Clinical psychological treatment effect evaluation method and system based on machine learning

Abstract

The invention relates to a clinical psychological treatment effect evaluation method and system based on machine learning, and the method comprises the steps: obtaining biological signals, behavior patterns and psychological state data of a patient during treatment, constructing a three-dimensional tensor structure with aligned timestamps, and carrying out the standardization of the three-dimensional tensor structure to generate a multi-dimensional data matrix; dimensionality reduction, reconstruction and verification are carried out on the matrix through a deep auto-encoder network, and unified feature vector representation is output; utilizing an improved support vector regression algorithm to establish a nonlinear mapping model of the feature vector and the curative effect score; according to the method, the curative effect score is predicted in real time, when the score is abnormal or fluctuation exceeds a threshold value, a personalized treatment scheme optimization mechanism based on the knowledge graph is triggered, deep fusion of multi-source heterogeneous data is achieved, evaluation precision and treatment adaptability are improved through a dynamic optimization mechanism, and intelligent decision support is provided for clinical psychological intervention.

Description

Technical Field

[0001] This invention belongs to the field of intelligent healthcare, and in particular relates to a method and system for evaluating the effectiveness of clinical psychological treatment based on machine learning. Background Technology

[0002] With the development of artificial intelligence in the healthcare field, a multimodal data-based technique for evaluating the effectiveness of psychotherapy has emerged. This technique integrates multi-source data, including physiological, behavioral, and psychological data, to achieve quantitative analysis of treatment effects. In traditional clinical psychotherapy, efficacy evaluation primarily relies on physician experience and patient self-assessment scales (such as HAMD and PHQ-9). The core process includes periodically collecting patients' subjective symptom descriptions, combining scale scores to determine interim efficacy, and adjusting treatment plans based on discrete assessment results. However, current evaluation methods have significant drawbacks: strong subjectivity and limited dimensions: scale assessments are easily influenced by patient description biases and physician subjective experience, lacking the fusion analysis of objective data such as biosignals (e.g., heart rate variability, EEG) and behavioral patterns (e.g., social frequency, movement trajectory), making it difficult to comprehensively reflect the true efficacy. Significant assessment lag: traditional methods rely on assessments at discrete time points (e.g., once a week), failing to capture dynamic changes during treatment in real time, leading to severe delays in adjusting treatment plans. Insufficient personalized decision support: Historical case data is not structured and correlated, making it difficult for doctors to quickly retrieve the optimal treatment plan for similar cases. Efficacy optimization relies on a trial-and-error mechanism, which is inefficient. Summary of the Invention

[0003] Therefore, it is necessary to provide a machine learning-based method and system for evaluating the effectiveness of clinical psychotherapy that can solve the above problems.

[0004] Firstly, this application provides a machine learning-based method for evaluating the effectiveness of clinical psychotherapy, including:

[0005] Data on patients' biosignals, behavioral patterns, and psychological states during treatment were acquired. A three-dimensional tensor data structure was constructed based on timestamp alignment, and a multi-dimensional data matrix was generated after Z-score normalization.

[0006] A deep autoencoder network is used to sequentially perform dimensionality reduction, reconstruction and verification on a multidimensional data matrix, and output a unified feature vector representation.

[0007] Based on the unified feature vector representation, a nonlinear mapping model is constructed using the support vector regression algorithm to establish a nonlinear mapping relationship between the unified feature vector representation and the treatment effect score.

[0008] The current unified feature vector is input into the nonlinear mapping model, and the corresponding treatment effect prediction score is calculated based on the nonlinear mapping relationship.

[0009] In one embodiment, a deep autoencoder network is used to sequentially perform dimensionality reduction, reconstruction, and verification processes on a multidimensional data matrix, outputting a unified feature vector representation, including:

[0010] The encoder of a deep autoencoder network maps a multidimensional data matrix to a low-dimensional feature space to generate feature vectors.

[0011] The reconstructed matrix is ​​obtained by reconstructing the feature vectors through the decoder of a deep autoencoder network;

[0012] The error between the reconstructed matrix and the multidimensional data matrix is ​​calculated. When the error is lower than a preset threshold, the feature vector is output as a unified feature vector representation that integrates multi-source information.

[0013] In one embodiment, the method further includes:

[0014] Real-time monitoring of treatment efficacy prediction scores triggers an adjustment mechanism when any of the following conditions are met:

[0015] The treatment outcome prediction score is lower than the preset score threshold;

[0016] Based on the scoring sequence of N consecutive time points within a preset time window, calculate the rate of change and the amplitude of fluctuation of the scores. If the rate of change is lower than the preset change threshold and the amplitude of fluctuation is higher than the preset fluctuation threshold.

[0017] In response to the trigger adjustment mechanism:

[0018] Using the current unified feature vector as the query vector, cases with similarity higher than the similarity threshold are retrieved from the pre-built knowledge graph; the knowledge graph contains feature vectors, treatment plans, and effect scores of historical cases.

[0019] Extract the treatment plan with the best treatment effect score from the retrieved similar cases and generate a personalized treatment optimization plan.

