A temporal lobe epilepsy resection surgery efficacy evaluation system based on brain image fusion features

Abstract

The present invention discloses a temporal lobe epilepsy resection surgery efficacy evaluation system based on brain image fusion features. The system collects and pre-processes multimodal brain imaging data before and after surgery, including T1-weighted images, diffusion tensor imaging (DTI), and functional magnetic resonance imaging (fMRI), and uses multi-time series analysis methods such as recurrent neural networks (RNN) and long short-term memory networks (LSTM) to generate a postoperative recovery model. The personalized efficacy evaluation module combines the patient's clinical characteristic data. The report generation and visualization module uses visualization technology to generate easy-to-understand reports from the evaluation data for the medical team to review and make decisions. Through multimodal data fusion and advanced algorithm models, the present invention improves the accuracy and reliability of efficacy evaluation, comprehensively reflects the changes in the patient's brain structure and function, has broad clinical application prospects, and significantly improves the patient's quality of life after surgery.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing and analysis, and in particular to a temporal lobe epilepsy resection surgery efficacy evaluation system based on brain image fusion features. Background Art

[0002] Temporal lobe epilepsy (TLE) is a common, medically intractable form of epilepsy involving key brain structures such as the amygdala. Seizures are typically controlled through surgical resection of the epileptic focus. As a serious brain disorder, epilepsy's diagnosis and efficacy assessment have a significant impact on patients' quality of life. Currently, the efficacy of TLE resection surgery primarily relies on preoperative and postoperative clinical symptom observation and analysis of single-modality brain imaging data, such as T1-weighted images or functional magnetic resonance imaging (fMRI). However, single-modality imaging analysis often fails to fully reflect changes in a patient's brain structure and function, resulting in inaccurate and unreliable efficacy assessments. While existing computer-assisted diagnosis systems can provide some diagnostic support, they lack the ability to integrate and analyze multimodal data. Furthermore, while medical data mining technology has demonstrated excellent performance in analyzing individual patient cases, its application in personalized efficacy assessment remains underdeveloped. Existing assessment methods also lack the ability to dynamically adjust and conduct multi-stage comprehensive analysis, making them inadequate for addressing complex brain disease states.

[0003] Therefore, there is an urgent need for a system that can integrate multimodal brain imaging data and use advanced algorithm models for dynamic and comprehensive evaluation to improve the evaluation level of the efficacy of temporal lobe epilepsy resection surgery and meet the needs of medical expert systems and medical data mining in the diagnosis and evaluation of brain diseases. Summary of the Invention

[0004] The purpose of the present invention is to provide a temporal lobe epilepsy resection surgery efficacy evaluation system based on brain image fusion features, which has the advantages of multimodal data fusion, dynamic feedback loop and staged comprehensive evaluation, and solves the problems of insufficient accuracy of efficacy evaluation and incomplete data analysis in the existing technology.

[0005] To achieve the above objectives, the present invention provides the following technical solutions: a system for evaluating the efficacy of temporal lobe epilepsy resection surgery based on brain image fusion features, the system comprising the following modules: a) a multimodal image data acquisition and preprocessing module responsible for acquiring and preprocessing preoperative and postoperative T1-weighted images, diffusion tensor imaging, and functional magnetic resonance imaging data, with the preprocessing steps including denoising, normalization, motion correction, and spatial smoothing;

[0006] b) a dynamic image fusion and analysis module, which receives the image data preprocessed by the multimodal image data acquisition and preprocessing module, applies a multi-time series analysis method to generate a postoperative recovery model, and transmits the postoperative recovery model to the personalized efficacy evaluation module;

[0007] c) A personalized efficacy evaluation module, which combines the postoperative recovery model generated by the dynamic image fusion and analysis module with the patient characteristic data to analyze and generate a personalized postoperative recovery report, which is then passed to the anomaly detection and intelligent early warning module;

[0008] d) Anomaly Detection and Intelligent Early Warning Module, which monitors abnormalities in the postoperative recovery report generated by the personalized efficacy evaluation module in real time, implements error compensation, issues early warning notifications, and transmits the detection results to the report generation and visualization module;

[0009] e) Data storage and management module, responsible for storing and managing multimodal imaging data, analysis results and patient information;

[0010] f) Report generation and visualization module, which receives data transmitted by the personalized efficacy evaluation module and the anomaly detection and intelligent early warning module, and generates a visualization report showing the postoperative recovery trend and personalized indicators.

[0011] Preferably, the multimodal image data acquisition and preprocessing module further comprises the following steps:

[0012] a) Convert the raw image data of preoperative and postoperative T1-weighted images, diffusion tensor imaging, and functional magnetic resonance imaging into a standard format; b) De-noise the converted image data using wavelet transform or Gaussian filtering; c) Standardize the denoised image data and calculate the mean of the image data and standard deviation , and apply the normalization formula: ,in is the mean of the image data, is the standard deviation of the image data, is the normalized pixel value; d) head motion correction is performed, and an edge detection-based algorithm is used to identify and correct head motion; e) the corrected image data is spatially smoothed, and a convolution operation is performed using a three-dimensional Gaussian kernel.

[0013] Preferably, the dynamic image fusion and analysis module further comprises the following steps:

[0014] a) receiving standardized image data from a multimodal image data preprocessing module; b) applying a multi-time series analysis method, using time series analysis or a recurrent neural network model to generate a postoperative recovery model; c) outputting the generated postoperative recovery model to a personalized efficacy evaluation module.

