Epilepsy auxiliary evaluation method and device based on heart rate variability, equipment and medium

Abstract

The application relates to the technical field of medical health, in particular to an epilepsy auxiliary evaluation method and device based on heart rate variability, equipment and a medium. The method comprises the following steps: acquiring at least two target physiological states in a current acquisition combination, including a resting state and one or two stimulation states; collecting electrocardio signals of a subject based on the target physiological states; performing heart rate variability analysis on each group of electrocardio signals based on a preset signal processing algorithm, obtaining a corresponding resting state HRV feature set and at least one stimulation state HRV feature set, and the HRV feature set comprises multiple feature items; determining a group of target feature items suitable for the current acquisition combination according to independent discrimination efficiency, correlation discrimination efficiency and / or global discrimination efficiency of each feature item on epilepsy; and generating a screening report for assisting in epilepsy risk evaluation based on the target feature items. While ensuring evaluation accuracy, operation efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of medical and health technology, and in particular to a method, device, equipment and medium for epilepsy auxiliary assessment based on heart rate variability. Background Technology

[0002] Epilepsy is a chronic neurological disorder caused by abnormal electrical activity in the brain. According to the World Health Organization, there are approximately 50 million people with epilepsy worldwide, with nearly 80% concentrated in low- and middle-income countries. Early identification and intervention of high-risk individuals are crucial for reducing the harm caused by epileptic seizures and improving prognosis. However, current clinical diagnosis mainly relies on advanced tools such as electroencephalography (EEG), neuroimaging, and genetic testing, as well as clinical symptom observation. These methods are expensive, complex, and have low compliance rates. This is particularly problematic in underdeveloped regions where high costs and low detection rates further limit diagnosis, leading to delays that increase the burden on patients and society.

[0003] A key aspect of epilepsy is autonomic nervous system dysfunction, a mechanism closely associated with adverse outcomes, including an increased risk of cardiovascular complications and sudden epileptic death. Therefore, there is a need to shift clinical practice from passive treatment to a more proactive predictive-prevention-personalized medicine (PPPM / 3PM) pathway. Heart rate variability (HRV) monitoring, as a non-invasive method, can be used to assess autonomic nervous function that is typically impaired in patients with epilepsy.

[0004] For example, patent application CN 120656745A discloses a system and method for mining and warning of HRV features in the epileptic seizure cycle. It collects multi-channel electrocardiogram signals through wearable devices, extracts RR intervals in layers, constructs high-dimensional manifolds and dynamic graphs, combines quantum state attention and chaotic pooling to identify key nodes and attractor patterns, extracts epileptic risk dynamic features, generates multidimensional risk scores and dynamically calibrates them, and finally outputs graded warnings and intervention suggestions to achieve intelligent prediction and management of epileptic seizures. It reveals the intrinsic geometric characteristics of HRV data through manifold mapping technology, and compared with traditional Euclidean space analysis methods, it can more accurately measure the true distance between different physiological states.

[0005] For example, patent application CN119073917A discloses a system for monitoring the course of consciousness disorders. A multi-dimensional data acquisition module uses multiple sensors installed on the patient's body to collect real-time physiological signal data such as electroencephalogram (EEG), heart rate variability (HRV), and electrical conductance of the skin (EDA). The collected data is preprocessed using edge computing devices. An adaptive course prediction model, based on the physiological signal data collected by the multi-dimensional data acquisition module and combined with the patient's historical disease data, uses an adaptive algorithm to train a personalized course prediction model. During the progression of the patient's disease, the model parameters are adaptively updated using machine learning algorithms to achieve dynamic prediction of the progression of consciousness disorders. A continuous course monitoring module uses high-frequency sampling technology to continuously acquire the patient's physiological signals and uses an anomaly detection algorithm to identify key events in the course of the disease in real time, including sudden decline in consciousness and epileptic seizures, generating warning signals and triggering an emergency response mechanism.

[0006] However, current HRV detection and analysis require a high quantity and accuracy of data, and the overall process is time-consuming and complex. Especially in underdeveloped areas with scarce medical resources, the high time and labor costs and cumbersome procedures make it difficult to promote and apply the technology. Summary of the Invention

[0007] The main objective of this application is to provide a method, device, equipment, and medium for auxiliary assessment of epilepsy based on heart rate variability. To solve the aforementioned technical problems, this application specifically adopts the following technical solution:

[0008] A first aspect of this application is to provide a method for epilepsy auxiliary assessment based on heart rate variability, the method comprising:

[0009] S101, acquire at least two target physiological states in the current acquisition combination, the target physiological states including resting state and one or two stimulus states;

[0010] S102, based on the target physiological state, collect the subject's resting electrocardiogram signal and at least one set of stimulated electrocardiogram signals respectively;

[0011] S103, based on a preset signal processing algorithm, heart rate variability analysis is performed on each group of electrocardiogram signals to obtain the corresponding resting state HRV feature set and at least one stimulated state HRV feature set, wherein the HRV feature set includes multiple feature items;

[0012] S104, Based on the independent discriminant efficacy of each feature item for epilepsy under each target physiological state, as well as the associated discriminant efficacy between at least two target physiological states and / or the global discriminant efficacy under all physiological states, determine a set of target features applicable to the current acquisition combination.

[0013] S105, Based on the target feature items in the resting-state HRV feature set and the stimulated-state HRV feature set, generate a screening report for assisting in epilepsy risk assessment.

[0014] In some embodiments, the stimulus state is induced by any of the following: sustained grip force action, Valsalva maneuver, or upright tilting maneuver.

[0015] In some embodiments, the target features include a first type of feature and a second type of feature; the first type of feature is determined based on global discrimination performance, and the second type of feature is dynamically adjusted based on the acquisition combination.

[0016] In some embodiments, the feature items in the HRV feature set include RMSSD, SDNN, NN50, pNN50, Triangular.index, AvgRR, HF, LF, TP, SD1, SD2, AC, and DC; the first type of feature items includes RMSSD and SD1.

[0017] In some embodiments, S104 includes: determining the number of second-class feature items to be selected based on the number of HRV feature sets; calculating the association discriminant efficacy of each of the two types of feature items based on the independent discriminant efficacy of each feature item for epilepsy under the target physiological state; and sorting the feature items from high to low and selecting the corresponding number of feature items as target feature items according to the number of second-class feature items selected and the association discriminant efficacy.

[0018] In some embodiments, the method further includes: obtaining clinical diagnostic reports of epileptic subjects and non-epileptic subjects, and HRV feature sets in the resting state and at least one stimulus state, as training data; training a machine learning model based on the training data to obtain an epilepsy risk prediction model; inputting the specific parameter values ​​of the target feature items into the epilepsy risk prediction model to output a risk score for generating a screening report to assist in epilepsy risk assessment.

[0019] In some embodiments, the ECG signal acquisition device includes at least one of: a wearable device with ECG measurement function, a single-lead or multi-lead ECG signal acquisition device.

[0020] A second aspect of this application is to provide an epilepsy auxiliary assessment device based on heart rate variability, the device comprising:

[0021] A physiological state acquisition module is used to acquire at least two target physiological states in the current acquisition combination, wherein the target physiological states include a resting state and one or two stimulus states;

[0022] An electrocardiogram (ECG) signal acquisition module is used to acquire the subject's resting ECG signal and at least one set of stimulated ECG signals based on the target physiological state.

[0023] The heart rate variability analysis module is used to perform heart rate variability analysis on each group of electrocardiogram signals based on a preset signal processing algorithm, and obtain the corresponding resting state HRV feature set and at least one stimulated state HRV feature set, wherein the HRV feature set includes multiple feature items.

[0024] The target feature determination module is used to determine a set of target features applicable to the current acquisition combination based on the independent discriminant efficacy of each feature in each target physiological state, the associated discriminant efficacy between at least two target physiological states and / or the global discriminant efficacy in all physiological states.

[0025] The screening report generation module is used to generate a screening report for assisting epilepsy risk assessment based on the target feature items in the resting-state HRV feature set and the stimulated-state HRV feature set.