[0020] In one embodiment, the steps for constructing the pre-built knowledge graph include:

[0021] Obtain historical case datasets, which include patients' biosignal data, behavioral pattern data, psychological state data, treatment plans, and actual treatment effect scores;

[0022] Perform the following operations on the historical case dataset:

[0023] The entity types are divided into patient attribute entities, treatment plan entities, and effect score entities;

[0024] Establish a treatment relationship between the patient attribute entity and the treatment plan entity, and establish a corresponding effect relationship between the treatment plan entity and the effect score entity;

[0025] A graph structure is constructed based on entity type and relationship type, and historical case data is mapped to graph nodes and edge relationships to generate a pre-built knowledge graph.

[0026] In one embodiment, a nonlinear mapping model is constructed using a support vector regression algorithm to establish a nonlinear mapping relationship between a unified feature vector and a treatment effect score, including:

[0027] Based on the unified feature vector representation, the kernel function of the corresponding support vector regression algorithm is selected;

[0028] The optimal combination of hyperparameters for the support vector regression model is determined using a parameter optimization algorithm.

[0029] By using the optimal combination of hyperparameters and kernel functions to train a support vector regression model, a nonlinear mapping relationship between a unified feature vector representation and treatment effect scores is established.

[0030] In one embodiment, the support vector regression algorithm employs the following improved spatiotemporal weighted objective function:

[0031]

[0032] Where w represents the weight vector, b represents the bias term, C represents the regularization parameter, and L ∈ (y i f(x) i ))=max(0,|y i -f(x i )|-∈) denotes the ∈-insensitive loss function, λ represents the time decay factor, and λ represents the time decay coefficient. β represents the confidence weight for multi-source data, D represents the total number of categories in the data sources, and β represents the confidence weight for multi-source data. d f(x) represents the reliability coefficient of the d-th type of data source. i ) represents a nonlinear mapping model for x i The predicted output value, t i Represents a timestamp, t current Indicates the current time, x i Represents a unified eigenvector, y i This indicates the treatment effectiveness score.

[0033] In one embodiment, the method further includes:

[0034] Obtain actual treatment outcome scores for patients after the implementation of the current treatment plan;

[0035] Calculate the deviation between the actual treatment effect score data and the treatment effect prediction score output by the nonlinear mapping model;

[0036] When the deviation value exceeds the preset deviation threshold, the model weight update parameters and feature extraction weight update parameters are generated based on the deviation value through an incremental learning algorithm.

[0037] By using weight update parameters and feature extraction weight update parameters, the weight parameters of the support vector regression algorithm and the feature extraction weights of the deep autoencoder network are adjusted.

[0038] Secondly, this application also provides a machine learning-based clinical psychotherapy efficacy evaluation system, including:

[0039] The data preprocessing module is used to acquire data on patients' biological signals, behavioral patterns and psychological states during treatment. It constructs a three-dimensional tensor data structure based on timestamp alignment and generates a multi-dimensional data matrix after Z-score normalization.

[0040] The feature fusion module is used to perform dimensionality reduction, reconstruction and verification on a multidimensional data matrix through a deep autoencoder network, and output a unified feature vector representation.

[0041] The model training module is used to construct a nonlinear mapping model based on a unified feature vector representation and a support vector regression algorithm, and to establish a nonlinear mapping relationship between the unified feature vector representation and the treatment effect score.

[0042] The efficacy prediction module is used to input the current unified feature vector representation into the nonlinear mapping model and calculate the corresponding treatment effect prediction score based on the nonlinear mapping relationship.

[0043] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method for evaluating the effectiveness of clinical psychological therapy based on machine learning.

[0044] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the aforementioned method for evaluating the effectiveness of clinical psychological therapy based on machine learning.

[0045] The aforementioned method, system, computer equipment, and storage medium for evaluating the effectiveness of clinical psychological treatment based on machine learning address the problem of single-dimensional traditional assessment by acquiring data on patients' biosignals, behavioral patterns, and psychological states during treatment and constructing a timestamp-aligned three-dimensional tensor structure. This achieves deep fusion of multi-source heterogeneous data. A deep autoencoder network is used to perform dimensionality reduction and reconstruction verification on the standardized multi-dimensional data matrix, automatically extracting cross-modal objective features, eliminating subjective bias, and generating a unified feature vector representation. A nonlinear mapping model between the unified feature vector and the efficacy score is established based on the support vector regression algorithm, overcoming the lag limitation of traditional discrete assessments and achieving real-time quantitative output of dynamic efficacy during treatment. The efficacy prediction score is calculated in real-time by inputting the current feature vector, providing continuous data support for clinical intervention and improving assessment efficiency and the timeliness of treatment adjustments. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a flowchart of a machine learning-based clinical psychotherapy efficacy evaluation method according to the present invention.