[0015] Preferably, the personalized efficacy evaluation module further comprises the following steps: a) receiving the postoperative recovery model output by the dynamic image fusion and analysis module; b) combining the patient's clinical characteristic data and applying a comprehensive evaluation algorithm to generate a personalized postoperative recovery report. The specific calculation formula is as follows:

[0016] in 、 、 is the dynamic adjustment coefficient; 、 is the error correction term;

[0017] c) outputting the personalized postoperative recovery report to the abnormality detection and intelligent early warning module.

[0018] Preferably, the anomaly detection and intelligent early warning module further comprises the following steps:

[0019] a) Real-time monitoring of abnormalities in the postoperative recovery report generated by the personalized efficacy evaluation module; b) Application of the error propagation and compensation formula for anomaly detection, the specific formula is as follows:

[0020] ,in is the error propagation amount, is the error correction coefficient; Control the scope of error propagation; c) Issue early warning notifications based on the test results and transmit the test results to the report generation and visualization module.

[0021] Preferably, the data storage and management module further comprises the following steps:

[0022] a) Stores the original and pre-processed versions of multimodal imaging data; b) Stores the postoperative recovery model generated by the dynamic image fusion and analysis module; c) Stores the evaluation report and related parameters generated by the personalized efficacy evaluation module; d) Ensures data security and privacy protection through encrypted storage and access control mechanisms.

[0023] Preferably, the report generation and visualization module further includes the following steps: a) receiving the detection results from the anomaly detection and intelligent early warning module; b) generating a postoperative recovery trend chart and a personalized indicator chart based on the evaluation results and detection data through data visualization technology; c) supporting report output in multiple formats, including PDF, HTML and interactive visualization interface.

[0024] Preferably, the multimodal image data acquisition and preprocessing module further includes the following algorithm steps: a) performing preliminary measurement calibration, the specific calculation formula is as follows:

[0025] ,in, is the normalized pixel value; is the weight of image preprocessing; is the adjustment coefficient of environmental parameters; is the environmental parameter;

[0026] b) Perform secondary precision calibration. The specific calculation formula is as follows:

[0027] ,in is the dynamic weight of the external parameter; is an external parameter; is the adjustment coefficient, which is used to control the impact of input on output; The function is used for nonlinear correction and smoothing of excessively large input values.

[0028] Preferably, the dynamic image fusion and analysis module further includes the following algorithm steps:

[0029] a) Use recurrent neural network models to perform time series prediction on multimodal imaging data;

[0030] b) Applying long short-term memory network structures to capture long-term dependencies in image data;

[0031] c) Generate a postoperative recovery model as input for personalized efficacy assessment.

[0032] Preferably, the personalized efficacy evaluation module further includes the following algorithm steps:

[0033] a) Obtain the patient's clinical characteristics, including but not limited to age, gender, medical history, and preoperative seizure frequency;

[0034] b) combining the patient's clinical characteristics data with the postoperative image restoration model to generate a comprehensive evaluation value using a weighted average method;

[0035] c) Based on the comprehensive evaluation value, the support vector machine classifier is used to predict the patient's postoperative seizure risk;

[0036] d) Implement a phased evaluation, specifically including: i. Generate performance indicators for each phase, calculated as follows: ,in is the performance index of stage i; is the number of samples for each stage; is the evaluation value of the jth sample in the i-th stage; is the reference value of the jth sample;

[0037] ii. Comprehensive evaluation index, calculated using the following formula: ,in The weight of each stage;

[0038] e) Implement feedback loops, including:

[0039] i. Receive accurate calibration results ;

[0040] ii. Adjust the input data according to the following formula , forming a feedback loop:

[0041] ,in is the adjusted input data; is the original input data; is the feedback gain coefficient;

[0042] iii. Adjust the input data Re-input into the multimodal image data acquisition and preprocessing module to dynamically optimize the evaluation results.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The present invention provides more comprehensive brain structure and function information by integrating multiple brain imaging data such as T1-weighted images, DTI and fMRI, thereby improving the accuracy of efficacy evaluation.

[0045] The system introduces a feedback loop to dynamically optimize the evaluation results by adjusting the input data, thereby enhancing the adaptability and real-time performance of the system and ensuring the reliability of the evaluation results.

[0046] The present invention implements a staged evaluation and obtains the final comprehensive evaluation index through the weighted sum of the performance indicators of each stage, thereby realizing the comprehensive analysis of multi-level data and improving the accuracy and consistency of the evaluation.

[0047] Through the error propagation and compensation model, the system can effectively adjust the difference between the preliminary calibration results and the reference value, minimize the error in the evaluation process, and improve the credibility of the evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is the overall architecture diagram of the system of the present invention;

[0049] Figure 2 This is a flowchart of multimodal image data acquisition and preprocessing of the present invention;

[0050] Figure 3 This is a workflow diagram of the dynamic image fusion and analysis module of the present invention;

[0051] Figure 4 This is a workflow diagram of the personalized efficacy evaluation module of the present invention;

[0052] Figure 5 This is a workflow diagram of the anomaly detection and intelligent early warning module of the present invention;

[0053] Figure 6 This is a schematic diagram of the data storage and management module architecture of the present invention;

[0054] Figure 7 A workflow diagram of the report generation and visualization module of the present invention;

[0055] Figure 8 This is the initial measurement calibration and secondary precision calibration process of the present invention;

[0056] Figure 9 This is a structural diagram of the recursive neural network and long short-term memory network of the present invention;

[0057] Figure 10 This is a feedback loop mechanism diagram of the personalized efficacy evaluation module of the present invention. DETAILED DESCRIPTION

[0058] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0059] See also Figures 1 to 10 This invention provides a temporal lobe epilepsy resection surgery efficacy assessment system based on brain image fusion features. This system aims to accurately and comprehensively evaluate the efficacy of temporal lobe epilepsy resection surgery by integrating multimodal brain imaging data and utilizing advanced algorithmic models. The system primarily consists of six modules: multimodal image data acquisition and preprocessing module; dynamic image fusion and analysis module; personalized efficacy assessment module; anomaly detection and intelligent early warning module; data storage and management module; and report generation and visualization module.