[0026] A third aspect of this application is to provide a computer device, the device comprising:

[0027] Memory, used to store computer programs;

[0028] A processor is configured to execute the computer program and, in executing the computer program, implement the steps of the epilepsy auxiliary assessment method based on heart rate variability as provided in any embodiment of this application.

[0029] A fourth aspect of this application is that a computer-readable storage medium is also provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the epilepsy auxiliary assessment method based on heart rate variability provided in any embodiment of this application.

[0030] Beneficial effects:

[0031] This application provides a method, device, equipment, and medium for epilepsy auxiliary assessment based on heart rate variability. Based on the free combination of resting state and any stimulus state, it simplifies the screening process for epilepsy auxiliary assessment using heart rate variability, improves operational efficiency while ensuring assessment accuracy, and is especially suitable for scenarios with limited medical resources and high timeliness requirements, such as primary care, emergency triage, and community follow-up.

[0032] Specifically, only ECG signals in the resting state and any stimulation state need to be collected, significantly reducing the complexity and time cost of data acquisition. In underdeveloped or primary healthcare environments, operators only need to master fewer testing procedures and standards, improving feasibility and helping to increase the operational accuracy of a single test, thus improving overall data quality.

[0033] Based on this, according to the actual physiological state combinations collected and the differences in the three-level discrimination efficiency of different HRV feature items, target feature items for risk assessment are dynamically screened. While minimizing the loss of key information, the interference of irrelevant or weakly related features on epilepsy assessment is effectively suppressed, thereby improving the accuracy of screening.

[0034] The collection and combination of HRV data are flexible and varied, and the types of HRV features are also diverse. Modeling and validating the discriminative value of each feature item for each application scenario is prohibitively expensive, especially in resource-constrained scenarios such as primary care, emergency triage, and community follow-up, where the goal is primarily initial screening or risk assessment of at-risk individuals. The practicality and feasibility of modeling and validating each feature individually are limited. Therefore, three types of discriminative efficacy are derived based on single and all physiological states to quickly extract cross-state robust and reliable subsets of HRV features, which can, to some extent, suppress the interference of irrelevant or weakly correlated features in epilepsy assessment.

[0035] Furthermore, the embodiments of this application include a two-level data reduction of HRV feature set and feature items within the feature set. To further balance screening efficiency and reliability, an adaptive adjustment strategy for target feature items is introduced: when only a small number of physiological state HRV feature sets are collected, the number of target feature items is appropriately increased to cover a wider range of dimensions and make up for insufficient information; while when the collected HRV feature set is rich, the focus is on a highly relevant set of simplified feature items to avoid introducing difference noise between multiple feature sets. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. The elements or parts in the drawings are not necessarily drawn to scale. Obviously, the drawings described below are some embodiments of this application; for those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0037] Figure 1 This is a schematic diagram illustrating the performance of a single paradigm model and a full paradigm model in epilepsy diagnosis, as provided in the embodiments of this application.

[0038] Figure 2 This is a schematic diagram illustrating the ranking of the importance of various features under different physiological states, provided in an embodiment of this application.

[0039] Figure 3 This is a schematic flowchart of an epilepsy auxiliary assessment method based on heart rate variability provided in an embodiment of this application;

[0040] Figure 4 This is a schematic diagram illustrating the performance of a combinatorial paradigm model provided in this application for epilepsy diagnosis;

[0041] Figure 5 This is a schematic diagram of calibration curves under various combined paradigm models provided in the embodiments of this application;

[0042] Figure 6 This is a schematic diagram of decision curves under various combined paradigm models provided in the embodiments of this application;

[0043] Figure 7 A schematic block diagram of an epilepsy auxiliary assessment device based on heart rate variability provided in this application embodiment;

[0044] Figure 8 This is a schematic block diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0046] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0047] In this document, suffixes such as “module,” “part,” or “unit” used to denote elements are used only for illustrative purposes and have no specific meaning in themselves. Therefore, “module,” “part,” or “unit” may be used interchangeably.

[0048] In this document, the terms "upper," "lower," "inner," "outer," "front," "rear," "one end," and "the other end," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0049] In this document, unless otherwise explicitly specified and limited, the terms "installed," "equipped with," and "connected," etc., should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, a direct connection, or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0050] In this document, the term “and / or” includes any and all combinations of one or more of the listed related items.

[0051] In this article, the term "multiple" means two or more, that is, it includes two, three, four, five, etc.

[0052] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0053] As used in this specification, the term "about" typically means + / - 5% of the value, more typically + / - 4%, more typically + / - 3%, more typically + / - 2%, even more typically + / - 1%, even more typically + / - 0.5% of the value.

[0054] In this specification, certain embodiments may be disclosed in a range-bound format. It should be understood that this "range-bound" description is merely for convenience and brevity and should not be construed as a rigid limitation on the disclosed range. Therefore, the description of a range should be considered as having specifically disclosed all possible subranges and the individual numerical values ​​within that range. For example, a description of the range 1 to 6 should be considered as having specifically disclosed subranges such as from 1 to 3, from 1 to 4, from 1 to 5, from 2 to 4, from 2 to 6, from 3 to 6, etc., and the individual numbers within that range, such as 1, 2, 3, 4, 5, and 6. This rule applies regardless of the breadth of the range.

[0055] Currently, mainstream HRV testing usually needs to be performed in multiple paradigms, including resting and stimulated states. The complete procedure takes about 25–30 minutes. If the time for pre-procedure explanation, demonstration, electrode replacement, and subject adaptation is included, the actual time is often longer.

[0056] In some embodiments, the resting state refers to a physiological state in which the subject remains awake, quiet, and without external intervention, maintaining calm breathing and refraining from active physical activity, including but not limited to supine or seated resting states. The resting state test procedure lasts for 5 to 10 minutes, and the subject remains awake, breathes calmly, and avoids body movement during the test.

[0057] In some embodiments, a stimulus state refers to a physiological state in which a cardiovascular or autonomic nervous system response is actively elicited through specific physiological or physical interventions. For example, the stimulus state can be induced by any of the following methods: sustained grip force, Valsalva maneuver, or upright tilting maneuver, and the stimulus states induced by different methods have different physiological characteristics, that is, each method corresponds to a specific stimulus state.

[0058] Among them, the sustained grip force exercise can be that the subject holds the pressure sensor for 2 to 3 minutes with 30%–50% of the maximum grip force; the Valsalva maneuver can be that the subject maintains a pressure of 40 mmHg against the closed airway for 15 seconds during exhalation, and repeats 2–3 times; the upright tilting maneuver can be that the subject quickly turns from a supine position to an upright position (such as tilting at 60°–80°), which takes about 25–30 minutes.

[0059] The lengthy process is difficult to adapt to scenarios that emphasize efficiency, such as primary care, emergency triage, or community follow-up. Especially in underdeveloped areas with scarce medical resources, the high time and labor costs, complex operational requirements, and relatively low detection rate make HRV testing difficult to popularize, thus causing diagnostic delays.

[0060] Based on this, embodiments of this application provide a method, device, equipment, and medium for epilepsy auxiliary assessment based on heart rate variability. Based on the free combination of resting state and any stimulus state, it simplifies the screening process for epilepsy auxiliary assessment using heart rate variability, improves operational efficiency while ensuring assessment accuracy, and is especially suitable for scenarios with limited medical resources and high timeliness requirements, such as primary care, emergency triage, and community follow-up.

[0061] In some embodiments, the features in the HRV feature set can be flexibly configured according to actual assessment needs. For example, a set of all standard HRV parameters covering the time domain, frequency domain, and nonlinear dimensions can be extracted from the resting-state and selected-stimulation-state ECG signals, as detailed in related technologies.

[0062] In practical applications, the application value or reliability of feature terms varies. While ensuring core discriminative capabilities, such feature terms can be proactively excluded to improve evaluation accuracy and avoid introducing data noise. For example, features such as the nonlinear index DFAα1 / α2 and the frequency domain indices VLF and LF / HF ratio are unstable in short-term records or subject to controversy regarding their physiological interpretation. Removing these unreliable features actually helps reduce noise and improve the model's interpretability and predictive ability.