[0048] Figure 2 This is a structural diagram of a machine learning-based clinical psychotherapy efficacy evaluation system according to the present invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0050] This invention discloses a machine learning-based method for evaluating the effectiveness of clinical psychological treatment. The related hardware architecture includes: sensors for collecting patient biosignals (such as ECG / EEG), mobile terminals / camera devices for monitoring behavioral patterns, electronic questionnaire terminals for recording psychological states, and edge computing terminals and cloud servers equipped with deep autoencoder and support vector regression algorithms. Multi-source data streams (biosignals, behavioral video streams, and psychological scale inputs) are synchronized in real time through a sensor network. The edge computing terminal or cloud server aligns timestamps, constructs a three-dimensional tensor, performs multidimensional matrix standardization, deep autoencoder feature fusion, and SVR efficacy prediction, ultimately pushing the scoring results to the doctor's terminal.

[0051] In one embodiment, such as Figure 1 As shown, a method for evaluating the effectiveness of clinical psychological therapy based on machine learning is provided. This embodiment illustrates the application of this method to an edge computing terminal. It is understood that this method can also be applied to cloud servers, and to systems including both edge computing terminals and cloud servers, and is implemented through the interaction between the edge computing terminal and the cloud server. In this embodiment, the method includes the following steps:

[0052] S01: Acquire data on the patient's biosignals, behavioral patterns, and psychological state during treatment. Construct a three-dimensional tensor data structure based on timestamp alignment, and generate a multi-dimensional data matrix after Z-score standardization.

[0053] This approach acquires biosignals (such as heart rate and EEG), behavioral patterns (such as social frequency and movement trajectory), and psychological state data (such as scale scores and questionnaire feedback) from patients during treatment, covering three dimensions of information: physiological, behavioral, and psychological, thus addressing the problem of single-source assessment data in traditional methods. Using timestamps as a benchmark, data from different sources and frequencies are synchronized to a unified timeline, constructing a three-dimensional tensor structure of time, data type, and feature dimensions. This ensures the spatiotemporal consistency of multi-source data, laying the foundation for subsequent fusion analysis. The three-dimensional tensor is standardized using the Z-score algorithm to eliminate dimensional differences between data dimensions, transforming it into a multi-dimensional data matrix with a mean of 0 and a standard deviation of 1, improving data comparability and model training efficiency.

[0054] S02 uses a deep autoencoder network to sequentially perform dimensionality reduction, reconstruction, and verification on a multidimensional data matrix, outputting a unified feature vector representation.

[0055] This method utilizes the encoder structure of a deep autoencoder to map a standardized multidimensional data matrix to a low-dimensional feature space. Through multi-layer nonlinear transformations, it automatically extracts common features across modalities (biological signals, behavioral patterns, and psychological states), generating compressed feature vectors. A decoder then performs an inverse mapping of these compressed feature vectors to reconstruct the original dimensional data matrix, fusing them to generate a unified feature vector representation. This self-supervised learning mechanism automatically completes feature dimensionality reduction and cross-modal fusion, avoiding the subjectivity of manual feature engineering and improving the objectivity and generalization ability of data representation.

[0056] S03, based on the unified feature vector representation, uses the support vector regression algorithm to construct a nonlinear mapping model and establish a nonlinear mapping relationship between the unified feature vector representation and the treatment effect score.

[0057] Based on the nonlinear distribution characteristics of a unified feature vector, radial basis functions (RBF) or multinomial kernel functions are selected to map linearly inseparable problems in low-dimensional feature spaces to high-dimensional spaces, transforming them into linearly separable problems and overcoming the representational limitations of traditional linear models. Parameter optimization algorithms such as grid search and Bayesian optimization can be used to globally optimize hyperparameters such as regularization parameters and kernel function parameters of the support vector regression algorithm, determining the optimal parameter combination and improving the model's fitting ability and generalization performance for multi-source features. An improved spatiotemporal weighted objective function can be adopted, introducing a time decay factor and multi-source data confidence weights, assigning higher weights to recent data and reliable data sources. Simultaneously, an insensitive loss function suppresses the influence of outliers, establishing a dynamically adaptive nonlinear mapping relationship. This achieves temporal and multi-dimensional accurate modeling of treatment effects, solving the problems of insufficient linear model fitting and inadequate utilization of data temporal characteristics in traditional assessments, and providing a dynamic mapping model for real-time prediction of treatment efficacy.

[0058] S04. Input the current unified feature vector representation into the nonlinear mapping model, and calculate the corresponding treatment effect prediction score based on the nonlinear mapping relationship.

[0059] The process involves inputting a unified feature vector representation, processed by a deep autoencoder, into a trained support vector regression nonlinear mapping model. Based on the established spatiotemporal weighted nonlinear mapping relationship (i.e., the mapping rule defined by the improved objective function), the input feature vector is calculated. A time decay factor automatically assigns higher weights to recent data, multi-source confidence weights strengthen the influence of reliable data sources, and an ε-insensitive loss function suppresses prediction bias, achieving dynamic adaptive scoring inference. The output is a predicted treatment effect score corresponding to the current feature vector, which quantitatively reflects the patient's current treatment progress. By continuously inputting real-time feature vectors, dynamic tracking of treatment effects can be achieved, addressing the lag problem of traditional discrete assessments and providing immediate data support for clinical intervention.