[0060] The Multimodal Imaging Data Acquisition and Preprocessing module is responsible for collecting different types of brain imaging data obtained from patients before and after surgery and performing a series of preprocessing steps on these data to ensure data quality and consistency. The core task of this module is to convert the raw imaging data into a standardized format suitable for subsequent analysis and eliminate noise and motion artifacts that may affect the evaluation results. Its specific implementation steps include:

[0061] 1. Data Acquisition: T1-weighted images: Provide high-resolution images of brain structure, clearly showing anatomical details. Diffusion Tensor Imaging (DTI): Used to analyze the direction and integrity of white matter fiber tracts and assess changes in nerve fibers. Functional Magnetic Resonance Imaging (fMRI): Monitors brain activity and assesses postoperative functional recovery.

[0062] 2. Data preprocessing: Data format conversion: Convert the raw image data of different modalities into a unified standard format (such as NIfTI format) for subsequent processing and analysis. Denoising: Use wavelet transform or Gaussian filtering methods to remove random noise in the image and improve image quality. Wavelet transform: Effectively remove noise in different frequency ranges through multi-scale decomposition. Gaussian filtering: Use Gaussian kernel to smooth the image and reduce high-frequency noise. Standardization: Calculate the mean of the image data and standard deviation , and apply the normalization formula: ,in is the mean of the image data, is the standard deviation of the image data, is the standardized pixel value; this standardization step ensures the comparability of imaging data between different patients and eliminates the impact of individual differences.

[0063] Head motion correction: An edge-detection-based algorithm identifies and corrects artifacts caused by the patient's head movement during image acquisition, ensuring geometric consistency. An edge detection algorithm identifies edge features in the image to detect the degree of head motion and apply appropriate corrections. Spatial smoothing: A convolution operation using a three-dimensional Gaussian kernel is used to spatially smooth the corrected image data, further improving image quality and reducing residual noise and artifacts.

[0064] Multimodal data integration: By integrating T1-weighted images, DTI and fMRI data, the patient's brain structure and functional status are fully reflected, providing a rich information basis for subsequent evaluation. Efficient preprocessing process: Using denoising, standardization, head motion correction and spatial smoothing methods, the quality and consistency of image data are significantly improved, ensuring the accuracy and reliability of the evaluation results. Parameters are clear and traceable: Parameters in all preprocessing steps (such as mean and standard deviation ) have clear sources and calculation methods to ensure the rationality and traceability of the calculation process.

[0065] Furthermore, the dynamic image fusion and analysis module receives pre-processed multimodal image data and applies multi-time series analysis methods to generate a postoperative recovery model. The core of this module is to conduct in-depth analysis of image data using advanced neural network algorithms (recurrent neural networks (RNNs) and long short-term memory (LSTMs)) to capture dynamic changes during postoperative recovery. Its specific implementation steps include:

[0066] 1. Data Reception and Fusion: Receiving Preprocessed Data: The Dynamic Image Fusion and Analysis Module receives standardized T1-weighted images, DTI, and fMRI data from the Multimodal Image Data Acquisition and Preprocessing Module. Data Fusion: This aligns the imaging data from different modalities spatially and temporally, integrating them into a unified multidimensional dataset for subsequent analysis and modeling.

[0067] 2. Application of Multiple Time Series Analysis Methods: Recurrent Neural Networks (RNNs): Suitable for processing sequential data and capable of capturing temporal dependencies within the data. Through recurrent connections, RNNs retain information from the previous step at each step in the sequence, making them suitable for analyzing time series changes in imaging data. Long Short-Term Memory (LSTM): A specialized RNN structure that better captures long-term dependencies and avoids the vanishing gradient problem found in traditional RNNs. By introducing forget gates, input gates, and output gates, LSTMs can selectively remember or forget information, making them suitable for complex time series prediction tasks.

[0068] 3. Postoperative Recovery Model Generation: Model Training and Prediction: Utilizing pre-processed multimodal imaging data, an RNN or LSTM model is trained to learn the dynamic patterns of change during a patient's postoperative recovery. Recovery Model Output: The generated postoperative recovery model reflects the changing trends of a patient's brain structure and function at different time points after surgery, providing a basis for personalized efficacy evaluation.

[0069] Furthermore, the personalized efficacy assessment module aims to combine the postoperative recovery model generated by the dynamic image fusion and analysis module with the patient's clinical characteristics data to conduct a comprehensive analysis and generate a personalized postoperative recovery report. By applying a comprehensive assessment algorithm, this module can provide accurate efficacy assessments based on the patient's specific situation and predict the risk of postoperative seizures. Its specific implementation steps include:

[0070] 1. Data Receipt and Integration: Postoperative Recovery Model Receipt: The Personalized Efficacy Evaluation Module receives the postoperative recovery model generated by the Dynamic Image Fusion and Analysis Module. This model reflects the changes in the patient's brain structure and function at different time points after surgery. Clinical Characteristic Data Acquisition: The patient's clinical characteristic data, including but not limited to age, gender, medical history, and preoperative seizure frequency, are collected. This data is obtained through the electronic medical record system or manual input, and the accuracy and completeness of the data are ensured.

