Efficacy prediction model training method, efficacy prediction method, and electronic device

Abstract

The present application relates to the field of efficacy prediction, and in particular to a kind of efficacy prediction model training method and efficacy prediction method and electronic equipment. Including: obtaining the clinical information corresponding to target patient, electroencephalogram data, electrocardiogram data and the sample type corresponding to target patient;Based on the clinical information corresponding to target patient, electroencephalogram data, electrocardiogram data, generate the initial feature corresponding to target patient;Based on the sample type corresponding to target patient, the initial feature is optimized and handled, and the target feature corresponding to target patient is generated;Based on the corresponding relationship between the target feature corresponding to target patient and sample type, generate the training data set corresponding to target patient;Initial efficacy prediction network is trained using training data set, and target efficacy prediction model is generated.The above method guarantees the accuracy of the target efficacy prediction model obtained by training, so as to guarantee the accuracy of predicting the treatment efficacy of patients using the target efficacy prediction model.

Description

Technical Field

[0001] This invention relates to the field of efficacy prediction, specifically to a method for training an efficacy prediction model, as well as an efficacy prediction method and electronic device. Background Technology

[0002] Epilepsy is a common neurological disorder affecting over 40 million people worldwide. While medication can effectively control seizures in most patients, approximately 30% experience unresponsive seizures despite treatment with at least two medications, either alone or in combination. This type of epilepsy is termed drug-resistant epilepsy. Drug-resistant epilepsy can cause severe and irreversible damage to the nervous system, particularly the brain development of children, significantly impacting the quality of life for both the patient and their family.

[0003] Vagus nerve stimulation (VNS) is an adjunctive treatment for refractory epilepsy. By stimulating the vagus nerve on the left side of the neck, it can alleviate seizures to some extent. However, its effectiveness varies greatly among patients; only 50%-60% of patients experience a reduction in seizure frequency of more than 50% with VNS, and only 10% achieve complete seizure remission. Given the aforementioned therapeutic characteristics of VNS, and considering the potential risks of implantation surgery, the long treatment course, and the high cost, preoperative evaluation to determine patient suitability for VNS therapy would benefit patients economically and in terms of quality of life, and allow for better allocation of medical resources.

[0004] Previous studies have examined the correlation between preoperative information and postoperative efficacy, including patient demographics, clinical history, clinical disease information (etiology, seizure type, lesion location, etc.), and physiological signals from preoperative electrocardiograms (ECG), electroencephalograms (EEG), and magnetic resonance imaging (MRI). Among these, scalp EEG is of paramount value. Scalp EEG is the gold standard for epilepsy diagnosis and assessment. Clinically, experts visually identify brain discharges and seizure patterns displayed on EEG, combining this with experience to arrive at diagnostic and assessment conclusions. In quantitative EEG studies, time-frequency analysis, network function analysis, and entropy analysis can be used to interpret EEG signals at the biomarker level, thereby characterizing the underlying physiological characteristics. Furthermore, dynamic ECG signals, obtained from freely moving subjects, can also reflect their physiological condition to some extent. Since the vagus nerve is involved in the regulation of cardiac autonomic nerve function, analyzing ECG signals can indirectly reflect the influence of the vagus nerve. This effect can be interpreted at the level of biomarkers by calculating heart rate variability (HRV).

[0005] However, there is currently no effective method for predicting the preoperative efficacy of VNS (vaginal nerve resuscitation) in clinical practice. Correlation analysis and prediction of long-term VNS efficacy suffer from inconsistencies in conclusions, poor predictive accuracy, and weak generalization. Furthermore, most of the aforementioned studies focus on single-type biomarkers, lacking comprehensive analysis of multi-angle and multi-modal indicators, making it difficult to improve predictive accuracy. Summary of the Invention

[0006] In view of this, embodiments of the present invention provide a method for training a therapeutic efficacy prediction model, as well as a therapeutic efficacy prediction method and electronic device, aiming to solve the problem of low accuracy in therapeutic efficacy prediction in the prior art.

[0007] According to a first aspect, embodiments of the present invention provide a method for training a therapeutic efficacy prediction model, comprising:

[0008] Acquire clinical information, EEG data, ECG data, and sample type corresponding to the target patient. The sample type is determined based on the long-term efficacy of the target patient and includes positive and negative samples.

[0009] Based on the clinical information, EEG data, and ECG data of the target patient, generate the initial features corresponding to the target patient;

[0010] Based on the sample type corresponding to the target patient, the initial features are optimized to generate the target features corresponding to the target patient.

[0011] Based on the correspondence between the target features and sample types corresponding to the target patients, a training dataset corresponding to the target patients is generated;

[0012] The initial efficacy prediction network is trained using the training dataset to generate the target efficacy prediction model.

[0013] The efficacy prediction model training method provided in this invention acquires clinical information, EEG data, and ECG data corresponding to the target patient, as well as the sample type corresponding to the target patient. Based on the clinical information, EEG data, and ECG data corresponding to the target patient, initial features corresponding to the target patient are generated, ensuring the accuracy and diversity of the generated initial features. Then, based on the sample type corresponding to the target patient, the initial features are optimized to generate target features corresponding to the target patient, ensuring the accuracy of the generated target features. Based on the correspondence between the target features and the sample type corresponding to the target patient, a training dataset corresponding to the target patient is generated, ensuring the accuracy of the generated training dataset and the accuracy of the data volume, avoiding the influence of too many invalid features on the training dataset and preventing a large training dataset, which would result in a long training time for training the initial efficacy prediction network using the training dataset. The initial efficacy prediction network is trained using the training dataset to generate the target efficacy prediction model, ensuring the accuracy of the generated target efficacy prediction model. The above method ensures the accuracy of the trained target efficacy prediction model, thereby ensuring the accuracy of predicting the treatment efficacy of patients using the target efficacy prediction model.

[0014] In conjunction with the first aspect, in the first embodiment of the first aspect, based on the clinical information, EEG data, and ECG data corresponding to the target patient, initial features corresponding to the target patient are generated, including:

[0015] Based on clinical information, an initial clinical feature is generated; the initial clinical feature includes multiple initial sub-clinical features.

[0016] The EEG data was analyzed to calculate the initial functional connectivity features, initial network topology features, and initial entropy features corresponding to the target patient. The initial functional connectivity features include multiple initial sub-functional connectivity features; the initial network topology features include multiple initial sub-network topology features; and the initial entropy features include multiple initial sub-entropy features.

[0017] The electrocardiogram data is analyzed to calculate the initial heart rate variability characteristics corresponding to the target patient; the initial heart rate variability characteristics include multiple initial sub-heart rate variability characteristics;

[0018] Initial features are constituted based on initial clinical features, initial functional connectivity features, initial network topology features, initial entropy features, and initial heart rate variability features.

[0019] The efficacy prediction model training method provided in this invention generates initial clinical features based on clinical information, ensuring the accuracy of the generated initial clinical features. It analyzes electroencephalogram (EEG) data to calculate the initial functional connectivity features, initial network topology features, and initial entropy features corresponding to the target patient, ensuring the accuracy of these calculated features. It also analyzes electrocardiogram (ECG) data to calculate the initial heart rate variability features corresponding to the target patient, ensuring the accuracy of the calculated features. Based on the initial clinical features, initial functional connectivity features, initial network topology features, initial entropy features, and initial heart rate variability features, initial features are constructed, ensuring the accuracy and diversity of the generated initial features, and guaranteeing that the generated initial features include analysis results from multiple perspectives and modalities.

[0020] In conjunction with the first embodiment of the first aspect, in the second embodiment of the first aspect, based on the sample type corresponding to the target patient, the initial features are optimized to generate target features corresponding to the target patient, including:

[0021] Using a pre-defined testing method, significant differences between the initial features corresponding to positive and negative samples in the target patients were analyzed to identify candidate features from the initial features. The candidate features included candidate clinical features, candidate functional connectivity features, candidate network topology features, candidate entropy features, and candidate heart rate variability features.

[0022] A recursive feature elimination method based on a pre-defined binary classification algorithm is used to select candidate features and generate target features corresponding to the target patient.