[0060] The aforementioned machine learning-based clinical psychological treatment efficacy evaluation method acquires data on patients' biosignals, behavioral patterns, and psychological states during treatment and constructs a three-dimensional tensor structure based on timestamp alignment. This is then standardized using Z-scores to generate a multi-dimensional data matrix, addressing the limitations of traditional assessments that rely on single dimensions and subjective scales, and achieving deep fusion of multi-source heterogeneous data. A deep autoencoder network is used to reduce, reconstruct, and validate the multi-dimensional data matrix, outputting a unified feature vector representation that integrates multi-source information. This eliminates the subjective bias of manual feature engineering and enables automatic extraction of cross-modal objective features. A nonlinear mapping model incorporating a spatiotemporal weighting mechanism is constructed using a support vector regression algorithm to establish a dynamic mapping relationship between the unified feature vector and the efficacy score, overcoming the lag limitations of traditional discrete assessments. The current feature vector is input into the model to calculate the efficacy prediction score in real time. Combined with a subsequently triggered knowledge graph retrieval mechanism, this provides continuous quantitative data support for clinical intervention and optimizes personalized treatment plans, improving assessment efficiency, the timeliness of treatment adjustments, and the accuracy of decision support. This addresses the technical shortcomings of traditional methods, such as assessment lag and insufficient personalized decision-making.

[0061] In one embodiment, a deep autoencoder network is used to sequentially perform dimensionality reduction, reconstruction, and verification processes on a multidimensional data matrix, outputting a unified feature vector representation, including:

[0062] S11, the encoder of the deep autoencoder network maps the multidimensional data matrix to the low-dimensional feature space to generate feature vectors;

[0063] S12, the feature vector is reconstructed through the decoder of the deep autoencoder network to obtain the reconstruction matrix;

[0064] S13, calculate the error between the reconstructed matrix and the multidimensional data matrix. When the error is lower than a preset threshold, output the feature vector as a unified feature vector representation of the fused multi-source information.

[0065] Specifically, the encoder of a deep autoencoder network can be used to map the multidimensional data matrix generated after Z-score normalization to a low-dimensional feature space. Through multi-layer nonlinear transformation, cross-modal common features of biological signals, behavioral patterns, and psychological states are automatically extracted and feature vectors are generated. The decoder of the deep autoencoder network performs inverse mapping on these feature vectors to reconstruct the original dimensional reconstruction matrix. The error between the reconstruction matrix and the original multidimensional data matrix is ​​calculated. When the error is lower than a preset threshold, it indicates that the feature vector effectively retains the key information of the original data. This feature vector is then output as a unified feature vector representation that fuses multi-source information. This process automatically completes feature dimensionality reduction and cross-modal fusion through a self-supervised learning mechanism, avoiding the subjectivity of manual feature engineering and achieving an objective mapping from multimodal data to a unified feature space. This provides standardized input features for subsequent efficacy prediction models.

[0066] In one embodiment, the method further includes:

[0067] S21, real-time monitoring of treatment effect prediction score, triggers adjustment mechanism when any of the following conditions are met:

[0068] S21.1, The treatment effect prediction score is lower than the preset score threshold;

[0069] S21.2, Based on the scoring sequence of N consecutive time points within a preset time window, calculate the scoring change rate and fluctuation amplitude. If the change rate is lower than the preset change threshold and the fluctuation amplitude is higher than the preset fluctuation threshold.

[0070] S22, in response to the trigger adjustment mechanism:

[0071] S22.1, using the current unified feature vector as the query vector, retrieve cases with similarity higher than the similarity threshold in the pre-built knowledge graph; the knowledge graph contains feature vectors, treatment plans, and effect scores of historical cases;

[0072] S22.2 Extract the treatment plan with the best treatment effect score from the retrieved similar cases and generate a personalized treatment optimization plan.

[0073] For example, the treatment effect prediction score output by the nonlinear mapping model is tracked in real time. When the treatment effect prediction score falls below a preset score threshold, or when the rate of change of the score calculated based on a score sequence of N consecutive time points within a preset time window (N is preset according to the specific needs of clinical psychotherapy, data collection frequency, and timeliness requirements of efficacy evaluation) is lower than a preset change threshold and the fluctuation amplitude is higher than a preset fluctuation threshold, an adjustment mechanism is triggered (the preset change threshold and preset fluctuation threshold are preset based on knowledge of the clinical psychotherapy domain and the characteristics of historical efficacy data). In response to this trigger, the current unified feature vector table is used. Using the query vector as a reference, the system retrieves cases with similarity scores higher than a pre-built knowledge graph containing historical case feature vectors, treatment plans, and effect scores. These cases are set to 0.85 and can be determined or modified by domain experts or through cross-validation algorithms, taking into account the balance between case matching accuracy and recall in clinical psychological treatment. The system then extracts the treatment plan with the best treatment effect score from the retrieved similar cases and makes adaptive adjustments (such as dose optimization and adjustment of intervention combination) based on the current patient's personalized characteristics (e.g., real-time fluctuations in biosignals and dynamic changes in psychological state). This generates a personalized treatment optimization plan for the current patient.