[0063] For example, the HRV feature set includes 13 main features, including RMSSD, SDNN, NN50, pNN50, Triangular index, AvgRR, HF, LF, TP, SD1, SD2, AC, and DC. RMSSD refers to the root mean square of the difference between adjacent normal RR intervals, used to characterize short-term heart rate variability and parasympathetic activity. SDNN refers to the standard deviation of all normal RR intervals, reflecting the overall level of heart rate variability. NN50 refers to the number of times the absolute value of the difference between adjacent normal RR intervals is greater than 50 milliseconds, used to measure the frequency of rapid heart rate fluctuations. pNN50 refers to the percentage of NN50 to the total number of normal RR intervals, used to standardize the expression of transient heart rate variability. Triangular Index refers to the triangular index calculated based on the RR interval histogram, defined as the ratio of the total number of RR intervals to the height of the highest bar in the histogram, used for geometric assessment of overall heart rate variability. AvgRR: refers to the average of all normal RR intervals, reflecting the subject's baseline heart rate cycle. HF refers to the power in the high-frequency range (0.15–0.4 Hz) of the heart rate variability power spectrum, primarily reflecting the modulation effect of the parasympathetic nervous system on the heart. LF refers to the power in the low-frequency range (0.04–0.15 Hz) of the heart rate variability power spectrum, regulated jointly by the sympathetic and parasympathetic nervous systems. TP refers to the total heart rate variability power, the sum of the power in all frequency bands within the power spectrum not exceeding 0.4 Hz, characterizing the overall autonomic nervous system's regulatory capacity. SD1 refers to the standard deviation perpendicular to the RR(n) = RR(n+1) line in the Poincaré scatter plot, reflecting instantaneous heart rate variability. SD2 refers to the standard deviation along the RR(n) = RR(n+1) line in the Poincaré scatter plot, reflecting long-term heart rate variability. AC refers to heart rate acceleration capacity, quantified by phase-rectified signal averaging techniques to represent the average change amplitude during the heart rate rise phase, characterizing the autonomic nervous system's ability to regulate heart rate acceleration. DC refers to heart rate deceleration capacity, which quantifies the average change in heart rate during the deceleration phase using phase rectified signal averaging technology, reflecting the vagus nerve-mediated cardioprotective regulatory function.

[0064] In some embodiments, the independent and global discriminant power of each pre-validated feature for epilepsy is obtained. In the experimental study, 200 patients with epilepsy (PWE Group) and 200 healthy subjects (HC Group) were included. The specific demographic and clinical characteristics of the study participants are shown in Table 1.

[0065] Table 1. Demographic and clinical characteristics of study participants

[0066]

[0067] In both the HC and PWE groups, women comprised 60.5% of the population, exhibiting a completely concordant sex distribution and showing no significant differences in any baseline demographic or anthropometric parameters, including age, weight, height, body mass index (BMI), body surface area, and waist circumference (all p>0.05). Among patients with epilepsy, focal epilepsy was the most common (62.5%), followed by generalized epilepsy (23.5%) and epilepsy of unknown type (14.0%). The median duration of epilepsy was 11.1 years (interquartile range: 4.4–17.8), and the median severity score (NHS3) was 6.0 (interquartile range: 3.0–9.0). Drug-resistant epilepsy accounted for 36.5%. Most patients (64.5%) experienced at least one seizure per month, with 52.5% experiencing seizures only during wakefulness, 9.5% only during sleep, and 38.0% experiencing seizures in both awake and sleep states.

[0068] Electrocardiogram (ECG) signals were collected from the HC group and the PWE group under different physiological conditions. For example, a standardized sequence (such as the CARTS guidelines) was used: 5 minutes of supine rest, standing, deep breathing (which could be excluded due to low compliance), Valsalva maneuver, and sustained grip strength (target 30% MVC, duration 5 minutes, with real-time feedback). Recovery time was included in each segment, and the total duration was controllable. Heart rate variability (HRV) feature sets were obtained based on the collected ECG signals. Specific statistical comparisons between the PWE and HC groups are shown in Tables 2-1 and 2-2.

[0069] Table 2-1 Comparison of HRV characteristics in supine resting position, upright tilt position, and Valsalva test

[0070]

[0071] Table 2-2 Comparison of HRV characteristics in sustained grip strength and full-range comprehensive test

[0072]

[0073] As shown in Tables 2-1 and 2-2, in all analytical paradigms, the heart rate variability of epilepsy patients was significantly reduced compared with the HC group. The key results of the 13 major features are analyzed as follows.

[0074] Regarding the Supine Resting Test (SWT), at baseline, the PWE group showed significant reductions in all 13 major features compared to the HC group. These included, but were not limited to, overall HRV (e.g., SDNN: 27.55 vs 35.30 ms), parasympathetic regulation indicators (e.g., RMSSD: 29.55 vs 35.85 ms), and indicators influenced by sympathetic activity (e.g., SD2: 33.05 vs 40.65), all p < 0.001. This pattern suggests that even at rest, there is impaired vagal regulation and an abnormal sympathetic-parasympathetic balance.

[0075] Regarding the Active Standing Test, also known as the upright tilting action test, under the upright stress caused by changes in body position, the PWE group still showed a significantly weakened autonomic nervous response. For example, the PWE group was significantly lower than the HC group in all 13 major characteristic items (p<0.001), indicating that the PWE group had impaired adaptability to postural changes.

[0076] Regarding the Valsalva Test, also known as the Valsalva Action Test, in the baroreceptor reflex challenge, the PWE group again showed significantly lower scores than the control group in all 13 major features (p < 0.01), further confirming that the PWE group had a weakened autonomic response to stress stimuli.

[0077] Regarding the Sustained Handgrip Test, which primarily induces sympathetic nerve responses, the PWE group also showed significantly lower values ​​in all 13 key characteristics compared to the control group (p < 0.01).

[0078] Regarding the whole experiment, when the entire recording period was analyzed as a single time period, the PWE group consistently showed significant reductions across all 13 major HRV parameters, further confirming the existence of a stable and persistent pattern of autonomic dysfunction.

[0079] The experimental results consistently indicate that epilepsy patients have widespread cardiac autonomic dysfunction, characterized by reduced overall heart rate variability and weakened responses to various physiological stresses.

[0080] Based on the HRV feature set of heart rate variability and demographic and anthropometric indicators, five different paradigms of logistic regression models were constructed, including: four single-paradigm models corresponding to supine resting, active standing, Valsalva maneuver, and sustained grip strength paradigms, respectively; and one full-paradigm model corresponding to the above four paradigms, learning cross-state synergistic effects. The logistic regression model is a generalized linear model used to estimate the probability that a sample belongs to a certain category (such as epilepsy patients or healthy controls), also known as logistic regression.

[0081] The 400 research participants were randomly divided into a training set (n = 280) and a test set (n = 120) in a 7:3 ratio. The logistic regression model was trained on the training set, and the model performance was evaluated on the test set. For single-paradigm models, the corresponding paradigm's HRV feature set was used for training and evaluation. For multi-paradigm models, HRV feature sets from all four paradigms were used for training and evaluation.

[0082] Based on the predicted probabilities and true labels output by each Logistic regression model, a corresponding ROC curve (Receiver Operating Characteristic Curve) is plotted. The ROC curve is a graphical tool used to evaluate the performance of binary classification models. Its vertical axis represents sensitivity (True Positive Rate, TPR), and its horizontal axis represents the false positive rate (FPR), i.e., 1 / TPR = 1 / TPR. Specificity (1) Specificity). The area under the curve (AUC) is used to quantify the overall discriminative ability of the model. The closer the AUC is to 1, the stronger the model's discriminative performance. Meanwhile, sensitivity reflects the ability to identify patients with epilepsy, and specificity reflects the ability to correctly exclude healthy individuals.