[0023] The efficacy prediction model training method provided in this invention utilizes a preset testing method to perform inter-group significance difference analysis on the initial features corresponding to positive samples and negative samples in the target patients. Candidate features are determined from the initial features, ensuring the accuracy of the candidate features and removing redundant features to avoid their influence on the candidate features. Then, a recursive feature elimination method based on a preset binary classification algorithm is used to select features from the candidate features, generating target features corresponding to the target patients. This ensures the accuracy of the generated target features and removes redundant features from the candidate features, avoiding their influence on the target features. This ensures the accuracy of the training data set generated based on the target features, thereby guaranteeing the accuracy of the target efficacy prediction model trained using the training dataset.

[0024] In conjunction with the second implementation of the first aspect, in the third implementation of the first aspect, a recursive feature elimination method based on a preset binary classification algorithm is used to select candidate features and generate target features corresponding to the target patient, including:

[0025] Based on the candidate features, generate a first candidate feature set that includes all candidate features;

[0026] The first candidate feature set is input into a preset binary classification algorithm to train the preset binary classification algorithm, thereby obtaining the first candidate binary classification model and determining the weight coefficients corresponding to each candidate feature in the first candidate feature set.

[0027] Based on the weight coefficients of each candidate feature in the first candidate feature set, the candidate feature with the smallest weight coefficient is removed from the candidate clinical features, candidate functional connectivity features, candidate network topology features, candidate entropy features, and candidate heart rate variability features respectively, to obtain the second candidate feature set.

[0028] This process is repeated until the number of candidate features in the set reaches the preset requirement.

[0029] Obtain the model accuracy corresponding to each candidate binary classification model, and determine the candidate features in the candidate feature set corresponding to the candidate binary classification model with the highest model accuracy as backup features;

[0030] Based on the backup features, determine the target features.

[0031] The efficacy prediction model training method provided in this invention generates a first candidate feature set including all candidate features based on candidate features. The first candidate feature set is input into a preset binary classification algorithm to train the algorithm, resulting in a first candidate binary classification model. The weight coefficients corresponding to each candidate feature in the first candidate feature set are determined, ensuring the accuracy of these weight coefficients. Based on the weight coefficients of each candidate feature in the first candidate feature set, the candidate feature with the smallest weight coefficient is removed from candidate clinical features, candidate functional connectivity features, candidate network topology features, candidate entropy features, and candidate heart rate variability features, respectively, resulting in a second candidate feature set. This ensures the accuracy and importance of each candidate feature in the second candidate feature set. This process is repeated until the number of candidate features in the second candidate feature set reaches a preset requirement. The model accuracy corresponding to each candidate binary classification model is obtained, and the candidate features in the candidate feature set corresponding to the candidate binary classification model with the highest model accuracy are selected as backup features, ensuring the accuracy of the selected backup features. Finally, based on the backup features, target features are determined, ensuring both the accuracy and diversity of the selected target features.

[0032] In conjunction with the third embodiment of the first aspect, in the fourth embodiment of the first aspect, the target feature is determined based on the alternative features, including:

[0033] By utilizing the relationship between the backup features corresponding to the target patient and the sample type corresponding to the target patient, the preset binary classification algorithm is trained again, and the weight coefficients of each backup feature are determined.

[0034] Remove standby features whose weight coefficients are less than a preset weight threshold to obtain the target features.

[0035] The efficacy prediction model training method provided in this embodiment of the invention utilizes the relationship between the backup features corresponding to the target patient and the sample type corresponding to the target patient to retrain the preset binary classification algorithm and determine the weight coefficients of each backup feature, ensuring the accuracy of the determined weight coefficients of each backup feature. Backup features with weight coefficients less than a preset weight threshold are deleted to obtain the target features, ensuring the accuracy of the obtained target features.

[0036] In conjunction with the first aspect, in the fifth embodiment of the first aspect, an initial efficacy prediction network is trained using a training dataset to generate a target efficacy prediction model, including:

[0037] The initial efficacy prediction network was validated using the training dataset;

[0038] When the initial efficacy prediction network meets the preset requirements, the initial efficacy prediction network is trained using the training dataset to generate the target efficacy prediction model.

[0039] The efficacy prediction model training method provided in this embodiment of the invention uses a training dataset to verify the initial efficacy prediction network, ensuring that the initial efficacy prediction network is suitable for efficacy prediction. When the initial efficacy prediction network meets the preset requirements, the initial efficacy prediction network is trained using the training dataset to generate the target efficacy prediction model, thus ensuring the accuracy of the target efficacy prediction model trained using the initial efficacy prediction network.

[0040] In conjunction with the fifth embodiment of the first aspect, in the sixth embodiment of the first aspect, the initial efficacy prediction network is validated using a training dataset, including:

[0041] Divide the training dataset into a predetermined number of training subsets;

[0042] Each time, a subset of the training dataset is selected as the validation dataset, and the initial efficacy prediction network is trained using the remaining training datasets to obtain the initial efficacy prediction model corresponding to the initial efficacy prediction network.

[0043] This process is repeated until each training subset is used as a validation subset to train the initial efficacy prediction network.

[0044] The model accuracy of each initial efficacy prediction model obtained from training is obtained. When the average model accuracy of each initial efficacy prediction model is greater than the preset model accuracy threshold, the initial efficacy prediction network is determined to meet the preset requirements.

[0045] The efficacy prediction model training method provided in this invention divides the training dataset into a preset number of training subsets. Each time, one subset is selected as a validation subset, and the remaining subsets are used to train the initial efficacy prediction network, resulting in an initial efficacy prediction model. This ensures the accuracy of the initial efficacy prediction model. This process is repeated until all training subsets are used as validation subsets to train the initial efficacy prediction network. Then, the model accuracy of each initial efficacy prediction model is obtained. When the average model accuracy of each initial efficacy prediction model is greater than a preset model accuracy threshold, the initial efficacy prediction network is determined to meet the preset requirements, ensuring the accuracy of the determined initial efficacy prediction network.

[0046] According to a second aspect, embodiments of the present invention also provide a method for predicting therapeutic efficacy, comprising:

[0047] Obtain target clinical information, target EEG data, and target ECG data for patients whose treatment efficacy is to be predicted;

[0048] Based on the target clinical information, target EEG data, and target ECG data, generate the target features to be predicted for patients whose efficacy is to be predicted.

[0049] The target features to be predicted are input into the target efficacy prediction model, and the predicted efficacy corresponding to the patient to be predicted is output; wherein, the target efficacy prediction model is generated according to the efficacy prediction model training method in the first aspect or any of the embodiments of the first aspect.

[0050] The efficacy prediction method provided in this invention acquires target clinical information, target EEG data, and target ECG data corresponding to the patient whose efficacy is to be predicted. Based on the target clinical information, target EEG data, and target ECG data, predictable target features corresponding to the patient whose efficacy is to be predicted are generated, ensuring the accuracy of the generated predictable target features. Then, the predictable target features are input into the target efficacy prediction model, and the predicted efficacy corresponding to the patient whose efficacy is to be predicted is output, ensuring the accuracy of the output predicted efficacy corresponding to the patient whose efficacy is to be predicted.

[0051] According to a third aspect, embodiments of the present invention also provide a therapeutic efficacy prediction model training device, comprising:

[0052] The first acquisition module is used to acquire the clinical information, EEG data, ECG data and sample type corresponding to the target patient. The sample type is determined according to the long-term efficacy of the target patient and includes positive samples and negative samples.

[0053] The first generation module is used to generate initial features corresponding to the target patient based on the clinical information, EEG data, and ECG data of the target patient.

[0054] The optimization module is used to optimize the initial features based on the sample type corresponding to the target patient, and generate the target features corresponding to the target patient.

[0055] The second generation module is used to generate a training dataset for the target patient based on the correspondence between the target features and the sample types.

[0056] The training module is used to train the initial efficacy prediction network using the training dataset and generate the target efficacy prediction model.