[0074] In one embodiment, the steps for constructing the pre-built knowledge graph include:

[0075] S31, Obtain historical case dataset, which includes patients' biosignal data, behavioral pattern data, psychological state data, treatment plans, and actual treatment effect scores;

[0076] S32, Perform the following operations on the historical case dataset:

[0077] S32.1, the entity types are divided into patient attribute entities, treatment plan entities, and effect score entities;

[0078] S32.2, Establish a treatment relationship between the patient attribute entity and the treatment plan entity, and establish a corresponding effect relationship between the treatment plan entity and the effect score entity;

[0079] S32.3 Construct a graph structure based on entity type and relationship type, map historical case data into graph nodes and edge relationships, and generate a pre-constructed knowledge graph.

[0080] Specifically, the process involves acquiring historical case datasets containing patient biosignal data, behavioral pattern data, psychological state data, treatment plans, and actual treatment effect scores, covering multi-dimensional clinical information to provide structured raw data for the knowledge graph. Through semantic modeling, entity types are categorized into patient attribute entities (including patient characteristics such as biosignals, behavioral patterns, and psychological states), treatment plan entities (including specific interventions, dosages, and cycles), and effect score entities (including quantitative efficacy scores and timestamps). Treatment relationships are established between patient attribute entities and treatment plan entities to represent the specific treatment plans received by patients. Corresponding effect relationships are established between treatment plan entities and effect score entities to clarify the association mapping between treatment plans and efficacy scores. Semantic connections between data are constructed through relationship modeling. Based on entity and relationship types, a graph structure is built, mapping the various dimensions of information in the historical case data to graph nodes (entities) and edge relationships (semantic connections), generating a pre-constructed knowledge graph containing historical case feature vectors, treatment plans, and effect scores. Through entity-relationship modeling of multi-dimensional data, unstructured historical cases are transformed into a structured knowledge network, providing efficient data support for feature vector-based similar case retrieval and personalized treatment plan generation.

[0081] In one embodiment, a nonlinear mapping model is constructed using a support vector regression algorithm to establish a nonlinear mapping relationship between a unified feature vector and a treatment effect score, including:

[0082] S41, Based on the unified feature vector representation, select the kernel function corresponding to the support vector regression algorithm;

[0083] S42, the optimal combination of hyperparameters for the support vector regression model is determined through a parameter optimization algorithm;

[0084] S43. By using the optimal combination of hyperparameters and kernel functions to train the support vector regression model, a nonlinear mapping relationship between the unified feature vector representation and the treatment effect score is established.

[0085] For example, the corresponding support vector regression algorithm kernel function is selected based on the nonlinear distribution characteristics of the unified feature vector. For instance, a radial basis function (RBF) kernel or a multinomial kernel is used for high-dimensional nonlinear features. The kernel function maps the linearly inseparable problem in the low-dimensional feature space to the high-dimensional space to achieve linear separability, solving the problem of insufficient fitting ability of traditional linear models for complex features. Through parameter optimization algorithms such as grid search, Bayesian optimization, or particle swarm optimization, the regularization parameters, kernel function parameters, and loss parameters such as insensitive loss parameters of the support vector regression model are globally optimized. The optimal hyperparameter combination is determined based on the cross-validation results of historical case data, avoiding the subjectivity of manual parameter tuning and improving the model's generalization ability. The support vector regression model is trained using the determined optimal hyperparameter combination and kernel function. By introducing a spatiotemporal weighting mechanism in the objective function (including time decay factor and multi-source data confidence weight), the influence of recent data and reliable data sources is strengthened, and a dynamic nonlinear mapping relationship between the unified feature vector representation and treatment effect score is established. This achieves accurate modeling of the correlation between multi-source features and efficacy, breaking through the representation limitations of linear models in traditional discrete assessment.

[0086] In one embodiment, S51, the support vector regression algorithm employs the following improved spatiotemporal weighted objective function:

[0087]

[0088] Where w represents the weight vector, b represents the bias term, C represents the regularization parameter, and L ∈ (y i f(x) i ))=max(0,|y i -f(x i )|-∈) denotes the ∈-insensitive loss function, λ represents the time decay factor, and λ represents the time decay coefficient. β represents the confidence weight for multi-source data, D represents the total number of categories in the data sources, and β represents the confidence weight for multi-source data. d f(x) represents the reliability coefficient of the d-th type of data source. i ) represents a nonlinear mapping model for x i The predicted output value, t i Represents a timestamp, t current Indicates the current time, x i Represents a unified eigenvector, y i This indicates the treatment effectiveness score.