[0071] 2. Application of comprehensive evaluation algorithm: The following comprehensive evaluation formula is used to generate a personalized postoperative recovery report: in 、 、 is the dynamic adjustment coefficient; 、 is the error correction term; It is a comprehensive evaluation value, representing the overall efficacy evaluation result; To provide preliminary measurement and calibration results, reflecting basic evaluation indicators; To accurately calibrate the results and reflect the further optimized evaluation indicators; It is to perform logarithmic transformation on the precise calibration results, smooth the evaluation values, and reduce the impact of extreme values.

[0072] 3. Seizure risk prediction: Support vector machine (SVM) classifier application: based on comprehensive evaluation value Predict the patient's postoperative seizure risk using a pre-trained SVM classifier. The SVM classifier analyzes the relationship between the comprehensive assessment value and historical data to determine the patient's seizure risk level (e.g., high risk, medium risk, low risk). Prediction result integration: The SVM classifier's prediction results are integrated into a personalized postoperative recovery report, which is provided to the medical team as a reference for decision-making.

[0073] 4. Report Generation and Delivery: Report Generation: Based on the above analysis results, a detailed personalized postoperative recovery report is generated, including comprehensive assessment values, seizure risk prediction results, and corresponding recommended measures. Report Delivery: The generated personalized postoperative recovery report is delivered to the Abnormal Detection and Intelligent Early Warning Module as a basis for further monitoring and optimization.

[0074] Furthermore, the Anomaly Detection and Intelligent Early Warning Module is responsible for real-time monitoring of anomalies in the postoperative recovery reports generated by the Personalized Efficacy Evaluation Module, implementing error compensation, and issuing early warning notifications. This module ensures the accuracy and reliability of evaluation results through error propagation and compensation mechanisms, while also promptly notifying the medical team of potential problems through early warning mechanisms. Its specific implementation steps include:

[0075] 1. Abnormality Monitoring: Real-time Data Reception: Receives postoperative recovery reports generated by the personalized efficacy evaluation module and monitors various evaluation indicators within the reports. Abnormality Detection: Receives evaluation indicators and analyzes them in real time to identify those that fall outside the normal range or exhibit abnormal changes. Using pre-set thresholds and rules, the system determines which evaluation results require attention.

[0076] 2. Error propagation and compensation: Application of error propagation formula:

[0077] ,in is the error propagation amount, is the error correction coefficient, which reflects the influence of different error sources; Parameters that control the range of error propagation and determine the rate at which the error decays; is the reference value, usually historical data or standard value. Error compensation implementation: calculated by the above formula , adjust the preliminary assessment results, compensate for possible errors, and ensure the accuracy of the assessment results.

[0078] 3. Warning notification: Warning trigger condition setting: According to the error propagation and abnormal detection results, set the warning trigger conditions. For example, when When a threshold is exceeded or an assessment indicator deviates significantly from a reference value, an alert is triggered. Alert Notification Generation: Once an alert condition is triggered, the system automatically generates an alert notification, including a description of the abnormality, possible cause analysis, and recommended countermeasures. Notification Delivery: The generated alert notification is sent to the Report Generation and Visualization Module for further display to the medical team, and immediate notification is provided via SMS, email, and other means.

[0079] Furthermore, the data storage and management module is responsible for the storage and management of all imaging data, analysis results, and patient information in the system, ensuring data security and privacy protection. This module uses efficient data management strategies and security mechanisms to ensure data integrity and accessibility, supporting stable system operation and efficient data utilization. Its specific implementation steps include:

[0080] 1. Data Storage: Raw Data Storage: This stores the original versions of multimodal imaging data to ensure data integrity and traceability. This includes pre- and post-operative T1-weighted images, DTI, and fMRI data. Pre-processed Data Storage: This stores image data pre-processed by the Multimodal Imaging Data Acquisition and Pre-processing Module to facilitate subsequent analysis and use. Analysis Result Storage: This stores the post-operative recovery model generated by the Dynamic Image Fusion and Analysis Module, the evaluation report generated by the Personalized Therapeutic Efficacy Evaluation Module, and related parameters.

[0081] 2. Data Management: Classification and Indexing: Classify and index stored data based on patient ID, image type, time point, and other information to ensure rapid retrieval and efficient data management. Backup and Recovery: Regularly back up all data and establish a data recovery mechanism to prevent data loss or damage.

[0082] 3. Data Security and Privacy Protection: Encrypted Storage: All stored data is encrypted using high-strength encryption algorithms (such as AES-256) to ensure data security during storage and transmission. Access Control: Strict access control mechanisms are implemented to ensure only authorized personnel can access sensitive data. User authentication and permission management prevent unauthorized access and data leaks. Logging: All data access and operation logs are recorded for audit and tracking purposes, ensuring transparency and traceability of data operations.

[0083] Furthermore, the Report Generation and Visualization Module integrates the data generated by the Personalized Efficacy Evaluation Module and the Anomaly Detection and Intelligent Early Warning Module, generating reports that are easy to understand and analyze using visualization technology. These reports include postoperative recovery trend charts and personalized indicator charts, aiming to provide the medical team with intuitive data presentations to assist in clinical decision-making. The specific implementation steps include:

[0084] 1. Data Reception and Integration: Receive test results from the Abnormality Detection and Intelligent Early Warning Module, including abnormalities detected during the evaluation process and their associated data. Receive evaluation data from the Personalized Efficacy Evaluation Module, including the personalized postoperative recovery report generated by the Personalized Efficacy Evaluation Module, including comprehensive evaluation values and seizure risk prediction results. Data Integration: Data from both modules is integrated to ensure that all relevant information is fully reflected in the report generation process.