[0057] The efficacy prediction model training device provided in this embodiment of the invention acquires clinical information, EEG data, and ECG data corresponding to the target patient, as well as the sample type corresponding to the target patient. Based on the clinical information, EEG data, and ECG data corresponding to the target patient, initial features corresponding to the target patient are generated, ensuring the accuracy and diversity of the generated initial features. Then, based on the sample type corresponding to the target patient, the initial features are optimized to generate target features corresponding to the target patient, ensuring the accuracy of the generated target features. Based on the correspondence between the target features and the sample type corresponding to the target patient, a training dataset corresponding to the target patient is generated, ensuring the accuracy of the generated training dataset and the accuracy of the data volume, avoiding the influence of too many invalid features on the training dataset, and preventing the training dataset from becoming too large, resulting in a long training time for training the initial efficacy prediction network using the training dataset. The initial efficacy prediction network is trained using the training dataset to generate the target efficacy prediction model, ensuring the accuracy of the generated target efficacy prediction model. The above device ensures the accuracy of the trained target efficacy prediction model, thereby ensuring the accuracy of predicting the treatment efficacy of patients using the target efficacy prediction model.

[0058] According to a fourth aspect, embodiments of the present invention also provide a therapeutic efficacy prediction device, comprising:

[0059] The second acquisition module is used to acquire the target clinical information, target EEG data and target ECG data corresponding to the patients whose efficacy is to be predicted.

[0060] The third acquisition module is used to generate the target features to be predicted for patients whose efficacy is to be predicted based on the target clinical information, target EEG data and target ECG data.

[0061] The output module is used to input the target features to be predicted into the target efficacy prediction model and output the predicted efficacy corresponding to the patient to be predicted; wherein, the target efficacy prediction model is generated according to the efficacy prediction model training method in the first aspect or any embodiment of the first aspect.

[0062] The efficacy prediction device provided in this invention acquires target clinical information, target EEG data, and target ECG data corresponding to the patient whose efficacy is to be predicted. Based on the target clinical information, target EEG data, and target ECG data, it generates predictable target features corresponding to the patient, ensuring the accuracy of the generated predictable target features. Then, the predictable target features are input into the target efficacy prediction model, and the predicted efficacy corresponding to the patient is output, ensuring the accuracy of the output predicted efficacy corresponding to the patient.

[0063] According to a fifth aspect, an embodiment of the present invention provides an electronic device, including a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the efficacy prediction model training method of the first aspect or any embodiment of the first aspect and the efficacy prediction method of the second aspect.

[0064] According to a sixth aspect, embodiments of the present invention provide a computer-readable storage medium storing computer instructions for causing a computer to execute the efficacy prediction model training method of the first aspect or any embodiment of the first aspect and the efficacy prediction method of the second aspect. Attached Figure Description

[0065] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0066] Figure 1 This is a flowchart of the training method for the efficacy prediction model provided in the embodiments of the present invention;

[0067] Figure 2 This is a flowchart of a method for training a therapeutic efficacy prediction model provided in another embodiment of the present invention;

[0068] Figure 3 This is a flowchart of the initial feature extraction provided by another embodiment of the present invention;

[0069] Figure 4 This is a flowchart of a method for training a therapeutic efficacy prediction model provided in another embodiment of the present invention;

[0070] Figure 5 This is a flowchart of a method for training a therapeutic efficacy prediction model provided in another embodiment of the present invention;

[0071] Figure 6 This is a flowchart of an efficacy prediction method provided by another embodiment of the present invention;

[0072] Figure 7 This is a functional block diagram of the efficacy prediction model training device provided in the embodiments of the present invention;

[0073] Figure 8 This is a functional block diagram of the efficacy prediction device provided in the embodiments of the present invention;

[0074] Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of the present invention. Detailed Implementation

[0075] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0076] It should be noted that the method for training the efficacy prediction model provided in this application can be executed by a device for training the efficacy prediction model. This device can be implemented as part or all of an electronic device through software, hardware, or a combination of both. The electronic device can be a server or a terminal. In this application, the server can be a single server or a server cluster composed of multiple servers. The terminal in this application can be a smartphone, personal computer, tablet computer, wearable device, or other intelligent hardware device such as a smart robot. The following method embodiments will use an electronic device as an example for explanation.

[0077] In one embodiment of this application, such as Figure 1 As shown, a method for training a therapeutic efficacy prediction model is provided. Taking the application of this method to electronic devices as an example, the method includes the following steps:

[0078] S11. Obtain the clinical information, EEG data, ECG data, and sample type corresponding to the target patient.

[0079] The sample type is determined based on the long-term efficacy for the target patients, and includes positive and negative samples.

[0080] In one optional embodiment of this application, the electronic device can receive clinical information, EEG data, ECG data, and sample type corresponding to the target patient input by the user; the electronic device can also receive clinical information, EEG data, ECG data, and sample type corresponding to the target patient sent by other devices.

[0081] In another optional embodiment of this application, the electronic device can receive clinical information corresponding to the target patient input by the user. Then, based on the connection with the detection device, it acquires the electroencephalogram (EEG) data and electrocardiogram (ECG) data corresponding to the target patient collected by the detection device, and then receives the sample type labeled by the user for the target patient.

[0082] In one optional embodiment of this application, the target patient may be a patient who meets preset requirements. For example, preset requirements may include: age under 16 years old; diagnosed with refractory epilepsy; regular programming for at least 1 year; stimulation current ≥1mA; scalp EEG recording within 6 months prior to surgery, with an effective duration of more than 4 hours (of which the awake duration is more than 40 minutes), etc. This application embodiment does not specifically limit the preset requirements.

[0083] The clinical information corresponding to the target patient may include: demographic information (gender, BMI), epilepsy diagnosis (age of onset, course of disease, etiology, syndrome diagnosis, MRI examination results), seizure information (seizure frequency, seizure type), treatment status (whether resection surgery has been performed, medication), etc. This application does not specifically limit the clinical information.

[0084] Among them, the sample types corresponding to the target patients are defined according to the long-term clinical treatment effect of VNS. Patients with a reduction of ≥50% in the frequency of attacks are defined as therapeutic responders (R) and marked as positive samples; patients with a reduction of <50% in the frequency of attacks are defined as therapeutic non-responders (NR) and marked as negative samples.

[0085] S12. Based on the clinical information, EEG data, and ECG data of the target patient, generate the initial features corresponding to the target patient.

[0086] In one optional embodiment of this application, the electronic device can generate initial clinical features based on clinical information, generate initial EEG features corresponding to EEG data based on EEG data, and then generate initial ECG features corresponding to ECG data based on ECG data.

[0087] This step will be explained in detail below.

[0088] S13. Based on the sample type corresponding to the target patient, optimize the initial features to generate the target features corresponding to the target patient.

[0089] In one optional embodiment of this application, the electronic device can generate multiple sets of initial feature combinations based on initial features. Then, based on the sample type corresponding to the target patient, a preset binary classification algorithm is trained using each set of initial feature combinations to obtain an initial binary classification model corresponding to each initial feature combination. Then, based on the model accuracy corresponding to each initial binary classification model, the optimal initial feature combination is determined from the multiple sets of initial feature combinations, thereby optimizing the initial features and generating target features corresponding to the target patient.

[0090] This step will be explained in detail below.

[0091] S14. Based on the correspondence between the target features and sample types corresponding to the target patients, generate the training dataset corresponding to the target patients.

[0092] Specifically, after determining the target features corresponding to the target patient, the electronic device can generate a training dataset for the target patient based on the correspondence between the target features and the sample types corresponding to the target patient. The sample types corresponding to the target patient are the label information of the target features.

[0093] S15. Use the training dataset to train the initial efficacy prediction network and generate the target efficacy prediction model.

[0094] Specifically, after generating the training dataset corresponding to the target patient, the electronic device can use the training dataset to train the initial efficacy prediction network and generate the target efficacy prediction model.

[0095] The initial efficacy prediction network can be any of the following binary classification networks: Naive Bayes network, Logistic Regression (LR) network, KNN (K Nearest Neighbors) network, Decision Tree network, SVM (Support Vector Machine) network, K-Means clustering network, etc. The embodiments of this application do not specifically limit the initial efficacy prediction network.

