Auxiliary epilepsy diagnosis and treatment system based on intelligent wearable device

Abstract

The invention relates to the technical field of intelligent wearable equipment, in particular to an epilepsy auxiliary diagnosis and treatment system based on intelligent wearable equipment, which comprises a data acquisition module, an atlas construction module, a feature extraction module, a state judgment module and a strategy execution module, and is characterized in that the atlas construction module is used for constructing a personalized behavior physiological atlas based on a test data set; the feature extraction module is used for extracting a real-time feature vector of a current time window based on the monitoring data flow; the state judgment module is used for generating a state judgment result about the current behavior intention of the user; a personalized physiological reference map of a user is established by executing an active behavior test, a deliberate inhibition behavior adopted by the user due to perception of a thought is intelligently identified, and monitoring data corresponding to the behavior is isolated, so that the core problem that the performance of an epilepsy auxiliary diagnosis and treatment model is degraded due to reverse modeling of the user behavior in long-term use is effectively solved; and the response reliability of the monitoring system in a real attack scene is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent wearable devices, and particularly relates to an epilepsy auxiliary diagnosis and treatment system based on an intelligent wearable device. BACKGROUND

[0002] Epilepsy is a common chronic neurological disease, and patients need long-term management. An auxiliary diagnosis and treatment system based on an intelligent wearable device, such as a smart watch, a bracelet, a special patch and the like, continuously monitors various physiological and behavioral signals such as electrocardiogram, electromyogram, movement, skin electricity and the like, and uses an algorithm model to realize automatic detection, early warning and log recording of an epileptic seizure, greatly improving the safety and quality of life of patients and providing an objective diagnosis and treatment basis for doctors.

[0003] However, as users wear and rely on the device for a long time, especially for patients with consciousness-retained focal seizures, a potential and insufficiently emphasized technical problem gradually emerges: user behavior counter-molding and feature pollution. Specifically, when a user perceives possible pre-seizure symptoms (aura), the user may unconsciously or based on experience take counteractive behavior strategies, such as deliberately keeping the limbs still, controlling the breathing rhythm or performing specific relaxation movements, trying to suppress the occurrence of seizures or avoid false alarms of the device. These deliberate adaptive behaviors change the form of physiological signals. SUMMARY

[0004] To solve the technical problems in the background art, the present application proposes an epilepsy auxiliary diagnosis and treatment system based on an intelligent wearable device, and the specific scheme is as follows: An epilepsy auxiliary diagnosis and treatment system based on an intelligent wearable device, comprising: a data acquisition module: by executing a preset active behavior test protocol, the active behavior test protocol contains at least one standardized behavior task, acquiring multi-modal physiological signals of a user when performing at least one standardized behavior task, and generating a test data set; a graph construction module: based on the test data set, an individualized behavior physiological graph is constructed; the individualized behavior physiological graph at least contains a baseline physiological feature set for representing a natural relaxation state of a user; a feature extraction module: in the daily monitoring process of the intelligent wearable device, a monitoring data stream is obtained in real time, and a real-time feature vector of a current time window is extracted based on the monitoring data stream; a state determination module: the real-time feature vector is matched and analyzed with the baseline physiological feature set in the individualized behavior physiological graph, and a state determination result about the current behavior intention of the user is generated; The policy execution module executes a preset data management policy on the monitoring data of the current time window based on the state determination result; when the state determination result is the active behavior inhibition state, the data management policy comprises marking and isolating the corresponding monitoring data.

[0005] Further, in the data acquisition module, the active behavior test protocol comprises the following standardized behavior tasks performed in sequence: The first task is to guide the user to maintain natural relaxation; The second task is to guide the user to deliberately suppress the movement of the wearing limb; The third task is to guide the user to perform regular deep breathing.