[0085] 2. Data Visualization: Recovery Trend Chart Generation: Use data visualization tools (such as Matplotlib or D3.js) to generate charts showing postoperative recovery trends. These charts illustrate changes in the patient's brain structure and function at different time points, helping the medical team intuitively understand the recovery process. Personalized Indicator Chart Generation: Based on the results of personalized efficacy assessments, charts are generated to display individualized indicators (such as comprehensive assessment values and seizure risk levels). These indicator charts highlight the patient's specific recovery status and potential risks, supporting precision medicine.

[0086] 3. Report Output and Delivery: Multi-Format Support: The system supports report generation in a variety of formats, including PDF, HTML, and an interactive visualization interface. PDF format is suitable for printing and archiving, HTML format is convenient for online viewing, and the interactive interface supports dynamic analysis and exploration of data by the medical team. Report Delivery: Generated visualization reports are sent to the medical team's workstations or shared through a secure network platform, ensuring that medical staff can access and review assessment results in a timely manner, assisting in clinical decision-making.

[0087] Furthermore, the multimodal image data acquisition and preprocessing module not only performs basic data acquisition and preprocessing, but also performs preliminary measurement calibration and precision calibration to ensure high accuracy and consistency of image data. These algorithmic steps further enhance the reliability of image data in subsequent analysis through a refined calibration process. The specific implementation steps include:

[0088] 1. Preliminary measurement calibration: Formula application: ,in, is the normalized pixel value; is the weight of image preprocessing; is the adjustment coefficient of environmental parameters; It is an environmental parameter; by giving different pixel values different weights , the system can highlight important areas and reduce the impact of irrelevant areas; by adjusting the coefficient and environmental parameters , the system can compensate for the impact of external environmental changes on image data, ensuring the stability and consistency of evaluation results.

[0089] 2. Secondary precision calibration: Formula application: ,in is the dynamic weight of the external parameter; is an external parameter; is the adjustment coefficient, which is used to control the impact of input on output; The function is used for nonlinear correction and smoothing of excessive input values;

[0090] pass Dynamically adjust the weights of different external parameters, and the system can flexibly respond to various external interferences according to actual conditions; use The function implements nonlinear correction to ensure that the calibration process can effectively compensate for errors without causing new errors due to over-adjustment; Accuracy improvement: Secondary precision calibration further improves the accuracy and reliability of the evaluation results by comprehensively considering multiple external factors.

[0091] Furthermore, the dynamic image fusion and analysis module employs recurrent neural networks (RNNs) and long short-term memory (LSTM) structures to predict time series and capture long-term dependencies in multimodal imaging data. Through these advanced deep learning algorithms, the module generates a more accurate and comprehensive postoperative recovery model. The specific implementation steps include:

[0092] 1. Time Series Forecasting: Recurrent Neural Network (RNN) Model Application: Model Structure: RNN is a neural network structure suitable for processing sequential data, capable of capturing temporal dependencies in the data. Working Principle: RNN maintains a hidden state through recurrent connections. At each time step, the hidden state of the previous step is combined with the current input to generate a new hidden state and output.

[0093] Implementation Steps: Data Input: Preprocessed multimodal imaging data is fed into the RNN model as a time series. Hidden State Update: At each time step, the hidden state is updated based on the current input and the previous hidden state. Output Generation: Postoperative recovery indicators are generated for the corresponding time step, reflecting the patient's recovery status at that point in time. The RNN's recurrent structure enables it to memorize information from previous time steps, making it suitable for analyzing continuous changes during a patient's postoperative recovery. This structure enables the model to identify and exploit long-term dependencies in the imaging data, improving prediction accuracy.

[0094] 2. Capturing Long-Term Dependencies: Long Short-Term Memory (LSTM) Architecture Application: Model Structure: LSTM is a special RNN architecture designed to address the vanishing gradient problem of traditional RNNs when processing long-term dependencies. Working Principle: By introducing a forget gate, input gate, and output gate, LSTM can selectively retain or forget information, effectively capturing long-term dependencies.

[0095] Implementation Steps: Data Input: Preprocessed multimodal imaging data is fed into the LSTM model. Gating Mechanism Application: The forget gate, input gate, and output gate control the flow of information, selectively retaining important information and ignoring irrelevant information. Recovery Model Generation: A more accurate and stable postoperative recovery model is generated, reflecting the dynamic changes in patients during long-term recovery. The LSTM, through its gating mechanism, can effectively process and memorize data changes over long time spans, making it suitable for analyzing long-term recovery trends after surgery. This structure enables the model to maintain efficient and accurate prediction capabilities even when dealing with complex and long-term data series.

[0096] 3. Restore Model Output: Model Training and Validation: Data Segmentation: Divide the collected multimodal imaging data into a training set and a validation set to ensure the model's generalization capabilities. Model Training: Use the training set to train the RNN and LSTM models, optimizing model parameters by minimizing prediction error. Model Validation: Evaluate model performance on the validation set to ensure its prediction accuracy and stability on unseen data.

[0097] Recovery Model Generation: Model Application: Apply the trained RNN or LSTM model to the multimodal imaging data of a new patient to generate a personalized postoperative recovery model. Result Output: The recovery model includes structural and functional evaluation indicators at different time points, providing accurate data support for the personalized efficacy evaluation module.