[0096] The efficacy prediction model training method provided in this invention acquires clinical information, EEG data, and ECG data corresponding to the target patient, as well as the sample type corresponding to the target patient. Based on the clinical information, EEG data, and ECG data corresponding to the target patient, initial features corresponding to the target patient are generated, ensuring the accuracy and diversity of the generated initial features. Then, based on the sample type corresponding to the target patient, the initial features are optimized to generate target features corresponding to the target patient, ensuring the accuracy of the generated target features. Based on the correspondence between the target features and the sample type corresponding to the target patient, a training dataset corresponding to the target patient is generated, ensuring the accuracy of the generated training dataset and the accuracy of the data volume, avoiding the influence of too many invalid features on the training dataset and preventing a large training dataset, which would result in a long training time for training the initial efficacy prediction network using the training dataset. The initial efficacy prediction network is trained using the training dataset to generate the target efficacy prediction model, ensuring the accuracy of the generated target efficacy prediction model. The above method ensures the accuracy of the trained target efficacy prediction model, thereby ensuring the accuracy of predicting the treatment efficacy of patients using the target efficacy prediction model.

[0097] In one embodiment of this application, such as Figure 2As shown, a method for training a therapeutic efficacy prediction model is provided. Taking the application of this method to electronic devices as an example, the method includes the following steps:

[0098] S21. Obtain the clinical information, EEG data, ECG data, and sample type corresponding to the target patient.

[0099] The sample type is determined based on the long-term efficacy for the target patients, and includes positive and negative samples.

[0100] For details on this step, please refer to [link / reference]. Figure 1 The details of S11 will not be elaborated here.

[0101] S22. Based on the clinical information, EEG data, and ECG data of the target patient, generate the initial features corresponding to the target patient.

[0102] In an optional embodiment of this application, the above-mentioned step S22, "generating initial features corresponding to the target patient based on the clinical information, EEG data, and ECG data corresponding to the target patient," may include the following steps:

[0103] S221. Generate initial clinical features based on clinical information; the initial clinical features include multiple initial sub-clinical features.

[0104] Specifically, electronic devices can read clinical information and generate initial clinical features based on that information.

[0105] For example, electronic devices can read clinical information, and the initial clinical features generated based on the clinical information may include age characteristics, epilepsy severity characteristics, seizure frequency, seizure type, treatment status characteristics, etc.

[0106] S222. Analyze the EEG data and calculate the initial functional connectivity features, initial network topology features, and initial entropy features corresponding to the target patient.

[0107] Among them, the initial functional connectivity features include multiple initial sub-functional connectivity features; the initial network topology features include multiple initial sub-network topology features; and the initial entropy features include multiple initial sub-entropy features.

[0108] In one optional embodiment of this application, the EEG data acquired by the electronic device can be 19-channel scalp EEG data collected during the patient's preoperative examination. The effective duration of the EEG data must be greater than 4 hours (with a conscious duration greater than 40 minutes). After acquiring the EEG data, the electronic device can identify the EEG data and the patient's state video during the EEG data acquisition process. Based on the identification results, the EEG data can be labeled with information such as seizures, discharges, wakefulness / sleep, and events.

[0109] The electronic device can then preprocess the EEG data. The EEG data preprocessing process may include the following steps:

[0110] a. Extract EEG data during the interictal period, avoiding abnormal data such as spikes and sharp waves, and the total duration of the extracted EEG data should be at least 10 minutes.

[0111] b. Use a 1-30Hz bandpass filter to filter the intercepted EEG data, retaining only the EEG data with a frequency between 1-30Hz;

[0112] c. Reset the reference point for the filtered EEG data (for example, the average voltage of the 19-channel EEG data can be used as the reference point, i.e., 0 point);

[0113] d. Divide the processed EEG data into segments, using 2-second intervals as the time unit;

[0114] e. Examine the segmented EEG data and remove abnormal segments;

[0115] f. Separate the independent components of the signal using Independent Component Analysis (ICA) and remove artifacts from the EEG data (such as interference from eye movement, ECG, and other signals).

[0116] For EEG data of different frequency bands, the following EEG data biomarkers are used as indicators for predicting therapeutic efficacy. In the embodiments of this application, the signal frequency bands involved in the electronic device calculation are: δ (1-4Hz), θ (4-7Hz), α (7-13Hz), lowβ (13-20Hz), and highβ (20-29Hz).

[0117] Functional Connectivity (FC) analysis assesses the functional relationships between brain regions by measuring the statistical dependence of signals across different brain regions. This helps us understand the synergistic relationships between brain regions and, consequently, the higher cognitive mechanisms of the brain. In this application, embodiments can use EEG data to calculate three functional connectivity indices: Phase Locking Value (PLV), Phase Lag Index (PLI), and Weighted Phase-Lag Index (wPLI), generating initial functional connectivity features.

[0118] For example, the phase-locked value (PLV) of two signals is defined as follows:

[0119]

[0120] Where N is the number of time points, For t n The phase difference between the two signals is calculated based on the Hilbert transform at a given time point.

[0121] The phase hysteresis index (PLI) of two signals is also calculated based on the phase difference:

[0122]

[0123] The weighted hysteresis index (wPLI) adds weights to the phase difference and is defined as follows:

[0124]

[0125] Where Im(Φ xy ) is the imaginary part of the cross spectrum of signals x and y.

[0126] Furthermore, epilepsy is considered a network-based disorder; therefore, analyzing functional networks and applying graph theory to epileptic seizure networks may contribute to a better understanding of the dynamics of seizures. Embodiments of this application can primarily use the functional connectivity matrix (PLV, PLI, wPLI) as the network's adjacency matrix to calculate the following network topology metrics: degree, characteristic path length, efficiency, clustering coefficient, and betweenness, generating initial network topology features. These network topology metrics can describe the characteristics of the network at different global and local scales.

[0127] For example, the degree of a node is defined as the number of edges connected to the node, and the average degree of all nodes is called the average degree of the network, calculated using the following formula:

[0128]

[0129] Where k i It is the degree of node i.

[0130] The distance between two nodes in a network is defined as the shortest path length between them. The characteristic path length of the network is defined as the average distance between any two nodes in the network, as shown in the formula:

[0131]

[0132] Where d ij It is the distance between i and j.

[0133] Considering the infinite distance between two unconnected nodes, which presents computational difficulties, we can calculate the connection efficiency between nodes, defined as the reciprocal of the distance between them. The average connection efficiency across all nodes is called the global efficiency of the network, and the formula is:

[0134]

[0135] The clustering coefficient is a parameter used to measure the clustering of nodes, that is, the probability of interconnection between all neighboring nodes. The clustering coefficient C of a network is defined as the average of the clustering coefficients of all nodes, i.e.:

[0136]

[0137] Betweenness centrality measures the extent to which a node is "middle" of other "pairs" of nodes in a graph. If a node is between multiple pairs of nodes, it is likely to play an important "mediating" role and thus be considered central to the network; this is how we measure a node's betweenness centrality. The betweenness centrality of a node is defined as the ratio of the shortest path passing through a given node and connecting those two nodes to the total number of shortest paths between those two nodes.

[0138] Furthermore, entropy analysis, based on Shannon's information theory to measure the distribution of signal features, can quantitatively describe the activity of the complex brain system and effectively reflect the potential activity characteristics of the nervous system. Compared to healthy individuals, the neuronal activity in the brains of epileptic patients exhibits greater uncertainty and randomness, and VNS can promote changes in the nervous system state of the target population. Therefore, entropy can effectively reflect some state changes in the nervous system that are not easily observed. The entropy analysis metrics used in the embodiments of this application include approximate entropy (ApEn), sample entropy (SampEn), wavelet entropy (WE), and Hilbert-Huang Spectrum Entropy (HHSE), thereby generating initial entropy features.

[0139] The concept of approximate entropy was first proposed to measure the complexity of a signal sequence. By calculating the probability of different new patterns arising in the signal, the complexity of the system can be determined; the more complex the time series, the greater the approximate entropy, and vice versa. Sample entropy is an optimization algorithm based on approximate entropy.

[0140] Wavelet entropy (WE) can also measure the complexity of a signal system; the difference is that wavelet entropy is based on wavelet transform, which decomposes any signal into basis components at different scales and calculates the energy distribution of the signal at different scales.

