prediction error is homoscedastic, verifying the statistical reliability of the model. The comparison between predicted and actual values demonstrates the good accuracy of the model's predictions (r=0.76, P<0.001).
[0036] In this invention, the model performance can be evaluated by using the 5-fold cross-validation method, which can effectively avoid the impact of sample size on statistical power.
[0037] The above describes a method for constructing a schizophrenia symptom severity prediction model according to one or more embodiments of this specification. Based on the same idea, this specification also provides a corresponding system for constructing a schizophrenia symptom severity prediction model, including: The acquisition module is used to acquire the patient's historical EEG data during the exploration phase of the visuospatial memory task, and to construct a weighted undirected connectivity matrix based on the historical EEG data to reflect the connection strength between different brain regions. The determination module is used to obtain multiple key graph theory indicators based on a weighted undirected connection matrix and combined with graph theory algorithms. Among them, the multiple key graph theory indicators include clustering coefficients to characterize the local information processing capability of the network, path lengths to characterize the average distance of information transmission in the network, global efficiency to characterize the overall parallel information processing capability of the network, and local efficiency to characterize the fault tolerance capability of the network when local nodes are damaged. The module is used to take multiple key graph theory indicators as predictors, input the predictors and the dependent variable PANSS score into the constructed multiple linear regression model to determine the linear relationship between the multiple key graph theory indicators and the dependent variable PANSS score; based on the linear relationship between the multiple key graph theory indicators and the PANSS score, a schizophrenia symptom severity prediction model is obtained to predict the severity of schizophrenia symptoms in patients.
[0038] Specific limitations regarding the construction system for the schizophrenia symptom severity prediction model can be found in the above-mentioned limitations on the construction method of the schizophrenia symptom severity prediction model, and will not be repeated here. Each module in the above-mentioned schizophrenia symptom severity prediction model construction system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.
[0039] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this invention.
Claims
1. A method for constructing a predictive model for the severity of schizophrenia symptoms, characterized in that, include: Acquire historical EEG data of patients during the exploration phase of a visuospatial memory task, and construct a weighted undirected connectivity matrix based on the historical EEG data to reflect the connection strength between different brain regions; Based on a weighted undirected connectivity matrix and combined with graph theory algorithms, several key graph theory metrics are obtained. These metrics include clustering coefficients to characterize the local information processing capability of the network, path lengths to characterize the average distance of information transmission in the network, global efficiency to characterize the overall parallel information processing capability of the network, and local efficiency to characterize the fault tolerance capability of the network when local nodes are damaged. Multiple key graph theory indicators were obtained as predictors. The predictors and the dependent variable PANSS score were input into the constructed multiple linear regression model to determine the linear relationship between the multiple key graph theory indicators and the dependent variable PANSS score. Based on the linear relationship between the multiple key graph theory indicators and the PANSS score, a schizophrenia symptom severity prediction model was obtained to predict the severity of symptoms in schizophrenia patients.
2. The method for constructing a schizophrenia symptom severity prediction model as described in claim 1, characterized in that, By traversing each node in the weighted undirected connectivity matrix, the ratio of the actual number of connections between each node's neighboring nodes to the maximum number of connections for each node is determined, and the clustering coefficient is obtained.
3. The method for constructing a schizophrenia symptom severity prediction model as described in claim 1, characterized in that, By converting the weighted undirected connection matrix into a distance matrix, the shortest path length between all node pairs in the distance matrix is determined, and the average of the shortest path lengths is taken as the path length.
4. The method for constructing a schizophrenia symptom severity prediction model as described in claim 3, characterized in that, The shortest path length between all pairs of nodes in the distance matrix is determined using a shortest path algorithm, wherein the shortest path algorithm is Dijkstra's algorithm.
5. The method for constructing a schizophrenia symptom severity prediction model as described in claim 1, characterized in that, Obtain the shortest path length between all pairs of nodes in the weighted undirected connection matrix, and take the average of the reciprocals of the shortest path lengths as the global efficiency; for each removed node, obtain the global efficiency of the subgraph formed by its neighboring nodes, and take it as the local efficiency of the removed node.
6. The method for constructing a schizophrenia symptom severity prediction model as described in claim 1, characterized in that, The historical EEG data were obtained using wavelet transform denoising and independent component analysis (ICA).
7. The method for constructing a schizophrenia symptom severity prediction model as described in claim 1, characterized in that, The key graph theory metrics were standardized using Z-score.
8. A system for constructing a predictive model for the severity of schizophrenia symptoms, characterized in that, include: The acquisition module is used to acquire the patient's historical EEG data during the exploration phase of the visuospatial memory task, and to construct a weighted undirected connectivity matrix based on the historical EEG data to reflect the connection strength between different brain regions. The determination module is used to obtain multiple key graph theory indicators based on a weighted undirected connection matrix and combined with graph theory algorithms. Among them, the multiple key graph theory indicators include clustering coefficients to characterize the local information processing capability of the network, path lengths to characterize the average distance of information transmission in the network, global efficiency to characterize the overall parallel information processing capability of the network, and local efficiency to characterize the fault tolerance capability of the network when local nodes are damaged. The module is used to take multiple key graph theory indicators as predictors, input the predictors and the dependent variable PANSS score into the constructed multiple linear regression model to determine the linear relationship between the multiple key graph theory indicators and the dependent variable PANSS score; based on the linear relationship between the multiple key graph theory indicators and the PANSS score, a schizophrenia symptom severity prediction model is obtained to predict the severity of schizophrenia symptoms in patients.