[0098] Furthermore, the personalized efficacy assessment module further includes acquiring patient clinical characteristics, generating a comprehensive assessment score, predicting postoperative seizure risk, performing staged assessments, and implementing a feedback loop mechanism. These steps, by comprehensively considering the patient's clinical characteristics and imaging recovery model, enable a comprehensive and personalized assessment of the patient's postoperative recovery. Specific implementation steps include:

[0099] 1. Obtaining Patient Clinical Data: Data Collection: Collect basic patient information and clinical data, including but not limited to age, gender, medical history, and preoperative seizure frequency. This data is obtained through electronic medical record systems, questionnaires, or manual entry. Data Integration: Integrate the collected clinical data with the postoperative recovery model generated by the dynamic image fusion and analysis module to provide a comprehensive data foundation for comprehensive evaluation.

[0100] 2. Generate comprehensive evaluation value: Application of weighted average method: in 、 、 is the dynamic adjustment coefficient; 、 is the error correction term; It is a comprehensive evaluation value, representing the overall efficacy evaluation result; To provide preliminary measurement and calibration results, reflecting basic evaluation indicators; To accurately calibrate the results and reflect the further optimized evaluation indicators; It is to perform logarithmic transformation on the precise calibration results, smooth the evaluation values, and reduce the impact of extreme values.

[0101] The comprehensive evaluation value is obtained by weighted combination of preliminary evaluation results and accurate assessment results , comprehensively reflect the patient's postoperative recovery status; 、 、 The introduction of , enables the assessment process to be flexibly adjusted according to the specific circumstances and needs of different patients, ensuring the personalization and accuracy of the assessment results; 、 Compensate for errors in the evaluation results, reduce the impact of systematic errors on the final evaluation value, and improve the credibility of the evaluation results; The application of can effectively smooth the evaluation values, avoid the excessive impact of extreme values on the comprehensive evaluation results, and improve the stability of the evaluation.

[0102] 3. Postoperative seizure risk prediction: Support vector machine (SVM) classifier application: Model training: Use historical patient data (including comprehensive assessment values and postoperative seizure conditions) to train the SVM classifier and establish a correlation model between comprehensive assessment values and seizure risk. Model application: Based on the current patient's comprehensive assessment value , using the trained SVM classifier to predict the patient's postoperative seizure risk level (e.g., high risk, medium risk, low risk). Prediction result integration: The predicted seizure risk level is integrated into the personalized postoperative recovery report, providing early warning and intervention recommendations to the medical team.

[0103] The SVM classifier has excellent classification performance and can effectively process high-dimensional data, making it suitable for complex seizure risk prediction tasks. Through training and validation, the SVM model can accurately distinguish different risk levels, improving the accuracy and reliability of predictions.

[0104] 4. Staged evaluation: Generate performance indicators at each stage: ,in is the performance indicator of stage i, which measures the difference between the evaluation result of this stage and the reference value; is the number of samples for each stage; is the evaluation value of the jth sample in the i-th stage; is the reference value of the jth sample; calculation of comprehensive evaluation index: ,in The weight of each stage reflects the importance of different stages in the comprehensive evaluation; It is a comprehensive evaluation indicator, representing the overall efficacy evaluation result.

[0105] By calculating the performance indicators of each stage, the system can carefully evaluate the effects of different recovery stages and identify potential problems; The setting ensures that the evaluation results of different stages occupy corresponding proportions in the comprehensive evaluation according to their importance, thereby improving the comprehensiveness and accuracy of the evaluation results.

[0106] 5. Feedback loop mechanism: Receive accurate calibration results: Receive accurate calibration results from the secondary precision calibration step ; Adjust the input data to form a feedback loop: ,in is the adjusted input data; is the original input data; is the feedback gain coefficient, It is the derivative of the precise calibration result with respect to the input data, indicating the degree of influence of the input data change on the evaluation result.

[0107] Input data re-preprocessing: The adjusted input data The data is re-input into the multimodal image data acquisition and preprocessing module to dynamically optimize the evaluation results. Through the feedback loop mechanism, the system can dynamically adjust the input data according to the precise calibration results, optimize the evaluation process in real time, and improve the system's adaptability and the accuracy of the evaluation results. The settings ensure the appropriateness of input data adjustments, avoiding over-adjustments that could lead to system instability or distorted evaluation results. All parameters and steps in the feedback adjustment process are clearly defined, ensuring transparency and traceability of the adjustment process.

[0108] In summary, the present invention realizes a comprehensive, accurate and dynamic evaluation of the efficacy of temporal lobe epilepsy resection surgery through the collaborative work of a multimodal image data acquisition and preprocessing module, a dynamic image fusion and analysis module, a personalized efficacy evaluation module, anomaly detection and intelligent early warning module, a data storage and management module, and a report generation and visualization module. The system integrates a variety of brain imaging data, applies advanced deep learning algorithms, and combines personalized clinical characteristics to provide highly accurate and reliable evaluation results. At the same time, through error compensation and feedback loop mechanisms, the system has the ability to optimize and dynamically adjust in real time to ensure the stability of the evaluation process and the credibility of the results. The data storage and management module ensures the security and privacy protection of the data, and the report generation and visualization module provides the medical team with an intuitive data display to assist them in making clinical decisions. The overall system not only improves the accuracy and reliability of efficacy evaluation, but also significantly improves the quality of life of patients with temporal lobe epilepsy after surgery, and has broad clinical application prospects.