[0089] Specifically, The L2 regularization term of the weight vector is used to control model complexity and avoid overfitting; the spatiotemporal weighted loss term... The loss function introduces spatiotemporal weights, where: time decay factor Historical data is weighted according to time distance using an exponential decay mechanism, while recent data (t) i Approximately current time t current Higher weights address the issue of traditional models' delayed response to dynamic changes over time, and multi-source data confidence weights. The L2 norm of each data source (total number of categories is D) is aggregated and multiplied by the reliability coefficient β. d Differentiated weighting of reliability is applied to multi-source data such as biological signals and behavioral patterns to enhance the influence of high-reliability data sources and suppress interference from noisy data; ∈-insensitive loss function L ∈ (y i f(x) i ))=max(0,|y i -f(x i The method applies a penalty only to data points where the prediction error exceeds a threshold, improving the model's robustness to outliers. Through time-dimensional adaptation, multi-source data fusion optimization, and enhanced robustness and generalization capabilities, a dynamic and accurate nonlinear mapping relationship between a unified feature vector and efficacy scores is established, providing a quantitative basis for real-time efficacy prediction and addressing the lag problem in traditional assessments.

[0090] In one embodiment, the method further includes:

[0091] S61, Obtain the actual treatment effect score data of patients after the implementation of the current treatment plan;

[0092] S62, Calculate the deviation between the actual treatment effect score data and the treatment effect prediction score output by the nonlinear mapping model;

[0093] S63, When the deviation value exceeds the preset deviation threshold, the model weight update parameters and feature extraction weight update parameters are generated based on the deviation value through an incremental learning algorithm.

[0094] S63 adjusts the weight parameters of the support vector regression algorithm and the feature extraction weights of the deep autoencoder network by using weight update parameters and feature extraction weight update parameters.

[0095] For example, the actual treatment effect score data of patients obtained through clinical evaluation (such as scale scores, doctor's diagnosis) after the implementation of the current treatment plan is acquired. This data serves as the benchmark for the accuracy of model prediction. The deviation value between the actual treatment effect score data and the treatment effect prediction score output by the nonlinear mapping model is calculated. The model error is characterized by quantifying the difference between the predicted value and the true value. When the deviation value exceeds a deviation threshold preset based on historical data distribution and clinically acceptable error range, incremental learning algorithms (such as online gradient descent, incremental support vector machine) are used to generate model weight update parameters and feature extraction weight update parameters. The incremental learning algorithm updates the deviation-related parameters locally only through iteration, avoiding retraining the entire model and improving optimization efficiency. Using the generated weight update parameters and feature extraction weight update parameters, the weight parameters w and b of the support vector regression algorithm and the feature extraction weights of the deep autoencoder network are adjusted synchronously to achieve end-to-end model optimization from feature extraction to efficacy prediction. By analyzing the discrepancies between actual treatment data and predicted results, this approach addresses the challenge of traditional static models adapting to individual patient differences and temporal changes. Through incremental learning, the model parameters are dynamically updated, enhancing the system's adaptability to new data and improving predictive accuracy. This ensures the assessment model continuously optimizes as treatment progresses, providing more precise dynamic decision support for clinical psychotherapy.

[0096] The aforementioned machine learning-based clinical psychological treatment efficacy evaluation method acquires data on patients' biosignals, behavioral patterns, and psychological states during treatment and constructs a timestamp-aligned three-dimensional tensor structure. This is then standardized using Z-scores to generate a multidimensional data matrix. A deep autoencoder network is used to reduce the matrix's dimensionality, reconstruct it, and validate it to output a unified feature vector representation. A nonlinear mapping model is constructed using a support vector regression algorithm with a spatiotemporal weighting mechanism to achieve a dynamic correlation between the unified feature vector and the efficacy score. The current feature vector is input into the model to calculate the predicted score in real time. When the score is abnormal or fluctuates beyond a threshold, a knowledge graph-based similar case retrieval is triggered to extract the optimal treatment plan. Simultaneously, the deviation between the actual efficacy and the predicted score is fed back through an incremental learning algorithm to dynamically update the model weights and feature extraction parameters. This approach effectively solves the technical problems of traditional assessments, such as strong subjectivity, single dimension, significant assessment lag, and insufficient personalized decision support, through deep fusion of multi-source heterogeneous data, automatic extraction of cross-modal features, real-time quantitative prediction of efficacy, personalized plan generation based on knowledge graphs, and dynamic model optimization mechanisms. It provides precise, real-time, and adaptive intelligent decision support for clinical psychological intervention.

[0097] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0098] Based on the same inventive concept, this application also provides a machine learning-based clinical psychotherapy efficacy evaluation system for implementing the aforementioned machine learning-based clinical psychotherapy efficacy evaluation method. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more machine learning-based clinical psychotherapy efficacy evaluation system embodiments provided below can be found in the above-described limitations of the machine learning-based clinical psychotherapy efficacy evaluation method, and will not be repeated here.

[0099] In one exemplary embodiment, such as Figure 2 As shown, a machine learning-based clinical psychotherapy efficacy evaluation system is provided, including:

[0100] The data preprocessing module 101 is used to acquire biosignals, behavioral patterns and psychological state data of patients during treatment, construct a three-dimensional tensor data structure based on timestamp alignment, and generate a multi-dimensional data matrix after Z-score normalization.

[0101] The feature fusion module 102 is used to perform dimensionality reduction, reconstruction and verification processing on the multidimensional data matrix through a deep autoencoder network, and output a unified feature vector representation.