[0083] Please see Figure 1 , Figure 1 This is a schematic diagram illustrating the performance of a single paradigm model provided in this application for epilepsy diagnosis.

[0084] like Figure 1As shown in Figures A, C, E, G, and I, the AUC in the training set ranged from 0.71 to 0.74, with all single-paradigm models exhibiting moderate to good discriminative ability. Among them, the model based on the active upright test achieved the highest AUC of 0.74 (73.86%, 95% CI: 68.34%-79.30%), with an accuracy of 0.661, sensitivity of 0.743, and specificity of 0.574. The sustained grip strength test model also performed excellently, with an AUC of 0.73 (72.96%, 95% CI: 67.03%-78.45%), demonstrating a good balance between sensitivity of 0.722 and specificity of 0.566.

[0085] like Figure 1 As shown in Figures B, D, F, H, and J, the performance of all models declined in the test set. The sustained grip strength test model was the most robust, achieving the highest AUC of 0.71 (70.62%, 95% CI: 61.16%-79.92%), with a sensitivity of 0.750, specificity of 0.375, and accuracy of 0.550. The AUCs of the other four models ranged from 0.60 to 0.63, with the supine resting test model having the lowest AUC. The active standing test and Valsalva maneuver both had an AUC of 0.61, with only the overall experimental model reaching 0.63. Accuracy ranged from 0.550 to 0.575.

[0086] It should be understood that the performance difference between the training and test sets, and the relatively limited specificity in the test set, suggest that the model may be overfitting to some extent. Furthermore, relying on HRV features in a single state (such as only the resting state) for discrimination ignores the dynamic response characteristics of the autonomic nervous system under different physiological loads, making it difficult to fully capture the multidimensional interaction patterns of the sympathetic-parasympathetic nervous system and affecting the stability of feature discrimination.

[0087] Please see Figure 2 , Figure 2 This is a schematic diagram illustrating the ranking of the importance of various features under different physiological states, provided in an embodiment of this application. For example... Figure 2 As shown, the standardized coefficients of the Logistic regression model are used for evaluation. The absolute value of each feature term reflects the relative importance of each feature term in the epilepsy prediction assessment, that is, the independent discriminant power and global discriminant power of the feature term for epilepsy.

[0088] like Figure 2 As shown in Figure A, for the supine resting test, the parameters reflecting parasympathetic regulation are most important at rest. RMSSD is the most important, followed by TP and SD1. Notably, the anthropometric variable waist circumference ranks higher than several other HRV indicators.

[0089] like Figure 2 As shown in Figure B, for the active standing test, the importance of parasympathetic biomarkers is further amplified during the upright loading process involving postural changes. RMSSD remains the most dominant feature, with its importance score significantly increased, followed by SD1 and SD2, highlighting the high sensitivity of these indicators to vagal withdrawal during postural changes.

[0090] like Figure 2 As shown in the C-chart, for the Valsalva test, under this baroreceptor reflex load, SD1 became the most influential characteristic, followed by RMSSD. This indicates that short-term changes in vagal variability are a key indicator during respiratory maneuvers.

[0091] like Figure 2 As shown in the D-chart, for the sustained grip strength test, parasympathetic withdrawal markers still exhibit high discriminative ability during isometric contraction. RMSSD and SD1 remain the most important, while pNN50 and NN50 also demonstrate significant importance, ranking higher than in other paradigms.

[0092] like Figure 2 As shown in the figure, for the comprehensive experiment, after integrating all experimental paradigms, RMSSD and SD1 still overwhelmingly became the most prominent features. The importance of SDNN in the combined model has increased, suggesting its value in characterizing overall autonomic dysfunction under various physiological conditions. In contrast, demographic indicators (such as age and sex) have consistently contributed relatively little to the models.

[0093] Please see Figure 3 , Figure 3 This is a schematic flowchart illustrating an epilepsy auxiliary assessment method based on heart rate variability provided in an embodiment of this application, such as... Figure 3 As shown, this application provides an auxiliary assessment method for epilepsy based on heart rate variability. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0094] S101, acquire several target physiological states in the current acquisition combination, the target physiological states including resting state and at least one stimulus state.

[0095] In actual screening, a freely combinable data collection mechanism is provided, which can select one or two stimulus states and resting states as needed. On the one hand, it meets the basic requirements of epilepsy auxiliary assessment for multi-state HRV data, and on the other hand, it takes into account the operational feasibility in resource-constrained scenarios such as primary care, emergency triage and community follow-up.

[0096] In some embodiments, the most suitable stimulus-state evoked mode can be flexibly selected based on the subject's physical condition, on-site resource conditions, and operational timeliness, thereby improving the adaptability and execution efficiency of the screening process while ensuring data validity. For example, for elderly or frail individuals who are not suitable for upright tilting exercises, continuous grip strength exercises can be used; for community follow-ups where tilting bed equipment may not be available, portable and easy-to-use grip strength meters can be used; and in emergency triage scenarios, the Valsalva maneuver, which takes less time, is preferred.

[0097] S102, based on the target physiological state, collect the subject's resting electrocardiogram signal and at least one set of stimulated electrocardiogram signals respectively.

[0098] Specifically, resting-state electrocardiogram (ECG) signals of the subject are collected in a resting state, and at least one set of stimulated-state ECG signals of the subject are collected in one or two stimulated states.

[0099] S103, based on a preset signal processing algorithm, heart rate variability analysis is performed on each group of electrocardiogram signals to obtain the corresponding resting-state HRV feature set and at least one stimulated-state HRV feature set, wherein the HRV feature set includes multiple feature items.

[0100] The signal processing algorithm is used to extract continuous normal sinus RR interval sequences from the raw ECG signal. It can also be used to perform a series of standardized preprocessing and analysis steps to obtain numerical values ​​for various features. For example, standardized preprocessing includes, but is not limited to, denoising (e.g., bandpass filtering to remove baseline drift and high-frequency noise), R-wave detection (e.g., the Pan-Tompkins algorithm), and abnormal beat correction. For example, different types of indicators have different analysis steps: time-domain indicators are directly calculated by statistically analyzing the dispersion of the RR intervals; frequency-domain indicators use fast Fourier transform or autoregressive models to perform power spectrum analysis on the resampled RR sequence; and nonlinear indicators quantify the nonlinear characteristics of heart rate dynamics based on phase space reconstruction and complexity algorithms.

[0101] In HRV analysis, a feature term refers to a specific indicator that can quantify autonomic nervous system regulation, such as time-domain indicators (e.g., SDNN, RMSSD), frequency-domain indicators (e.g., LF, HF), or nonlinear indicators (e.g., SD1). Each feature term reflects different dimensions of physiological regulatory characteristics. For example, the HRV feature set may include only the 13 main features, or it may include other features besides the main features.

[0102] Specifically, after obtaining the ECG signals in the resting state and the selected stimulus state, the signal processing algorithm is used to perform HRV analysis on the two signals respectively. For example, in the acquisition combination of the resting state being supine rest and the stimulus state being continuous grip force action, the resting HRV feature set in the resting state is output, and the stimulus HRV feature set in the continuous grip force state is output at the same time. The feature terms are divided in the same way, but the values ​​are different due to the difference in physiological state.

[0103] S104, based on the independent discriminant efficacy of each feature in each target physiological state for epilepsy, as well as the associated discriminant efficacy between at least two target physiological states and / or the global discriminant efficacy in all physiological states, determine a set of target features applicable to the current acquisition combination.

[0104] Independent discriminant efficacy refers to the ability of a feature to distinguish between epileptic patients and non-epileptic subjects under a single specific physiological state; associated discriminant efficacy refers to the ability of a feature to jointly distinguish between epileptic patients and non-epileptic subjects among multiple target physiological states included in the current acquisition combination; global discriminant efficacy refers to the overall ability of a feature to distinguish between epileptic patients and non-epileptic subjects in a comprehensive test covering all standard physiological states (including resting, grip strength, Valsalva maneuver, upright tilt, etc.).