[0109] To further illustrate the specific implementation of the present invention, the following examples illustrate the implementation process and advantages of the present invention in different application scenarios. These examples will help to more intuitively understand the working principles and practical application effects of the system and verify the practicality and superiority of the present invention in clinical settings.

[0110] Example 1: Postoperative evaluation of temporal lobe epilepsy in a standard clinical setting

[0111] Temporal lobe epilepsy is one of the most common types of drug-resistant epilepsy, and surgical resection is an effective treatment. However, postoperative efficacy assessment typically relies on clinical symptom observation and single-modality imaging data analysis, which fails to fully reflect changes in brain structure and function, resulting in insufficient accuracy and reliability of the assessment.

[0112] This embodiment aims to integrate multimodal brain imaging data through the temporal lobe epilepsy resection surgery efficacy evaluation system of the present invention, utilize advanced algorithm models, provide a comprehensive and accurate postoperative recovery evaluation, improve the accuracy and reliability of the evaluation, and assist the medical team in making clinical decisions.

[0113] Implementation Method

[0114] Patient data acquisition: T1-weighted images, DTI, and fMRI data were collected before and after surgery.

[0115] Data preprocessing: Use the multimodal image data acquisition and preprocessing module to perform denoising, standardization, head motion correction and spatial smoothing on the collected data.

[0116] Image fusion and analysis: The dynamic image fusion and analysis module receives the preprocessed data and applies the RNN and LSTM models to generate a postoperative recovery model.

[0117] Personalized assessment: The personalized efficacy assessment module combines the recovery model and the patient's clinical characteristic data to generate a comprehensive assessment value and postoperative relapse risk prediction.

[0118] Anomaly monitoring and early warning: The anomaly detection and intelligent early warning module monitors the evaluation results in real time, compensates for errors, and issues early warning notifications when anomalies are detected.

[0119] Report Generation and Visualization: The report generation and visualization module integrates assessment and test data to generate multi-format visual reports for medical teams to review and make decisions.

[0120] Multimodal data integration: By integrating T1-weighted images, DTI and fMRI data, it comprehensively reflects the changes in brain structure and function.

[0121] Advanced deep learning algorithms: Using RNN and LSTM models to accurately capture the dynamic changes in postoperative recovery.

[0122] Personalized assessment and risk prediction: Combined with clinical characteristic data, it provides personalized efficacy evaluation and seizure risk prediction.

[0123] Real-time anomaly monitoring and early warning: Ensure the accuracy and reliability of assessment results through error compensation and early warning mechanisms.

[0124] In a standard clinical setting, this system is used to conduct postoperative recovery assessments for patients undergoing temporal lobe epilepsy resection surgery, assisting physicians in developing personalized postoperative management and intervention plans.

[0125] By implementing this system, we can more comprehensively and accurately assess the patient's postoperative recovery, detect abnormalities in a timely manner, improve the accuracy and reliability of efficacy evaluation, assist the medical team in making scientific decisions, and significantly improve the patient's quality of life after surgery.

[0126] Comparative experiment: To verify the superiority of the system of the present invention, a comparative experiment with the existing single-modality image evaluation method was conducted.

[0127]

[0128] Comparative results analysis:

[0129] Assessment accuracy and reliability: The system of the present invention significantly improves the assessment accuracy and reliability through multimodal data integration and advanced algorithms. Data integration capability: Compared with single-modal methods, this system can integrate a variety of brain imaging data to provide a more comprehensive assessment basis. Time dependency capture capability: Using RNN and LSTM models, the system can effectively capture and utilize time dependencies in the data to improve prediction capabilities. Risk prediction capability: The accuracy of seizure risk prediction of this system is higher than that of existing methods, providing more reliable clinical decision support. Timeliness of anomaly detection: Real-time monitoring and early warning mechanisms enable the system to respond quickly to abnormal situations and reduce potential risks. Data security: High-strength encryption and strict access control ensure the security and privacy protection of patient data. User-friendliness: Multi-format visual reports enable the medical team to understand the assessment results more intuitively, improving the practicality and ease of use of the system.

[0130] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A temporal lobe epilepsy resection surgery efficacy evaluation system based on brain image fusion features, characterized by: The system includes the following modules: a) Multimodal imaging data acquisition and preprocessing module, responsible for acquiring and preprocessing preoperative and postoperative T1-weighted images, diffusion tensor imaging, and functional magnetic resonance imaging data. Preprocessing steps include denoising, normalization, motion correction, and spatial smoothing. b) a dynamic image fusion and analysis module, which receives the image data preprocessed by the multimodal image data acquisition and preprocessing module, applies a multi-time series analysis method to generate a postoperative recovery model, and transmits the postoperative recovery model to the personalized efficacy evaluation module; c) A personalized efficacy evaluation module, which combines the postoperative recovery model generated by the dynamic image fusion and analysis module with the patient characteristic data to analyze and generate a personalized postoperative recovery report, which is then passed to the anomaly detection and intelligent early warning module; d) Anomaly Detection and Intelligent Early Warning Module, which monitors abnormalities in the postoperative recovery report generated by the personalized efficacy evaluation module in real time, implements error compensation, issues early warning notifications, and transmits the detection results to the report generation and visualization module; e) Data storage and management module, responsible for storing and managing multimodal imaging data, analysis results and patient information; f) Report generation and visualization module, which receives data from the personalized efficacy evaluation module and the anomaly detection and intelligent early warning module, and generates a visualization report showing postoperative recovery trends and personalized indicators; The personalized efficacy evaluation module further includes the following algorithm steps: a) Obtain the patient's clinical characteristics, including but not limited to age, gender, medical history, and preoperative seizure frequency; b) combining the patient's clinical characteristics data with the postoperative image restoration model to generate a comprehensive evaluation value using a weighted average method; c) Based on the comprehensive evaluation value, the support vector machine classifier is used to predict the patient's postoperative seizure risk; d) Implement a phased evaluation, specifically including: i. Generate performance indicators for each phase, calculated as follows: ,in is the performance index of stage i; is the number of samples for each stage; is the evaluation value of the jth sample in the i-th stage; is the reference value of the jth sample; ii. Comprehensive evaluation index, calculated using the following formula: ,in The weight of each stage; e) Implement feedback loops, including: i. Receive accurate calibration results ; ii. Adjust the input data according to the following formula , forming a feedback loop: ,in is the adjusted input data; is the original input data; is the feedback gain coefficient; iii. Adjust the input data Re-input into the multimodal image data acquisition and preprocessing module to dynamically optimize the evaluation results.