[0102] The model training module 103 is used to construct a nonlinear mapping model based on a unified feature vector representation and a support vector regression algorithm, and to establish a nonlinear mapping relationship between the unified feature vector representation and the treatment effect score.

[0103] The efficacy prediction module 104 is used to input the current unified feature vector representation into the nonlinear mapping model and calculate the corresponding treatment effect prediction score based on the nonlinear mapping relationship.

[0104] In one embodiment, the feature fusion module 102 is further configured to:

[0105] The encoder of a deep autoencoder network maps a multidimensional data matrix to a low-dimensional feature space to generate feature vectors.

[0106] The reconstructed matrix is ​​obtained by reconstructing the feature vectors through the decoder of a deep autoencoder network;

[0107] The error between the reconstructed matrix and the multidimensional data matrix is ​​calculated. When the error is lower than a preset threshold, the feature vector is output as a unified feature vector representation that integrates multi-source information.

[0108] In one embodiment, the efficacy prediction module 104 is further configured to:

[0109] Real-time monitoring of treatment efficacy prediction scores triggers an adjustment mechanism when any of the following conditions are met:

[0110] The treatment outcome prediction score is lower than the preset score threshold;

[0111] Based on the scoring sequence of N consecutive time points within a preset time window, calculate the rate of change and the amplitude of fluctuation of the scores. If the rate of change is lower than the preset change threshold and the amplitude of fluctuation is higher than the preset fluctuation threshold.

[0112] In response to the trigger adjustment mechanism:

[0113] Using the current unified feature vector as the query vector, cases with similarity higher than the similarity threshold are retrieved from the pre-built knowledge graph; the knowledge graph contains feature vectors, treatment plans, and effect scores of historical cases.

[0114] Extract the treatment plan with the best treatment effect score from the retrieved similar cases and generate a personalized treatment optimization plan.

[0115] In one embodiment, the efficacy prediction module 104 is further configured to:

[0116] Obtain historical case datasets, which include patients' biosignal data, behavioral pattern data, psychological state data, treatment plans, and actual treatment effect scores;

[0117] Perform the following operations on the historical case dataset:

[0118] The entity types are divided into patient attribute entities, treatment plan entities, and effect score entities;

[0119] Establish a treatment relationship between the patient attribute entity and the treatment plan entity, and establish a corresponding effect relationship between the treatment plan entity and the effect score entity;

[0120] A graph structure is constructed based on entity type and relationship type, and historical case data is mapped to graph nodes and edge relationships to generate a pre-built knowledge graph.

[0121] In one embodiment, the model training module 103 is further configured to:

[0122] Based on the unified feature vector representation, the kernel function of the corresponding support vector regression algorithm is selected;

[0123] The optimal combination of hyperparameters for the support vector regression model is determined using a parameter optimization algorithm.

[0124] By using the optimal combination of hyperparameters and kernel functions to train a support vector regression model, a nonlinear mapping relationship between a unified feature vector representation and treatment effect scores is established.

[0125] In one embodiment, the module training module is also used to construct a support vector regression algorithm using the following improved spatiotemporal weighted objective function:

[0126]

[0127] Where w represents the weight vector, b represents the bias term, C represents the regularization parameter, and L ∈ (y i f(x) i ))=max(0,|y i -f(x i )|-∈) denotes the ∈-insensitive loss function, λ represents the time decay factor, and λ represents the time decay coefficient. β represents the confidence weight for multi-source data, D represents the total number of categories in the data sources, and β represents the confidence weight for multi-source data. d f(x) represents the reliability coefficient of the d-th type of data source. i ) represents a nonlinear mapping model for x i The predicted output value, t i Represents a timestamp, t current Indicates the current time, x i Represents a unified eigenvector, y i This indicates the treatment effectiveness score.

[0128] In one embodiment, the system further includes a feedback optimization module for:

[0129] Obtain actual treatment outcome scores for patients after the implementation of the current treatment plan;

[0130] Calculate the deviation between the actual treatment effect score data and the treatment effect prediction score output by the nonlinear mapping model;

[0131] When the deviation value exceeds the preset deviation threshold, the model weight update parameters and feature extraction weight update parameters are generated based on the deviation value through an incremental learning algorithm.

[0132] By using weight update parameters and feature extraction weight update parameters, the weight parameters of the support vector regression algorithm and the feature extraction weights of the deep autoencoder network are adjusted.

[0133] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of a machine learning-based clinical psychotherapy efficacy evaluation method as described above.

[0134] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0135] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0136] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for evaluating the effectiveness of clinical psychotherapy based on machine learning, characterized in that, The method includes: Data on patients' biosignals, behavioral patterns, and psychological states during treatment were acquired. A three-dimensional tensor data structure was constructed based on timestamp alignment, and a multi-dimensional data matrix was generated after Z-score normalization. The multidimensional data matrix is ​​sequentially subjected to dimensionality reduction, reconstruction, and verification processes using a deep autoencoder network, and a unified feature vector representation is output. Based on the unified feature vector representation, a nonlinear mapping model is constructed using the support vector regression algorithm to establish a nonlinear mapping relationship between the unified feature vector representation and the treatment effect score. The current unified feature vector representation is input into the nonlinear mapping model, and the corresponding treatment effect prediction score is calculated based on the nonlinear mapping relationship.