[0141] HHSE (Hilbert Huang Entropy) is a spectral entropy analysis performed using the Hilbert-Huang Transform (HHT). It uses the Empirical Mode Decomposition (EMD) method to decompose nonlinear and non-stationary signals, and obtains the corresponding Hilbert spectrum through the Hilbert Transform. HHSE calculates the energy distribution probability of the signal at different frequencies.

[0142] S223. Analyze the electrocardiogram data and calculate the initial heart rate variability characteristics corresponding to the target patient.

[0143] The initial heart rate variability features include multiple initial sub-heart rate variability features.

[0144] Specifically, the data acquisition device can acquire the target patient's electroencephalogram (EEG) data at the same time as the target patient's electrocardiogram (ECG) data. Therefore, the electronic device can acquire the target patient's EEG data as well as the target patient's ECG data.

[0145] After acquiring electrocardiogram (ECG) data, the electronic device can perform data preprocessing. The ECG data preprocessing process may include the following steps:

[0146] a. Identify the ECG data, detect the R wave peak (QRS complex) of the ECG data, and obtain the ECG data corresponding to the RR interval sequence.

[0147] b. Divide the ECG data corresponding to the above RR interval sequence into segments, using 2s as a time unit.

[0148] c. Random segment selection. Before partial indicator extraction (such as before functional connectivity indicator calculation), 20% of the total number of segments are randomly selected with replacement, and feature values ​​are calculated for each segment as the basic unit (biomarker calculation for each segment), and the results are averaged (for example, the results of 100 segments corresponding to the patient's data).

[0149] This invention quantifies ECG characteristics through the calculation of heart rate variability. Specifically, this includes time-domain characteristics, frequency-domain characteristics, and nonlinear characteristics. Specific characteristics are shown in Table 1 below.

[0150] Table 1 Heart Rate Variability Indicators

[0151]

[0152] S224. Initial features are constituted based on initial clinical features, initial functional connectivity features, initial network topology features, initial entropy features, and initial heart rate variability features.

[0153] Specifically, after acquiring initial clinical features, initial functional connectivity features, initial network topology features, initial entropy features, and initial heart rate variability features, the electronic device assembles initial features.

[0154] For example, such as Figure 3 The diagram shown is a flowchart illustrating the process of extracting initial features from an electronic device.

[0155] S23. Based on the sample type corresponding to the target patient, optimize the initial features to generate the target features corresponding to the target patient.

[0156] In an optional embodiment of this application, the above-mentioned step S23, "optimizing the initial features based on the sample type corresponding to the target patient to generate target features corresponding to the target patient," may include the following steps:

[0157] S231. Using a pre-defined test method, perform inter-group significance analysis on the initial features corresponding to positive samples and the initial features corresponding to negative samples in the target patients, and determine candidate features from the initial features.

[0158] The candidate features include candidate clinical features, candidate functional connectivity features, candidate network topology features, candidate entropy features, and candidate heart rate variability features.

[0159] Specifically, electronic devices can use the Mann-Whitney U test to analyze the significant differences between the initial features corresponding to positive and negative samples in the target patients, and determine candidate features from the initial features.

[0160] The Mann-Whitney U test, also known as the Mann-Whitney rank-sum test, is a commonly used nonparametric test. It assumes that the two samples come from two identical populations except for their population means, and aims to test whether there is a significant difference between the means of these two populations. The test statistic is calculated by mixing and ranking the points from the two samples and then calculating the sum of their rank in each sample. The difference between the two samples is reflected in the difference in their rank sums. When this statistic is less than a critical value, the null hypothesis is rejected, and a significant difference is considered between the two groups.

[0161] S232. Based on the recursive feature elimination method of the preset binary classification algorithm, the candidate features are selected to generate the target features corresponding to the target patient.

[0162] In an optional embodiment of this application, the above-mentioned step S232, "selecting features from candidate features using a recursive feature elimination method based on a preset binary classification algorithm to generate target features corresponding to the target patient," may include the following steps:

[0163] (1) Generate a first candidate feature set that includes all candidate features based on the candidate features;

[0164] (2) Input the first candidate feature set into the preset binary classification algorithm, train the preset binary classification algorithm, obtain the first candidate binary classification model, and determine the weight coefficients corresponding to each candidate feature in the first candidate feature set;

[0165] (3) Based on the weight coefficients of each candidate feature in the first candidate feature set, delete the candidate feature with the smallest weight coefficient from the candidate clinical feature, candidate functional connectivity feature, candidate network topology feature, candidate entropy feature and candidate heart rate variability feature respectively, to obtain the second candidate feature set.

[0166] (4) Repeat this process until the number of candidate features in the set reaches the preset requirement;

[0167] (5) Obtain the model accuracy corresponding to each candidate binary classification model, and determine the candidate features in the candidate feature set corresponding to the candidate binary classification model with the highest model accuracy as backup features;

[0168] (6) Determine the target features based on the backup features.

[0169] Specifically, after determining candidate features from the initial features, the electronic device can generate a first candidate feature set including all candidate features. Then, the first candidate feature set is input into a preset binary classification algorithm to train the algorithm, resulting in a first candidate binary classification model. The weight coefficients corresponding to each candidate sub-clinical feature in the candidate clinical features, each candidate sub-functional connectivity feature in the candidate functional connectivity features, each candidate sub-network topology feature in the candidate network topology features, each candidate sub-entropy feature in the candidate entropy features, and each candidate sub-heart rate variability feature in the candidate heart rate variability features are determined within the first candidate feature set.

[0170] Then, based on the weight coefficients corresponding to each candidate sub-clinical feature, candidate sub-functional connectivity feature, candidate sub-network topology feature, candidate sub-entropy feature, and candidate sub-heart rate variability feature, the feature with the smallest weight coefficient is removed from each of the candidate sub-clinical feature, candidate sub-functional connectivity feature, candidate sub-network topology feature, candidate sub-entropy feature, and candidate sub-heart rate variability feature, respectively, to obtain a second candidate feature set. The first preset number can be 1, 2, or 3; this embodiment does not specifically limit the first preset number. For example, when the first preset number is 2, the electronic device removes the two features with the smallest weight coefficients from each of the candidate sub-clinical feature, candidate sub-functional connectivity feature, candidate sub-network topology feature, candidate sub-entropy feature, and candidate sub-heart rate variability feature, respectively, to obtain the second candidate feature set.

[0171] Then, the second candidate feature set is input into a preset binary classification algorithm to train the algorithm, resulting in a second candidate binary classification model. The weight coefficients for each candidate sub-clinical feature, each candidate sub-functional connectivity feature, each candidate sub-network topology feature, each candidate sub-entropy feature, and each candidate sub-heart rate variability feature in the candidate clinical feature set are determined. Based on these weight coefficients, the features with the smallest weight coefficients are removed from each of the following categories: candidate clinical feature, candidate sub-functional connectivity feature, candidate sub-network topology feature, candidate sub-entropy feature, and candidate sub-heart rate variability feature, respectively, to obtain the third candidate feature set.

[0172] This process is repeated until the number of candidates in the feature set reaches a preset requirement. The preset requirement can be either the number of candidates in the feature set reaches a first preset number, or the number of a certain type of candidate feature data reaches a second preset number.

[0173] Then, the electronic device can use the 10-fold cross-validation method to verify the model accuracy of each candidate binary classification model, determine the candidate features in the candidate feature set corresponding to the candidate binary classification model with the highest model accuracy as backup features, and determine the target features based on the backup features.

[0174] In an optional embodiment of this application, step (6) "determining the target feature based on the backup feature" may include the following steps:

[0175] (61) Using the relationship between the backup features corresponding to the target patient and the sample type corresponding to the target patient, the preset binary classification algorithm is trained again, and the weight coefficients of each backup feature are determined.

[0176] (62) Delete the standby features whose weight coefficients are less than the preset weight threshold to obtain the target features.

[0177] Specifically, the electronic device uses the relationship between the backup features corresponding to the target patient and the sample type corresponding to the target patient to retrain the preset binary classification algorithm and determine the weight coefficient of each backup feature. Then, based on the determined weight coefficients of each backup feature, backup features with weight coefficients less than a preset weight threshold are deleted to obtain the target feature.