2. The temporal lobe epilepsy resection surgery efficacy evaluation system based on brain image fusion features according to claim 1, characterized in that: The multimodal image data acquisition and preprocessing module further includes the following steps: a) Convert preoperative and postoperative T1-weighted images, diffusion tensor imaging, and functional magnetic resonance imaging raw image data into a standard format; b) performing denoising on the converted image data by using wavelet transform or Gaussian filtering; c) Standardize the denoised image data and calculate the mean of the image data and standard deviation , and apply the normalization formula: ,in is the mean of the image data, is the standard deviation of the image data, is the normalized pixel value; d) Perform head motion correction, using an edge detection-based algorithm to identify and correct head motion; e) Perform spatial smoothing on the corrected image data and perform convolution operation using a three-dimensional Gaussian kernel.

3. The temporal lobe epilepsy resection surgery efficacy evaluation system based on brain image fusion features according to claim 1, characterized in that: The dynamic image fusion and analysis module further includes the following steps: a) receiving standardized image data from a multimodal image data preprocessing module; b) Applying multiple time series analysis methods, using time series analysis or recurrent neural network models to generate postoperative recovery models; c) Output the generated postoperative recovery model to the personalized efficacy evaluation module.

4. The temporal lobe epilepsy resection surgery efficacy evaluation system based on brain image fusion features according to claim 1, characterized in that: The personalized efficacy evaluation module further includes the following steps: a) receiving the postoperative recovery model output by the dynamic image fusion and analysis module; b) Combined with the patient's clinical characteristics data, a comprehensive assessment algorithm is applied to generate a personalized postoperative recovery report. The specific calculation formula is as follows: in 、 、 is the dynamic adjustment coefficient; 、 is the error correction term; It is a comprehensive evaluation value, representing the overall efficacy evaluation result; To provide preliminary measurement and calibration results, reflecting basic evaluation indicators; To accurately calibrate the results; c) outputting the personalized postoperative recovery report to the abnormality detection and intelligent early warning module.

5. The temporal lobe epilepsy resection surgery efficacy evaluation system based on brain image fusion features according to claim 1, characterized in that: The anomaly detection and intelligent early warning module further includes the following steps: a) Real-time monitoring of abnormalities in the postoperative recovery report generated by the personalized efficacy evaluation module; b) Apply the error propagation and compensation formula to perform anomaly detection. The specific formula is as follows: ,in is the error propagation amount, is the error correction coefficient; Control the scope of error propagation; is the reference value, is the historical data or standard value, and error compensation is implemented; c) Issue early warning notifications based on the test results and transmit the test results to the report generation and visualization module.

6. The temporal lobe epilepsy resection surgery efficacy evaluation system based on brain image fusion features according to claim 1, characterized in that: The data storage and management module further comprises the following steps: a) Store raw and pre-processed versions of multimodal imaging data; b) storing the postoperative recovery model generated by the dynamic image fusion and analysis module; c) storing the evaluation report and related parameters generated by the personalized efficacy evaluation module; d) Ensure data security and privacy protection by adopting encryption storage and access control mechanisms.

7. The temporal lobe epilepsy resection surgery efficacy evaluation system based on brain image fusion features according to claim 1, characterized in that: The report generation and visualization module further comprises the following steps: a) Receive detection results from the anomaly detection and intelligent early warning module; b) Using data visualization technology to generate postoperative recovery trend charts and personalized indicator charts based on the evaluation results and test data; c) Supports report output in multiple formats, including PDF, HTML and interactive visual interface.

8. The temporal lobe epilepsy resection surgery efficacy evaluation system based on brain image fusion features according to claim 1, characterized in that: The multimodal image data acquisition and preprocessing module further includes the following algorithm steps: a) performing preliminary measurement calibration, the specific calculation formula is as follows: ,in, is the normalized pixel value; is the weight of image preprocessing; is the adjustment coefficient of environmental parameters; is the environmental parameter; Calibration results for preliminary measurements; b) Perform secondary precision calibration. The specific calculation formula is as follows: ,in is the dynamic weight of the external parameter; is an external parameter; is the adjustment coefficient, which is used to control the impact of input on output; The function is used for nonlinear correction and smoothing of excessive input values; For accurate calibration results.

9. The temporal lobe epilepsy resection surgery efficacy evaluation system based on brain image fusion features according to claim 1, characterized in that: The dynamic image fusion and analysis module further includes the following algorithm steps: a) Use recurrent neural network models to perform time series prediction on multimodal imaging data; b) Applying long short-term memory network structures to capture long-term dependencies in image data; c) Generate a postoperative recovery model as input for personalized efficacy assessment.

Images