[0006] Further, in the atlas construction module, a personalized behavior-physiology atlas is constructed: Based on the signals collected during the execution of the first task, a set of baseline physiological features is calculated; Based on the signals collected during the execution of the second task, a set of deliberate suppression features is calculated, wherein the set of deliberate suppression features includes a first feature index extracted from heart rate variability signals, and the first feature index is used to quantify the degree of autonomic nervous effort; Based on the signals collected during the execution of the third task, a set of deliberate breathing features is calculated, and the set of deliberate breathing features includes a second feature index used to quantify the regularity of the breathing signal.

[0007] Further, the first feature index is the ratio of low-frequency power to high-frequency power in frequency domain analysis of heart rate variability, and its calculation formula is: , wherein, is the signal power in the preset low-frequency range, is the signal power in the preset high-frequency range.

[0008] Further, the second feature index used to quantify the regularity of the breathing signal is the sample entropy of the respiratory interval sequence, and its calculation method is: for a respiratory interval sequence RR={rr1,rr2,...,rrN} with length N, set the pattern length as m and the tolerance as r, and the calculation of the sample entropy SampEn(m,r,N) follows the following steps: a, form N-m+1 m-dimensional vectors ; b, define as the number of vectors with a distance close to the vector within the tolerance r, and the ratio of the number to N-m; c, calculate d, calculate ; e, the sample entropy is defined as: ; wherein a lower sample entropy value represents a high regularity of the breathing pattern.

[0009] Further, in the state determination module, the real-time feature vector is matched and analyzed with the personalized behavior-physiological atlas, including: calculating a first similarity between the real-time feature vector and the intentional suppression feature set ; calculating a second similarity between the real-time feature vector and the intentional breathing feature set ; calculating a third similarity between the real-time feature vector and the reference physiological feature set ; if >θ1 or >θ1, and <θ2, it is determined that the state determination result is an active behavior suppression state; otherwise, it is determined to be a natural physiological state; wherein θ1 and θ2 are preset decision thresholds.

[0010] Further, the first similarity , the second similarity , and the third similarity are calculated using the cosine similarity method, and for two feature vectors A and B, the similarity is calculated as: , wherein A·B represents the dot product of the vector, and ||A|| and ||B|| represent the length of the vector.

[0011] Further, it further includes a reference calibration module: periodically or when a preset condition is met, the first task in the active behavior test protocol is re-executed to generate an iterative reference physiological signal, and the reference physiological feature set in the personalized behavior-physiological atlas is iterated based on the iterative reference physiological signal.

[0012] Further, periodically or when a preset condition is met, including: specifically including at least one of the following situations: a, reaching a preset fixed time period, the fixed time period is daily, weekly or monthly; b, the statistical difference between the natural state features extracted from the current monitoring data and the reference physiological feature set stored in the personalized behavior-physiological atlas exceeds a preset drift threshold.

[0013] Compared with the prior art, the present application can at least achieve the following beneficial effects: The present application introduces an active behavior test protocol, pre-acquires and constructs a personalized behavior physiological atlas unique to the user under benchmarking conditions, actively acquires the physiological fingerprint of the user that is real and not interfered by behavior strategy, and provides a key and objective comparison benchmark for subsequent differentiation between pathological signals and adaptive behavior signals; through matching analysis of the physiological characteristics monitored in real time and the personalized benchmark pre-established, a determination result about whether the user is currently in an active behavior inhibition state can be generated in real time, and the core of this determination is to identify the behavior pattern deliberately controlled by the user due to the perception of aura; once the active inhibition state is determined, the strategy execution module immediately triggers the preset data management strategy to mark and isolate the monitoring data of the corresponding time window; the above data isolation operation systematically excludes the physiological signal data contaminated or disguised by the adaptive behavior of the user from the training and optimization process of the core machine learning model for epilepsy event identification, ensuring long-term performance stability. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor. Among them: Figure 1 The system principle block diagram of the present application. DETAILED DESCRIPTION

[0015] The embodiments of the present application will be described in detail below, and the examples of the embodiments are represented in the drawings, wherein the same or similar symbols represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are only used to explain the present application, and cannot be understood as a limitation on the present application.