[0178] S24. Based on the correspondence between the target features and sample types corresponding to the target patients, generate the training dataset corresponding to the target patients.

[0179] For details on this step, please refer to [link / reference]. Figure 1 The details of S14 will not be elaborated here.

[0180] S25. Use the training dataset to train the initial efficacy prediction network and generate the target efficacy prediction model.

[0181] For details on this step, please refer to [link / reference]. Figure 1 The details of S15 will not be elaborated here.

[0182] The efficacy prediction model training method provided in this invention generates initial clinical features based on clinical information, ensuring the accuracy of the generated initial clinical features. It analyzes electroencephalogram (EEG) data to calculate the initial functional connectivity features, initial network topology features, and initial entropy features corresponding to the target patient, ensuring the accuracy of these calculated features. It also analyzes electrocardiogram (ECG) data to calculate the initial heart rate variability features corresponding to the target patient, ensuring the accuracy of the calculated features. Based on the initial clinical features, initial functional connectivity features, initial network topology features, initial entropy features, and initial heart rate variability features, initial features are constructed, ensuring the accuracy and diversity of the generated initial features, and guaranteeing that the generated initial features include analysis results from multiple perspectives and modalities.

[0183] Then, using a pre-defined test method, a significant difference analysis was performed between groups on the initial features corresponding to the positive samples and the initial features corresponding to the negative samples in the target patients. Candidate features were determined from the initial features, ensuring the accuracy of the candidate features determined from the initial features and removing redundant features from the initial features to avoid the influence of redundant features on the candidate features.

[0184] Then, based on the candidate features, a first candidate feature set including all candidate features is generated. This first candidate feature set is input into a preset binary classification algorithm to train it, resulting in a first candidate binary classification model. The weight coefficients corresponding to each candidate feature in the first candidate feature set are determined, ensuring the accuracy of these weight coefficients. Based on the weight coefficients of each candidate feature in the first candidate feature set, the candidate feature with the smallest weight coefficient is removed from candidate clinical features, candidate functional connectivity features, candidate network topology features, candidate entropy features, and candidate heart rate variability features, respectively, resulting in a second candidate feature set. This ensures the accuracy of the second candidate feature set and the importance of each candidate feature within it. This process is repeated until the number of candidate features in the set reaches a preset requirement. The model accuracy of each candidate binary classification model is obtained, and the candidate features in the candidate feature set corresponding to the candidate binary classification model with the highest accuracy are selected as backup features, ensuring the accuracy of the selected backup features. Then, by utilizing the relationship between the backup features corresponding to the target patient and the sample type corresponding to the target patient, the preset binary classification algorithm is trained again, and the weight coefficients of each backup feature are determined, ensuring the accuracy of the determined weight coefficients of each backup feature. Backup features with weight coefficients less than a preset weight threshold are deleted to obtain the target features, ensuring the accuracy of the obtained target features. Furthermore, redundant features are deleted from the candidate features to avoid the influence of redundant features on the target features, ensuring the accuracy of the data training set generated based on the target features, and thus ensuring the accuracy of the target efficacy prediction model trained using the training dataset.

[0185] In one embodiment of this application, such as Figure 4 As shown, a method for training a therapeutic efficacy prediction model is provided. Taking the application of this method to electronic devices as an example, the method includes the following steps:

[0186] S31. Obtain the clinical information, EEG data, ECG data, and sample type corresponding to the target patient.

[0187] The sample type is determined based on the long-term efficacy for the target patients, and includes positive and negative samples.

[0188] For details on this step, please refer to [link / reference]. Figure 2The details of S21 will not be elaborated here.

[0189] S32. Based on the clinical information, EEG data, and ECG data of the target patient, generate the initial features corresponding to the target patient.

[0190] For details on this step, please refer to [link / reference]. Figure 2 The details of S22 will not be elaborated here.

[0191] S33. Based on the sample type corresponding to the target patient, optimize the initial features to generate the target features corresponding to the target patient.

[0192] For details on this step, please refer to [link / reference]. Figure 2 The details of S23 will not be elaborated here.

[0193] S34. Based on the correspondence between the target features and sample types corresponding to the target patients, generate the training dataset corresponding to the target patients.

[0194] For details on this step, please refer to [link / reference]. Figure 2 The details of S24 will not be elaborated here.

[0195] S35. Train the initial efficacy prediction network using the training dataset to generate the target efficacy prediction model.

[0196] In an optional embodiment of this application, the above-mentioned step S35, "training an initial efficacy prediction network using a training dataset to generate a target efficacy prediction model," may include the following steps:

[0197] S351. Validate the initial efficacy prediction network using the training dataset.

[0198] In an optional embodiment of this application, the above-mentioned step S351, "validating the initial efficacy prediction network using the training dataset," may include the following steps:

[0199] (1) Divide the training dataset into a predetermined number of training subsets;

[0200] (2) Each time, select one part from the training subset as the validation subset, and use the remaining training subset to train the initial efficacy prediction network to obtain the initial efficacy prediction model corresponding to the initial efficacy prediction network.

[0201] (3) Repeat this process until each training subset is used as a validation subset to train the initial efficacy prediction network.

[0202] (4) Obtain the model accuracy of each initial efficacy prediction model obtained from training. When the average value of the model accuracy of each initial efficacy prediction model is greater than the preset model accuracy threshold, it is determined that the initial efficacy prediction network meets the preset requirements.

[0203] Specifically, the electronic device can divide the training dataset into a preset number of training subsets. Then, each time, one subset is selected as a validation subset, and the initial efficacy prediction network is trained using the remaining training subsets to obtain the initial efficacy prediction model. This process is repeated until all training subsets are used as validation subsets to train the initial efficacy prediction network. The electronic device then uses the validation subsets corresponding to each initial efficacy prediction model to validate the model and obtain the model accuracy of each trained initial efficacy prediction model. Finally, the model accuracy of each initial efficacy prediction model is averaged. When the average model accuracy of all initial efficacy prediction models is greater than a preset model accuracy threshold, the initial efficacy prediction network is determined to meet the preset requirements.

[0204] For example, the electronic device can split the training dataset into 10 training subsets. The electronic device can select one subset from the 10 subsets as the detection subset each time. For instance, it might select the first subset initially, then use the second to tenth subsets to train the initial efficacy prediction network, generating the first initial efficacy prediction model. This process continues, obtaining the second initial efficacy prediction model, and so on, until the tenth initial efficacy prediction model is obtained. The electronic device then obtains the model accuracy corresponding to each of the tenth initial efficacy prediction models, and calculates the average of the model accuracies of each initial efficacy prediction model. When the average model accuracy of each initial efficacy prediction model is greater than a preset model accuracy threshold, the initial efficacy prediction network is determined to meet the preset requirements.

[0205] The initial efficacy prediction network can be any of the following binary classification networks: Naive Bayes network, Logistic Regression (LR) network, KNN (K Nearest Neighbors) network, Decision Tree network, SVM (Support Vector Machine) network, K-Means clustering network, etc. The embodiments of this application do not specifically limit the initial efficacy prediction network.

[0206] S352. When the initial efficacy prediction network meets the preset requirements, the initial efficacy prediction network is trained using the training dataset to generate the target efficacy prediction model.

[0207] Specifically, when the initial efficacy prediction network meets the preset requirements, the initial efficacy prediction network is trained using the training dataset to generate the target efficacy prediction model.

[0208] Then, the electronic device can acquire a data validation set, perform feature processing on the data validation set to generate target features, input the target features into the target efficacy prediction model, perform external validation on the target efficacy prediction model, and ensure the model accuracy of the target efficacy prediction model.

[0209] For example, such as Figure 5 The diagram shows one optional workflow step for generating a target efficacy prediction model.

[0210] The efficacy prediction model training method provided in this invention divides the training dataset into a preset number of training subsets. Each time, one subset is selected as a validation subset, and the remaining subsets are used to train an initial efficacy prediction network, resulting in an initial efficacy prediction model. This ensures the accuracy of the initial efficacy prediction model. This process is repeated until all training subsets are used as validation subsets to train the initial efficacy prediction network. Then, the model accuracy of each initial efficacy prediction model is obtained. When the average model accuracy of each initial efficacy prediction model is greater than a preset model accuracy threshold, the initial efficacy prediction network is determined to meet the preset requirements, ensuring the accuracy of the determined initial efficacy prediction network. When the initial efficacy prediction network meets the preset requirements, the training dataset is used to train the initial efficacy prediction network to generate a target efficacy prediction model, ensuring the accuracy of the target efficacy prediction model trained using the initial efficacy prediction network.