[0105] In some embodiments, the independent discriminant efficacy of each HRV feature for epilepsy under each physiological state and the global discriminant efficacy in a comprehensive trial are obtained. For example, the discriminant efficacy can be calibrated and updated based on clinical data from different population groups (such as children with epilepsy, adults with refractory epilepsy, or specific etiological subtypes) to form a efficacy database that can be finely adjusted for different epidemiological or clinical contexts.

[0106] like Figure 2 As shown, the independent discriminant efficacy values ​​and relative rankings of each HRV feature under each physiological state (including resting state, sustained grip force action, Valsalva maneuver, and upright tilting action) are recorded, as well as the global discriminant efficacy values ​​and relative rankings of each HRV feature under the full paradigm.

[0107] It is evident that the discriminative efficacy of different features varies under different physiological states. For example, TP exhibits high independent discriminative efficacy in resting state and sustained grip strength movements, while in the Valsalva maneuver, TP's independent discriminative efficacy drops to the lowest among all features, indicating the least importance. In the comprehensive test, TP's global discriminative efficacy is moderate. It should be understood that the discriminative value of features is highly dependent on the physiological state. During the selection of target features, dynamic evaluation should be conducted by combining the specific collection combinations and their corresponding discriminative efficacy to avoid mechanically applying fixed feature sets.

[0108] In some embodiments, first-level features whose independent discrimination efficacy is higher than or equal to a first preset threshold under different target physiological states are selected as target features, so that the target features have basic discrimination ability under each state.

[0109] In some embodiments, the associated discriminant power of each of the two types of features is calculated based on the independent discriminant power of each feature under the target physiological state for epilepsy. For example, the associated discriminant power is obtained by taking the arithmetic mean of the independent discriminant powers under multiple target physiological states.

[0110] In some embodiments, primary feature items whose independent discrimination efficacy is higher than or equal to a first preset threshold under different target physiological states are selected; secondary feature items whose correlation discrimination efficacy is higher than or equal to a second preset threshold are selected from the primary feature items as target feature items. This ensures that the target feature items not only possess basic discrimination ability under various states, but also exhibit stable overall performance and high comprehensive efficacy.

[0111] For example, in the combination of resting supine position and sustained grip strength action, such as Figure 2 The feature terms shown show that the SDNN has different discrimination performance in the two physiological states. The independent discrimination performance under continuous grip force is less than the first preset threshold. It is filtered at the independent discrimination performance screening node to avoid entering the subsequent correlation discrimination performance screening node. It is retained because the average value is higher than or equal to the second preset threshold.

[0112] In some embodiments, first-level features with independent discrimination efficacy higher than or equal to a first preset threshold under different target physiological states are selected; if the independent discrimination efficacy of a feature is higher than the first preset threshold under some physiological states, but lower than the first preset threshold under other physiological states, for suspicious feature items that show such differences, third-level features with global discrimination efficacy higher than or equal to a third preset threshold are selected; the first-level features and third-level features are used as target features.

[0113] For example, in the combination of resting supine position and Valsalva maneuver, such as Figure 2 The feature term TP, as shown, exhibits extreme differences in discriminative efficacy between the two physiological states, thus it is a questionable feature term; in the combination of data collection during resting supine position and sustained grip force, such as Figure 2 The feature term SDNN shows a difference in discriminative efficacy under the two physiological states, making it a questionable feature term. These two feature terms are filtered at the independent discriminative efficacy screening node. To avoid the loss of key information due to over-filtering, the questionable feature term enters the global discriminative efficacy screening node. At this point, feature term TP is filtered because its global discriminative efficacy is less than the third preset threshold, while feature term SDNN is retained as a level 3 feature term because its global discriminative efficacy is higher than the third preset threshold.

[0114] It should be understood that global discriminative power assessment based on feature terms can assess the possibility of introducing noise. For example, feature terms with significantly low global discriminative power may be considered to have poor stability and high noise risk even if they perform well in individual states. Therefore, they should be excluded when determining conflicts.

[0115] In some embodiments, a minimum number of target features is set, for example, no less than 10.

[0116] In some embodiments, secondary and tertiary feature items are used as target feature items. When the number of target feature items does not meet the minimum requirement, primary feature items are used as target feature items, or the first preset threshold and / or the second preset threshold are lowered to preset values, and selection is performed again until the number of target feature items meets the minimum requirement. This avoids the loss of key physiological information due to over-filtering and effectively suppresses the interference of low-quality and unstable features on the model.

[0117] In some implementations, the first preset threshold is a benchmark value used to determine whether a feature has sufficient independent discriminative power under a specific physiological state, and may be a lower limit threshold for determining obvious ineffectiveness; the second preset threshold is a benchmark value used to determine whether a feature has sufficient correlation discriminative power among multiple target physiological states; and the third preset threshold is a benchmark value used to determine whether a feature has sufficient global discriminative power in the entire comprehensive test.

[0118] Furthermore, the specific values ​​of the first, second, and third preset thresholds can be dynamically set according to the actual distribution of HRV feature discrimination efficacy under different physiological states. Because different physiological states exhibit varying sensitivities to epilepsy and different autonomic nervous system response patterns, the overall level and dispersion of feature discrimination efficacy also differ under different physiological states. For example, ... Figure 2 As shown, under the Valsalva maneuver, the top-ranked feature, SD1, has a discriminative efficiency of only 4, while under the upright-tilting maneuver, the top-ranked feature, RMSSD, has a discriminative efficiency of 10.0. Therefore, based on the rationality and adaptability of the screening rules, the first preset threshold can be configured separately for each physiological state, and the second and third preset thresholds can be set according to the physiological state to be judged.

[0119] For example, by analyzing the central tendency in the efficacy database, such as calculating the median or average, we can use it as a benchmark indicator for judging the overall efficacy level. Based on this benchmark indicator, we can determine whether the efficacy in each physiological state is generally too high or too low, and apply a relatively strict threshold or a relatively relaxed threshold accordingly.

[0120] It should be noted that in practical applications, the combination and form of HRV data collection are flexible and varied, and the types of HRV features are also diverse. Modeling and validating the discriminative value of each feature item for each application scenario is prohibitively expensive, especially in resource-constrained scenarios such as primary care, emergency triage, and community follow-up, where the goal is primarily initial screening or risk assessment of at-risk individuals. The practicality and feasibility of modeling and validating each feature individually are limited. Therefore, three types of discriminative efficacy are derived based on single and all physiological states to quickly extract cross-state robust and reliable subsets of HRV features, which can, to some extent, suppress the interference of irrelevant or weakly correlated features in epilepsy assessment.

[0121] like Figure 2 As shown, when using the combination of supine resting and upright tilting data acquisition, the NN50's independent discrimination performance in upright tilting is less than the first preset threshold, and the AC's independent discrimination performance in supine resting is less than the first preset threshold, so they are filtered out. Based on the correlation discrimination performance and global discrimination performance, the target feature terms can be determined to include RMSSD, TP, SD1, LF, SD2, SDNN, DC, HF, AvgRR, and Triangular.index. Similarly, when using the combination of supine resting and continuous grip force action data acquisition, the target feature terms can be determined to include RMSSD, SD1, TP, LF, HF, SD2, PNN50, NN50, Triangular.index, and AvgRR.

[0122] S105, Based on the target feature items in the resting-state HRV feature set and the stimulated-state HRV feature set, generate a screening report for assisting in epilepsy risk assessment.

[0123] In some embodiments, specific parameter values ​​of the target feature are extracted from the resting-state HRV feature set and the stimulated-state HRV feature set, and a screening report for assisting epilepsy risk assessment is generated based on the specific parameter values ​​of the target feature.

[0124] Specifically, after determining a set of target features applicable to the current acquisition combination, specific parameter values ​​matching the names of these target features are extracted one by one from the previously generated resting-state HRV feature set and the corresponding stimulus-state HRV feature set. For example, the millisecond value of a subject's RMSSD in the resting state, or the specific value of HF power under continuous grip force.