2. The method according to claim 1, characterized in that, The process of sequentially performing dimensionality reduction, reconstruction, and verification on the multidimensional data matrix using a deep autoencoder network to output a unified feature vector representation includes: The encoder of the deep autoencoder network maps the multidimensional data matrix to a low-dimensional feature space to generate feature vectors. The feature vector is reconstructed using the decoder of the deep autoencoder network to obtain the reconstruction matrix; Calculate the error between the reconstructed matrix and the multidimensional data matrix. When the error is lower than a preset threshold, output the feature vector as a unified feature vector representation that integrates multi-source information.

3. The method according to claim 1, characterized in that, The method further includes: Real-time monitoring of treatment efficacy prediction scores triggers an adjustment mechanism when any of the following conditions are met: The predicted treatment effect score is lower than a preset score threshold; Based on the scoring sequence of N consecutive time points within a preset time window, calculate the scoring change rate and fluctuation amplitude. If the change rate is lower than a preset change threshold and the fluctuation amplitude is higher than a preset fluctuation threshold. In response to the aforementioned trigger adjustment mechanism: Using the current unified feature vector as the query vector, cases with similarity higher than a similarity threshold are retrieved from the pre-constructed knowledge graph; the knowledge graph contains feature vectors, treatment plans, and effect scores of historical cases. Extract the treatment plan with the best treatment effect score from the retrieved similar cases and generate a personalized treatment optimization plan.

4. The method according to claim 3, characterized in that, The construction steps of the pre-built knowledge graph include: Obtain a historical case dataset, which includes patients' biosignal data, behavioral pattern data, psychological state data, treatment plans, and actual treatment effect scores; Perform the following operations on the historical case dataset: The entity types are divided into patient attribute entities, treatment plan entities, and effect score entities; Establish a treatment relationship between the patient attribute entity and the treatment plan entity, and establish a corresponding effect relationship between the treatment plan entity and the effect score entity; Based on the entity type and relationship type, a graph structure is constructed, and historical case data is mapped to graph nodes and edge relationships to generate the pre-constructed knowledge graph.

5. The method according to claim 1, characterized in that, The step of constructing a nonlinear mapping model using the support vector regression algorithm to establish a nonlinear mapping relationship between the unified feature vector and the treatment effect score includes: Based on the unified feature vector representation, the kernel function of the corresponding support vector regression algorithm is selected; The optimal combination of hyperparameters for the support vector regression model is determined using a parameter optimization algorithm. The support vector regression model is trained using the optimal hyperparameter combination and kernel function to establish a nonlinear mapping relationship between the unified feature vector representation and the treatment effect score.

6. The method according to claim 5, characterized in that, The support vector regression algorithm employs the following improved spatiotemporal weighted objective function: Where w represents the weight vector, b represents the bias term, C represents the regularization parameter, and L ∈ (y i f(x) i ))=max(0,|y i -f(x i )|-∈) denotes the ∈-insensitive loss function, λ represents the time decay factor, and λ represents the time decay coefficient. β represents the confidence weight for multi-source data, D represents the total number of categories in the data sources, and β represents the confidence weight for multi-source data. d f(x) represents the reliability coefficient of the d-th type of data source. i ) represents a nonlinear mapping model for x i The predicted output value, t i Represents a timestamp, t current Indicates the current time, x i Represents a unified eigenvector, y i This indicates the treatment effectiveness score.

7. The method according to claim 5, characterized in that, The method further includes: Obtain actual treatment outcome scores for patients after the implementation of the current treatment plan; Calculate the deviation between the actual treatment effect score data and the treatment effect prediction score output by the nonlinear mapping model; When the deviation value exceeds a preset deviation threshold, model weight update parameters and feature extraction weight update parameters are generated based on the deviation value using an incremental learning algorithm. The weight parameters of the support vector regression algorithm and the feature extraction weights of the deep autoencoder network are adjusted using the weight update parameters and feature extraction weight update parameters.

8. A machine learning-based clinical psychotherapy efficacy evaluation system, characterized in that, The system includes: The data preprocessing module is used to acquire data on patients' biological signals, behavioral patterns and psychological states during treatment. It constructs a three-dimensional tensor data structure based on timestamp alignment and generates a multi-dimensional data matrix after Z-score normalization. The feature fusion module is used to sequentially perform dimensionality reduction, reconstruction and verification processing on the multidimensional data matrix through a deep autoencoder network, and output a unified feature vector representation. The model training module is used to construct a nonlinear mapping model based on the unified feature vector representation and the support vector regression algorithm, and to establish a nonlinear mapping relationship between the unified feature vector representation and the treatment effect score. The efficacy prediction module is used to input the current unified feature vector representation into the nonlinear mapping model and calculate the corresponding treatment effect prediction score based on the nonlinear mapping relationship.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images