[0211] To better illustrate the efficacy prediction method provided in the embodiments of this application, the embodiments of this application also provide an efficacy prediction method, such as... Figure 6 As shown, the method is applied to an electronic device as an example, and includes the following steps:

[0212] S41. Obtain target clinical information, target EEG data, and target ECG data for patients whose treatment efficacy is to be predicted.

[0213] In one optional embodiment of this application, the electronic device can receive target clinical information, target EEG data, and target ECG data corresponding to the patient whose efficacy is to be predicted, input by the user; the electronic device can also receive target clinical information, target EEG data, and target ECG data corresponding to the patient whose efficacy is to be predicted sent by other devices.

[0214] In another optional embodiment of this application, the electronic device can receive target clinical information corresponding to the patient whose treatment efficacy is to be predicted, input by the user. Then, based on the connection with the detection device, it acquires the target EEG data and target ECG data corresponding to the patient whose treatment efficacy is to be predicted, collected by the detection device.

[0215] This application does not specifically limit the method by which electronic devices acquire target clinical information, target EEG data, and target ECG data corresponding to patients whose therapeutic effects are to be predicted.

[0216] S42. Based on the target clinical information, target EEG data, and target ECG data, generate the target features to be predicted for patients whose efficacy is to be predicted.

[0217] Specifically, the electronic device generates the clinical features of the target to be predicted based on the target clinical information; analyzes the target EEG data to calculate the functional connectivity features, network topology features, and entropy features of the target to be predicted for the patients whose efficacy is to be predicted; and analyzes the target ECG data to calculate the heart rate variability features of the target to be predicted for the patients whose efficacy is to be predicted.

[0218] S43. Input the target features to be predicted into the target efficacy prediction model, and output the predicted efficacy for the patient to be predicted.

[0219] The target efficacy prediction model is generated according to the efficacy prediction model training method in the first aspect or any embodiment of the first aspect.

[0220] Specifically, the electronic device inputs the target features to be predicted into the target efficacy prediction model and outputs the predicted efficacy for the patient to be predicted.

[0221] The efficacy prediction method provided in this invention acquires target clinical information, target EEG data, and target ECG data corresponding to the patient whose efficacy is to be predicted. Based on the target clinical information, target EEG data, and target ECG data, predictable target features corresponding to the patient whose efficacy is to be predicted are generated, ensuring the accuracy of the generated predictable target features. Then, the predictable target features are input into the target efficacy prediction model, and the predicted efficacy corresponding to the patient whose efficacy is to be predicted is output, ensuring the accuracy of the output predicted efficacy corresponding to the patient whose efficacy is to be predicted.

[0222] It should be understood that, although Figure 1 , Figure 2 , Figure 4 as well as Figure 6 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 , Figure 2 , Figure 4 as well as Figure 6At least some of the steps in the process may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but may be executed at different times. The execution order of these steps or stages is not necessarily sequential, but may be executed in turn or alternately with other steps or at least some of the steps or stages in other steps.

[0223] like Figure 7 As shown, this embodiment provides a therapeutic efficacy prediction model training device, including:

[0224] The first acquisition module 51 is used to acquire clinical information, EEG data, ECG data and sample type corresponding to the target patient. The sample type is determined according to the long-term efficacy of the target patient and includes positive samples and negative samples.

[0225] The first generation module 52 is used to generate initial features corresponding to the target patient based on the clinical information, EEG data, and ECG data corresponding to the target patient.

[0226] Optimization module 53 is used to optimize the initial features based on the sample type corresponding to the target patient, and generate the target features corresponding to the target patient.

[0227] The second generation module 54 is used to generate a training dataset corresponding to the target patient based on the correspondence between the target features and sample types corresponding to the target patient.

[0228] Training module 55 is used to train the initial efficacy prediction network using the training dataset and generate the target efficacy prediction model.

[0229] In one embodiment of this application, the first generation module 52 is specifically used to generate initial clinical features based on clinical information; the initial clinical features include multiple initial sub-clinical features; analyze electroencephalogram (EEG) data to calculate initial functional connectivity features, initial network topology features, and initial entropy features corresponding to the target patient; the initial functional connectivity features include multiple initial sub-functional connectivity features; the initial network topology features include multiple initial sub-network topology features; the initial entropy features include multiple initial sub-entropy features; analyze electrocardiogram (ECG) data to calculate initial heart rate variability features corresponding to the target patient; the initial heart rate variability features include multiple initial sub-heart rate variability features; and construct initial features based on the initial clinical features, initial functional connectivity features, initial network topology features, initial entropy features, and initial heart rate variability features.

[0230] In one embodiment of this application, the optimization module 53 is specifically used to perform inter-group significant difference analysis on the initial features corresponding to positive samples and the initial features corresponding to negative samples in the target patient using a preset testing method, and to determine candidate features from the initial features; the candidate features include candidate clinical features, candidate functional connectivity features, candidate network topology features, candidate entropy features, and candidate heart rate variability features; and to perform feature selection on the candidate features based on a recursive feature elimination method using a preset binary classification algorithm to generate target features corresponding to the target patient.

[0231] In one embodiment of this application, the optimization module 53 is specifically used to generate a first candidate feature set including all candidate features based on the candidate features; input the first candidate feature set into a preset binary classification algorithm, train the preset binary classification algorithm to obtain a first candidate binary classification model, and determine the weight coefficients corresponding to each candidate feature in the first candidate feature set; based on the weight coefficients corresponding to each candidate feature in the first candidate feature set, delete the candidate feature with the smallest weight coefficient from candidate clinical features, candidate functional connectivity features, candidate network topology features, candidate entropy features, and candidate heart rate variability features, respectively, to obtain a second candidate feature set; repeat this process until the number of candidates in the candidate feature set reaches a preset requirement; obtain the model accuracy corresponding to each candidate binary classification model, and determine the candidate feature in the candidate feature set corresponding to the candidate binary classification model with the highest model accuracy as a backup feature; and determine the target feature based on the backup feature.

[0232] In one embodiment of this application, the optimization module 53 is specifically used to retrain the preset binary classification algorithm by utilizing the relationship between the backup features corresponding to the target patient and the sample type corresponding to the target patient, and to determine the weight coefficient of each backup feature; and to delete backup features with weight coefficients less than the preset weight threshold to obtain the target features.

[0233] In one embodiment of this application, the training module 55 is specifically used to verify the initial efficacy prediction network using the training dataset; when the initial efficacy prediction network meets the preset requirements, the initial efficacy prediction network is trained using the training dataset to generate the target efficacy prediction model.

[0234] In one embodiment of this application, the training module 55 is specifically used to divide the training dataset into a preset number of training subsets; each time, one subset is selected from the training subset as a validation subset, and the initial efficacy prediction network is trained using the remaining training subsets to obtain the initial efficacy prediction model corresponding to the initial efficacy prediction network; this process is repeated until each training subset is used as a validation subset to train the initial efficacy prediction network; the model accuracy of each initial efficacy prediction model is obtained, and when the average model accuracy of each initial efficacy prediction model is greater than a preset model accuracy threshold, it is determined that the initial efficacy prediction network meets the preset requirements.

[0235] like Figure 8 As shown, this embodiment provides a therapeutic efficacy prediction model training device, including:

[0236] The second acquisition module 61 is used to acquire target clinical information, target EEG data and target ECG data corresponding to patients whose efficacy is to be predicted.

[0237] The third acquisition module 62 is used to generate the target features to be predicted for patients whose efficacy is to be predicted based on the target clinical information, target EEG data and target ECG data.

[0238] The output module 62 is used to input the target features to be predicted into the target efficacy prediction model and output the predicted efficacy corresponding to the patient to be predicted; wherein, the target efficacy prediction model is generated according to the efficacy prediction model training method in the first aspect or any embodiment of the first aspect.