[0125] After obtaining the specific parameter values ​​corresponding to the target feature, these values ​​are input into a pre-trained epilepsy risk prediction model. The model outputs a risk score, and the epilepsy risk level is determined based on a preset risk threshold and the risk score, generating a screening report. The screening report may include the epilepsy risk level (e.g., low, medium, high), the risk score value, and a brief explanatory description for clinical decision support.

[0126] For example, after collecting electrocardiogram signals from subjects during a community follow-up study while they were supine at rest and under sustained grip strength, the specific parameter values ​​of the target feature items were extracted and input into an epilepsy risk prediction model. The output risk score was 0.82 (if the high-risk threshold is 0.7), and the screening report indicated "high risk of epilepsy, further examination recommended".

[0127] In some embodiments, the target features include a first type of feature and a second type of feature; the first type of feature is a mandatory feature under each physiological state, and the second type of feature is dynamically adjusted based on the acquisition combination.

[0128] In some embodiments, the type of feature includes RMSSD and SD1.

[0129] In some embodiments, a class of features can be determined based on the global discriminative efficacy determined in a comprehensive test. For example, features with high global discriminative efficacy, high ranking, and stable cross-state performance in the comprehensive test (e.g., RMSSD, SDNN, etc.) are presumed to have fundamental and universal discriminative value for epilepsy and are therefore classified as a class of features and fixedly included in the target feature set to minimize the loss of core discriminative information. For example, a class of features consists of features with a global discriminative efficacy greater than or equal to a fourth preset threshold, where the fourth preset threshold is greater than or equal to a third preset threshold.

[0130] In some embodiments, the selection of the two types of features can be performed using the target feature selection mechanism described in the foregoing embodiments.

[0131] In some embodiments, S104 includes: determining the number of second-class feature items to be selected based on the number of HRV feature sets; calculating the association discriminant efficacy of each of the two types of feature items based on the independent discriminant efficacy of each feature item for epilepsy under the target physiological state; and sorting the feature items from high to low and selecting the corresponding number of feature items as target feature items according to the number of second-class feature items selected and the association discriminant efficacy.

[0132] Specifically, the number of allowed selections of the second type of feature items is determined based on the actual number of HRV feature sets collected. When only a small number of HRV feature sets of physiological states are collected (such as two feature sets containing only the resting state and one stimulus state), the number of target feature items is appropriately increased to cover a wider range of dimensions and make up for insufficient information. When the collected HRV feature sets are rich, the focus is on a simplified feature set with high relevance to avoid introducing difference noise between multiple feature sets.

[0133] For each candidate binary feature item, the independent discrimination performance is integrated (e.g., by taking the arithmetic mean) to obtain the association discrimination performance. These values ​​are then sorted from highest to lowest, and finally, no more than the selected number of feature items are chosen as the target feature items. It should be noted that the selected number can correspond to the minimum requirement in the aforementioned embodiments, or it can be the maximum requirement, depending on the specific application scenario.

[0134] In some embodiments, an electrocardiogram (ECG) signal acquisition device refers to a hardware device used to record the electrical activity of a subject's body surface. The ECG signal acquisition device includes at least one of the following: a wearable device with ECG measurement function (such as a smart bracelet or a chest strap ECG patch), a single-lead ECG signal acquisition device (such as a portable handheld device), or a multi-lead ECG signal acquisition device (such as a clinical 12-lead device).

[0135] In practice, the acquisition equipment can be flexibly selected according to the application scenario: in grassroots or community follow-up, a lightweight and easy-to-operate single-lead wearable device can be used to complete continuous electrocardiogram recording; in hospital or research environments, multi-lead devices can be used to improve signal stability and R-wave detection accuracy.

[0136] In some embodiments, a wireless ECG patch is used, which can be used in conjunction with a respiratory flow sensor and a digital hand grip meter.

[0137] In some embodiments, the ECG sampling rate is ≥250Hz.

[0138] In some embodiments, the acquired signals, after uniform preprocessing, are used to extract HRV feature sets, which are then input into the epilepsy risk prediction model. For example, Pan-Tompkins is used to detect R waves, with adaptive artifact removal; RR sequences are resampled at 4Hz and detrended; and high-quality RR interval sequences are constructed.

[0139] Please see Figure 4 , Figure 4 This is a schematic diagram illustrating the performance of a combinatorial paradigm model provided in this application for epilepsy diagnosis.

[0140] The resting state was combined with any one of the three stimulus states, and logistic regression models were constructed accordingly. The 400 participants were randomly divided into a training set (n = 280) and a test set (n = 120) in a 7:3 ratio. The logistic regression model was trained on the training set using the corresponding paradigm's HRV feature set, and the model performance was evaluated on the test set. It should be noted that all features in the HRV feature set were used for training and evaluation without prior screening.

[0141] like Figure 4 As shown in Figures A, C, and E, within the training set, all three combined paradigm models exhibit similar and moderate to good diagnostic capabilities, with AUCs ranging from 0.74 to 0.76.

[0142] The key differences were found in the independent test set used to evaluate the generalization ability of the models, as shown in Figures B, D, and F in Figure 4. The first paradigm model combining supine rest and active standing had an AUC of 0.61, the second paradigm model combining supine rest and Valsalva maneuver had an AUC of 0.62, and the third paradigm model combining supine rest and sustained grip strength showed better and more robust performance, with an AUC of 0.68 (95% confidence interval: 0.59–0.77), and a sensitivity of 0.821 (95% confidence interval: 0.76–0.92) at a specificity of 0.484.

[0143] like Figure 2 In the single-paradigm model and the full-paradigm model, the combined paradigm models of the first and second groups showed a slight improvement in AUC compared to the corresponding single-paradigm models, and gradually approached the performance of the full-paradigm model, while maintaining a good balance between specificity and sensitivity; although the AUC of the third group was slightly lower than that of the single-paradigm model in the continuous grip strength test, it still remained higher than that of the full-paradigm model, and its specificity and sensitivity were both improved.

[0144] It should be understood that a full-paradigm model requires comprehensive collection of HRV data across multiple paradigms, providing more reference information for subsequent referrals, assessments, or interventions by healthcare professionals. However, it demands high levels of patient cooperation, time commitment, and adherence to clinical procedures, making implementation difficult. Furthermore, it may introduce redundant or even conflicting information, limiting model performance; for example, its AUC performance is not significantly superior to single-paradigm or multi-paradigm combined models. Conversely, a single-paradigm model is limited by reflecting only the heart rate variability characteristics under a specific physiological state, providing only a single dimension of information. It struggles to comprehensively reflect the complex pathological mechanisms of epilepsy, a neurological disease characterized by high heterogeneity and dynamic changes, and cannot adapt to the performance differences caused by fluctuations in emotion, environment, or physical condition in real-world clinical scenarios, thus limiting its reliability.

[0145] By integrating HRV data from resting state and different stimulus states, the multi-paradigm combination provides different levels of physiological information from the perspectives of basic autonomic nervous function and stress response. The combination of the two can better reflect the dynamic changes of the autonomic nervous system. To a certain extent, it enhances the ability to identify patients with different clinical phenotypes with less data, ensuring the clinical applicability of diagnosis and helping to provide more reliable auxiliary diagnosis. It also shows superior comprehensive performance in scenarios where medical resources are limited and timeliness is critical.

[0146] Preferably, the data collection combination is resting state and continuous grip force action. Combining continuous grip force with baseline resting state is the preferred method for identifying epilepsy patients and can effectively improve model sensitivity.

[0147] To further evaluate the performance of the combinatorial paradigm model, calibration plots and decision curve analysis (DCA) were used to assess calibration accuracy and clinical applicability. Please refer to the relevant technical documentation for details. Figures 5 to 6 , Figure 5 This is a schematic diagram of calibration curves under various combined paradigm models provided in the embodiments of this application. Figure 6 This is a schematic diagram of decision curves under various combined paradigm models provided in the embodiments of this application.

[0148] Calibration plots were generated on both the training and test sets to compare the consistency between the model's predicted probabilities and the actual outcomes. Figure 5 As shown in Figure C, in the test set, the combined paradigm model of supine resting and sustained grip strength showed the best calibration, and its bias correction curve closely matched the ideal 45° reference line, indicating high diagnostic reliability. Figure 5 In the A and B models, the other two models only showed mild over-prediction. Their test set curves were slightly below the ideal line overall, but they still had a certain degree of diagnostic reliability.

[0149] Furthermore, decision curve analysis was performed to evaluate the clinical net benefit of each model within different threshold probability ranges. For example... Figure 6 As shown in Figures B, D, and F, in the key test set analysis, the combined paradigm of supine resting and sustained grip strength consistently demonstrated the highest net benefit across a broad and clinically meaningful threshold range of approximately 0.20 to 0.75. Furthermore, as... Figure 6 Similar advantages were observed in categories A, C, and E in the training set. This suggests that using this model to guide clinical decision-making yields better outcomes compared to strategies that assess all patients or none at all.

[0150] To further improve the performance of the multi-paradigm combination model, in practical applications, target features for risk assessment can be dynamically selected based on the actual collected physiological state combinations and the differences in the three-level discriminative efficiency of different HRV feature items. A multi-paradigm model can then be built based on these target features. This approach can effectively suppress the interference of irrelevant or weakly correlated features on epilepsy assessment while minimizing the loss of key information, thereby improving the accuracy of screening.

[0151] In this embodiment, target features for risk assessment are dynamically selected based on the actual collected physiological state combinations, rather than simply using all HRV features. This adapts to the differences in the discriminative value of different HRV features for epilepsy under different physiological states, reducing potential noise and thus further improving the accuracy of epilepsy assessment. For example, a two-layer feature selection mechanism is designed: one type of feature ensures that core discriminative features are always retained; the other type of feature is dynamically adjusted based on the correlation discriminative efficacy under the current collection conditions. This effectively suppresses the interference of irrelevant or weakly correlated features on epilepsy assessment while minimizing the loss of key information, thereby improving the accuracy of screening.

[0152] In some embodiments, the method further includes: obtaining clinical diagnostic reports of epileptic subjects and non-epileptic subjects, and HRV feature sets in the resting state and at least one stimulus state, as training data; training a machine learning model based on the training data to obtain an epilepsy risk prediction model; inputting the specific parameter values ​​of the target feature items into the epilepsy risk prediction model to output a risk score for generating a screening report to assist in epilepsy risk assessment.

[0153] Specifically, HRV feature sets and clinical diagnostic reports (confirmed epilepsy or non-epilepsy) of clinically diagnosed epilepsy patients and healthy controls are collected, preprocessed to form structured training samples, and trained using highly interpretable algorithms such as logistic regression. A dedicated paradigm model (such as a model that is only applicable to the two combined paradigms of supine rest and continuous grip strength) or a general full paradigm model can be constructed.

[0154] It should be understood that the machine learning model used can be constructed using a comprehensive dataset covering a variety of physiological state combinations (including the complete four states) during the training phase. It has strong generalization ability and can be flexibly adapted to various subsequent actual collection conditions, enabling one-time training and reuse in multiple scenarios. This improves the efficiency and consistency of epilepsy screening at the grassroots level and realizes the rapid, standardized and intelligent assessment of epilepsy risk.

[0155] In some embodiments, training data can be preprocessed based on target features of different acquisition combinations to pre-screen training data that do not contain target features.

[0156] This application also provides a method for epilepsy prediction based on HRV, characterized in that the method includes: collecting resting-state electrocardiogram (ECG) signals of a subject in a resting state, and collecting at least one set of stimulated-state ECG signals of the subject under one or two stimulation states; performing heart rate variability analysis on the resting-state ECG signals and the stimulated-state ECG signals respectively based on a preset signal processing algorithm to obtain corresponding resting-state HRV feature sets and stimulated-state HRV feature sets, wherein the HRV feature sets include multiple feature items; determining a set of target feature items suitable for the current collection combination based on the independent discriminant efficacy of each feature item for epilepsy under each physiological state; extracting specific parameter values ​​of the target feature items from the resting-state HRV feature sets and the stimulated-state HRV feature sets; and generating a screening report for assisting epilepsy risk assessment based on the specific parameter values. Specific implementation methods of this method can be found in the foregoing embodiments and will not be repeated here.

[0157] This application also provides a multi-paradigm HRV acquisition method for epilepsy risk prediction. The method includes: acquiring at least two target physiological states in the current acquisition combination, the target physiological states including a resting state and one or two stimulation states; acquiring the subject's resting-state electrocardiogram (ECG) signal and at least one set of stimulation-state ECG signals based on the target physiological states; performing heart rate variability analysis on each set of ECG signals based on a preset signal processing algorithm to obtain corresponding resting-state HRV feature sets and stimulation-state HRV feature sets, the HRV feature sets including multiple feature items; and generating a screening report for assisting epilepsy risk assessment based on the resting-state HRV feature sets and the stimulation-state HRV feature sets. Specific implementation details of this method can be found in the foregoing embodiments and will not be repeated here.

[0158] This application also provides a method for predicting epilepsy risk using heart rate variability (HRV), characterized in that the method includes: collecting resting-state electrocardiogram (ECG) signals of a subject in a supine position and collecting stimulated-state ECG signals of the subject in a continuously held position; performing heart rate variability analysis on each set of ECG signals based on a preset signal processing algorithm to obtain corresponding resting-state HRV feature sets and stimulated-state HRV feature sets, wherein the HRV feature sets include multiple feature items; and generating a screening report for assisting epilepsy risk assessment based on the resting-state HRV feature sets and the stimulated-state HRV feature sets.

[0159] The resting supine state and the continuous holding state correspond to the specific implementation methods of the resting state and the stimulated state, respectively. The specific implementation method of the HRV application method for epilepsy risk prediction can be referred to the foregoing embodiments, and will not be repeated here.

[0160] Please see Figure 7 , Figure 7This is a schematic block diagram of an epilepsy auxiliary assessment device based on heart rate variability provided in an embodiment of this application. The epilepsy auxiliary assessment device based on heart rate variability can be configured in a server to execute the aforementioned epilepsy auxiliary assessment method based on heart rate variability.

[0161] like Figure 7 As shown in the illustration, this application also provides an epilepsy auxiliary assessment device 200 based on heart rate variability, the device comprising:

[0162] Physiological state acquisition module 201 is used to acquire at least two target physiological states in the current acquisition combination, wherein the target physiological states include a resting state and one or two stimulus states;

[0163] The electrocardiogram (ECG) signal acquisition module 202 is used to acquire the subject's resting ECG signal and at least one set of stimulated ECG signals based on the target physiological state.

[0164] The heart rate variability analysis module 203 is used to perform heart rate variability analysis on each group of electrocardiogram signals based on a preset signal processing algorithm, and obtain the corresponding resting state HRV feature set and at least one stimulated state HRV feature set, wherein the HRV feature set includes multiple feature items.

[0165] The target feature determination module 204 is used to determine a set of target features applicable to the current acquisition combination based on the independent discriminant efficacy of each feature in each target physiological state, the associated discriminant efficacy between at least two target physiological states and / or the global discriminant efficacy in all physiological states.

[0166] The screening report generation module 205 is used to generate a screening report for assisting epilepsy risk assessment based on the target feature items in the resting-state HRV feature set and the stimulated-state HRV feature set.

[0167] For example, the stimulus state is induced by any of the following: sustained gripping action, Valsalva maneuver, or upright tilting action.

[0168] For example, the target features include a first type of feature and a second type of feature; the first type of feature is determined based on global discrimination performance, and the second type of feature is dynamically adjusted based on the acquisition combination.

[0169] For example, the feature items in the HRV feature set include RMSSD, SDNN, NN50, pNN50, Triangular.index, AvgRR, HF, LF, TP, SD1, SD2, AC, and DC; the first type of feature items includes RMSSD and SD1.