[0239] For specific limitations and beneficial effects regarding the efficacy prediction model training device and the efficacy prediction device, please refer to the limitations of the efficacy prediction model training method and the efficacy prediction method mentioned above, which will not be repeated here. Each module in the aforementioned efficacy prediction model training device and efficacy prediction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in the electronic device in hardware form, or stored in the memory of the electronic device in software form, so that the processor can call and execute the corresponding operations of each module.

[0240] This invention also provides an electronic device having the above-described features. Figure 7 The therapeutic effect prediction model training device shown and Figure 8 The therapeutic effect prediction device shown.

[0241] like Figure 9 As shown, Figure 9 This is a schematic diagram of the structure of an electronic device provided in an optional embodiment of the present invention, such as... Figure 9As shown, the electronic device may include: at least one processor 71, such as a CPU (Central Processing Unit), at least one communication interface 73, memory 74, and at least one communication bus 72. The communication bus 72 is used to enable communication between these components. The communication interface 73 may include a display screen or a keyboard; optionally, the communication interface 73 may also include a standard wired interface or a wireless interface. The memory 74 may be high-speed RAM (Random Access Memory) or non-volatile memory, such as at least one disk storage device. Optionally, the memory 74 may also be at least one storage device located remotely from the aforementioned processor 71. The processor 71 may be combined with... Figure 7 as well as Figure 8 The described apparatus has an application program stored in memory 74, and the processor 71 calls the program code stored in memory 74 to perform any of the above method steps.

[0242] The communication bus 72 can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The communication bus 72 can be divided into an address bus, a data bus, and a control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0243] The memory 74 may include volatile memory, such as random-access memory (RAM); the memory may also include non-volatile memory, such as flash memory, hard disk drive (HDD) or solid-state drive (SSD); the memory 74 may also include a combination of the above types of memory.

[0244] The processor 71 can be a central processing unit (CPU), a network processor (NP), or a combination of CPU and NP.

[0245] The processor 71 may further include a hardware chip. This hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0246] Optionally, memory 74 is also used to store program instructions. Processor 71 can call the program instructions to implement the functions described in this application. Figure 1 , Figure 2 , Figure 4 The efficacy prediction model training method shown in the embodiments and Figure 6 The efficacy prediction method shown in the examples.

[0247] This invention also provides a non-transitory computer storage medium storing computer-executable instructions that can execute the efficacy prediction model training method and efficacy prediction method in any of the above method embodiments. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium may also include combinations of the above types of memory.

[0248] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for training a therapeutic efficacy prediction model, characterized in that, include: Acquire clinical information, EEG data, ECG data, and sample type corresponding to the target patient. The sample type is determined based on the long-term efficacy of the target patient and includes positive and negative samples. Based on the clinical information, EEG data, and ECG data corresponding to the target patient, initial features corresponding to the target patient are generated; Based on the sample type corresponding to the target patient, the initial features are optimized to generate the target features corresponding to the target patient; Based on the correspondence between the target features and the sample type corresponding to the target patient, a training dataset corresponding to the target patient is generated; The initial efficacy prediction network was trained using the training dataset to generate the target efficacy prediction model. Specifically, based on the clinical information, EEG data, and ECG data corresponding to the target patient, initial features corresponding to the target patient are generated, including: Based on the clinical information, an initial clinical feature is generated; the initial clinical feature includes multiple initial sub-clinical features; the initial sub-clinical features include at least age features, epilepsy severity features, seizure frequency, seizure type, and treatment status features. The EEG data is analyzed to calculate the initial functional connectivity features, initial network topology features, and initial entropy features corresponding to the target patient. The initial functional connectivity features include multiple initial sub-functional connectivity features; the initial network topology features include multiple initial sub-network topology features; and the initial entropy features include multiple initial sub-entropy features. The initial functional connectivity features include: phase lock value, hysteresis index, and weighted phase hysteresis index; the initial network topology features include: degree, feature path length and efficiency, clustering coefficient, and betweenness coefficient; and the initial entropy features include: approximate entropy, sample entropy, wavelet entropy, and Hilbert-Huang spectrum entropy. The electrocardiogram data is analyzed to calculate the initial heart rate variability characteristics corresponding to the target patient; the initial heart rate variability characteristics include multiple initial sub-heart rate variability characteristics; The initial features are constituted based on the initial clinical features, the initial functional connectivity features, the initial network topology features, the initial entropy features, and the initial heart rate variability features.

2. The method according to claim 1, characterized in that, The step of optimizing the initial features based on the sample type corresponding to the target patient to generate target features corresponding to the target patient includes: Using a pre-defined testing method, a significant difference analysis between groups is performed on the initial features corresponding to positive samples and the initial features corresponding to negative samples in the target patients to determine candidate features from the initial features; the candidate features include candidate clinical features, candidate functional connectivity features, candidate network topology features, candidate entropy features, and candidate heart rate variability features; The recursive feature elimination method based on the preset binary classification algorithm is used to select the candidate features and generate the target features corresponding to the target patient.

3. The method according to claim 2, characterized in that, The recursive feature elimination method based on a preset binary classification algorithm performs feature selection on the candidate features to generate target features corresponding to the target patient, including: Based on the candidate features, a first candidate feature set including all the candidate features is generated; The first candidate feature set is input into a preset binary classification algorithm to train the preset binary classification algorithm, thereby obtaining a first candidate binary classification model, and determining the weight coefficients corresponding to each candidate feature in the first candidate feature set. Based on the weight coefficients corresponding to each candidate feature in the first candidate feature set, the candidate feature with the smallest weight coefficient is deleted from the candidate clinical feature, the candidate functional connectivity feature, the candidate network topology feature, the candidate entropy feature, and the candidate heart rate variability feature, respectively, to obtain the second candidate feature set; This process is repeated until the number of candidate features in the set reaches the preset requirement. Obtain the model accuracy corresponding to each candidate binary classification model, and determine the candidate features in the candidate feature set corresponding to the candidate binary classification model with the highest model accuracy as backup features; The target feature is determined based on the backup feature.

4. The method according to claim 3, characterized in that, The step of determining the target feature based on the backup feature includes: By utilizing the relationship between the backup features corresponding to the target patient and the sample type corresponding to the target patient, the preset binary classification algorithm is trained again, and the weight coefficients of each backup feature are determined. The target feature is obtained by deleting the backup features whose weight coefficients are less than a preset weight threshold.

5. The method according to claim 1, characterized in that, The step of training an initial efficacy prediction network using the training dataset to generate a target efficacy prediction model includes: The initial efficacy prediction network was validated using the training dataset. When the initial efficacy prediction network meets the preset requirements, the initial efficacy prediction network is trained using the training dataset to generate the target efficacy prediction model.

6. The method according to claim 5, characterized in that, The validation of the initial efficacy prediction network using the training dataset includes: The training dataset is divided into a predetermined number of training sub-datasets; Each time, a subset of the training dataset is selected as a validation dataset, and the initial efficacy prediction network is trained using the remaining training datasets to obtain the initial efficacy prediction model corresponding to the initial efficacy prediction network. This process is repeated until each of the training subsets is used as a validation subset to train the initial efficacy prediction network. The model accuracy of each of the initial efficacy prediction models obtained from training is obtained. When the average model accuracy of each of the initial efficacy prediction models is greater than a preset model accuracy threshold, it is determined that the initial efficacy prediction network meets the preset requirements.

7. A method for predicting therapeutic efficacy, characterized in that, include: Obtain target clinical information, target EEG data, and target ECG data for patients whose treatment efficacy is to be predicted; Based on the target clinical information, the target EEG data, and the target ECG data, generate the target features to be predicted for the patients whose treatment efficacy is to be predicted; The target feature to be predicted is input into the target efficacy prediction model, and the predicted efficacy corresponding to the patient to be predicted is output; wherein, the target efficacy prediction model is generated by the efficacy prediction model training method according to any one of claims 1-6.

8. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores computer instructions, and the processor executes the computer instructions to perform the efficacy prediction model training method of any one of claims 1-6 and the efficacy prediction method of claim 7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the efficacy prediction model training method according to any one of claims 1-6 and the efficacy prediction method according to claim 7.

Images