[0170] For example, the target feature determination module 204 further includes: a two-category feature item submodule, an association discrimination efficiency submodule, and a feature item selection submodule.

[0171] The second-class feature item submodule is used to determine the number of second-class feature items to be selected based on the number of HRV feature sets.

[0172] The association discriminant efficacy submodule is used to calculate the association discriminant efficacy of each of the two types of feature items based on the independent discriminant efficacy of each feature item for epilepsy under the target physiological state.

[0173] The feature selection submodule is used to sort and select a corresponding number of feature items as target feature items based on the number of selected second-type feature items and the association discrimination performance, from high to low.

[0174] For example, the epilepsy auxiliary assessment device 200 based on heart rate variability also includes: a training data submodule, a model training submodule, and a risk scoring submodule.

[0175] The training data submodule is used to obtain clinical diagnostic reports of epileptic and non-epileptic subjects, as well as HRV feature sets in the resting state and at least one stimulus state, as training data.

[0176] The model training submodule is used to train a machine learning model based on the training data to obtain an epilepsy risk prediction model.

[0177] The risk scoring submodule is used to input the specific parameter values ​​of the target feature items into the epilepsy risk prediction model to output a risk score, which is used to generate a screening report to assist in epilepsy risk assessment.

[0178] For example, the electrocardiogram (ECG) signal acquisition device includes at least one of the following: a wearable device with ECG measurement function, a single-lead or multi-lead ECG signal acquisition device.

[0179] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the above-described apparatus and its modules and units can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0180] The methods and apparatus of this application can be used in a wide variety of general-purpose or special-purpose computing system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer terminal devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc.

[0181] Please see Figure 8 , Figure 8 This is a schematic block diagram illustrating the structure of a computer device according to an embodiment of this application. The computer device may be a terminal device or a server.

[0182] For example, the above-described method and apparatus can be implemented as a computer program, which can be used in, for example... Figure 8 It runs on the computer device shown.

[0183] like Figure 8 As shown, the computer device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include non-volatile storage media and internal memory.

[0184] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions that, when executed, cause a processor to perform the steps of the methods provided in any of the foregoing embodiments.

[0185] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0186] The internal memory provides an environment for the execution of a computer program in a non-volatile storage medium, which, when executed by a processor, causes the processor to perform the steps of the methods provided in any of the foregoing embodiments.

[0187] This network interface is used for network communication, such as sending assigned tasks.

[0188] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0189] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps:

[0190] S101, acquire at least two target physiological states in the current acquisition combination, the target physiological states including resting state and one or two stimulus states;

[0191] S102, based on the target physiological state, collect the subject's resting electrocardiogram signal and at least one set of stimulated electrocardiogram signals respectively;

[0192] S103, based on a preset signal processing algorithm, heart rate variability analysis is performed on each group of electrocardiogram signals to obtain the corresponding resting state HRV feature set and at least one stimulated state HRV feature set, wherein the HRV feature set includes multiple feature items;

[0193] S104, Based on the independent discriminant efficacy of each feature item for epilepsy under each target physiological state, as well as the associated discriminant efficacy between at least two target physiological states and / or the global discriminant efficacy under all physiological states, determine a set of target features applicable to the current acquisition combination.

[0194] S105, Based on the target feature items in the resting-state HRV feature set and the stimulated-state HRV feature set, generate a screening report for assisting in epilepsy risk assessment.

[0195] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps:

[0196] Acquire at least two target physiological states from the current acquisition combination, wherein the target physiological states include a resting state and one or two stimulus states;

[0197] Based on the target physiological state, the subject's resting electrocardiogram (ECG) signal and at least one set of stimulated ECG signals were collected respectively;

[0198] Based on a preset signal processing algorithm, heart rate variability analysis is performed on each group of electrocardiogram signals to obtain the corresponding resting-state HRV feature set and the stimulated-state HRV feature set. The HRV feature set includes multiple feature items.

[0199] Based on the resting-state HRV feature set and the stimulated-state HRV feature set, a screening report is generated to assist in epilepsy risk assessment.

[0200] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps:

[0201] Resting-state electrocardiogram (ECG) signals were collected from the subjects while they were in a supine position at rest, and stimulated-state ECG signals were collected from the subjects while they were in a continuous holding position.

[0202] Based on a preset signal processing algorithm, heart rate variability analysis is performed on each group of electrocardiogram signals to obtain the corresponding resting-state HRV feature set and the stimulated-state HRV feature set. The HRV feature set includes multiple feature items.

[0203] Based on the resting-state HRV feature set and the stimulated-state HRV feature set, a screening report is generated to assist in epilepsy risk assessment.

[0204] For example, the processor is used to run a computer program stored in a memory, and is also used to implement the steps of the method provided in any embodiment of this application, which will not be repeated here.

[0205] This application also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, and the processor executing the program instructions to implement the steps of the method provided in any embodiment of this application.

[0206] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0207] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A heart rate variability-based epilepsy assessment device, characterized in that, The device includes: The physiological state acquisition module is used to acquire at least two target physiological states in the current acquisition combination. The target physiological states include a resting state and one or two stimulus states. The stimulus states are induced by any one of the following: sustained grip force action, Valsalva maneuver, and upright tilting action. An electrocardiogram (ECG) signal acquisition module is used to acquire the subject's resting ECG signal and at least one set of stimulated ECG signals based on the target physiological state. The heart rate variability analysis module is used to perform heart rate variability analysis on each group of electrocardiogram signals based on a preset signal processing algorithm, and obtain the corresponding resting state HRV feature set and at least one stimulated state HRV feature set, wherein the HRV feature set includes multiple feature items. The target feature determination module is used to determine a set of target features applicable to the current acquisition combination based on the independent discriminant efficacy of each feature in each target physiological state, the associated discriminant efficacy between at least two target physiological states and / or the global discriminant efficacy in all physiological states. The screening report generation module is used to generate a screening report for assisting epilepsy risk assessment based on the target feature items in the resting-state HRV feature set and the stimulated-state HRV feature set. The target feature items include a first type of feature items and a second type of feature items; the first type of feature items are determined based on global discrimination performance, and the second type of feature items are dynamically adjusted based on the acquisition combination. The target feature determination module further includes: The second-class feature item submodule is used to determine the number of second-class feature items to be selected based on the number of HRV feature sets. The association discriminant efficacy submodule is used to calculate the association discriminant efficacy of each of the two types of feature items based on the independent discriminant efficacy of each feature item for epilepsy under the target physiological state. The feature selection submodule is used to sort and select a corresponding number of feature items as target feature items based on the number of selected second-type feature items and the association discrimination performance, from high to low.

2. The apparatus according to claim 1, characterized in that, The features in the HRV feature set include RMSSD, SDNN, NN50, pNN50, Triangular.index, AvgRR, HF, LF, TP, SD1, SD2, AC, and DC; the first type of features includes RMSSD and SD1.

3. The apparatus according to claim 1, characterized in that, The device further includes: The training data submodule is used to obtain clinical diagnostic reports of epileptic and non-epileptic subjects, as well as HRV feature sets in the resting state and at least one stimulus state, as training data. The model training submodule is used to train a machine learning model based on the training data to obtain an epilepsy risk prediction model. The risk scoring submodule is used to input the specific parameter values ​​of the target feature items into the epilepsy risk prediction model to output a risk score, which is used to generate a screening report to assist in epilepsy risk assessment.

4. The apparatus according to claim 1, characterized in that, The electrocardiogram (ECG) signal acquisition device includes at least one of the following: a wearable device with ECG measurement function, a single-lead or multi-lead ECG signal acquisition device.

5. A computer device, characterized in that, The device includes: Memory, used to store computer programs; A processor is configured to execute the computer program and, in executing the computer program, implement the module functions of the epilepsy auxiliary assessment device based on heart rate variability as described in any one of claims 1 to 4.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to perform the module functions of the epilepsy auxiliary assessment device based on heart rate variability as described in any one of claims 1 to 4.

Images