[0016] Please refer to Figure 1 The present application provides an epilepsy auxiliary diagnosis and treatment system based on a smart wearable device, comprising a data acquisition module, an atlas construction module, a feature extraction module, a state determination module and a strategy execution module. Data acquisition module: by executing a preset active behavior test protocol, the active behavior test protocol contains at least one standardized behavior task, acquires multi-modal physiological signals of the user when performing at least one standardized behavior task, and generates a test data set; In the data acquisition module, the active behavior test protocol includes the following standardized behavior tasks performed in sequence: First task, guide the user to keep natural relaxation; The second task is to guide the user to consciously suppress the movement of the wearing limb. The third task is to guide the user to perform regular deep breathing.

[0017] It should be noted that the active behavior test protocol is a structured interaction process guided by the intelligent wearable device for the user to complete. The core purpose is to actively induce and record the physiological response pattern of the user under specific intended behavior in a controlled condition, to provide benchmark data for subsequent construction of personalized behavior physiological atlas. The protocol includes three standardized behavior tasks performed in sequence, each task has clear behavior instructions and physiological state targets: The first task (natural relaxation task): This task aims to obtain the physiological baseline of the user in a completely relaxed and unconscious control state. The device guides the user into a resting state through human-computer interaction interface (such as screen display text, play voice prompts or specific vibration sequence). For example, the prompt can be: please sit quietly, relax your whole body, and breathe naturally for one minute. In this task, the user should avoid any conscious body control or focus on breathing, in order to capture the physiological signals closest to their daily resting state.

[0018] The second task (intentional suppression task): This task aims to induce and record the physiological features related to the effort cost of the user's active suppression of limb movement. The device guides the user to consciously and maximally keep the limb (usually the wrist) of the wearing part in a static state. For example, the prompt can be: now, consciously keep the arm wearing the device completely still, as if it were fixed, for 30 seconds. This task requires the user to focus on suppressing all subtle autonomous or involuntary movements, and the signals collected at this time will contain the autonomic nervous system activity features different from natural rest caused by suppression behavior.

[0019] The third task (intentional breathing task): This task aims to induce and record the physiological features of the user when performing regular and consciously controlled breathing patterns. The device guides the user to follow a clear and slow rhythm for deep breathing. For example, through dynamic visual instructions on the screen (such as a slowly expanding and contracting sphere) or an audio metronome, the user is guided to perform 6-8 deep abdominal breaths per minute for 1 minute. In this task, the high regularity and consciousness control of breathing are the key, and its physiological signal pattern will be significantly different from natural breathing or episode-related abnormal breathing.

[0020] It should be noted that during the execution of each of the above tasks, the intelligent wearable device synchronously collects multi-modal physiological signals. These signals come from multiple biological sensors integrated in the device, which together constitute a comprehensive description of the user's physiological state. Specifically, it can include: Motion signal: collected by inertial measurement units (e.g., accelerometer, gyroscope), used to quantify the amplitude, frequency, and stability (e.g., variance) of limb movements. In the intentional suppression task, this signal should approach a zero-noise level.

[0021] Electrocardiogram / heart rate variability signal: can be obtained indirectly or directly through optical pulse wave (PPG) sensors or dedicated electrodes. Used to extract heart rate and heart rate variability frequency-domain indicators (e.g., LF, HF power) reflecting autonomic nervous balance.

[0022] Respiratory signal: can be extracted from PPG waveforms through specific algorithms to detect volume changes caused by respiration, or indirectly obtained by monitoring the micro-movements of the chest and abdomen through accelerometers. Used to analyze respiratory frequency, depth, and regularity (e.g., calculate sample entropy).

[0023] Skin conductance signal: collected by skin conductance response sensors, reflecting changes in skin conductivity caused by sympathetic nervous activity, an important indicator of emotional arousal or effort level.

[0024] Test data set refers to the collection of time-synchronized multi-modal physiological signal raw data segments corresponding to the three tasks described above after completing a complete active behavior test protocol. This data set is the raw material for subsequent feature extraction and analysis.

[0025] Atlas construction module: based on the test data set, construct a personalized behavior physiological atlas; the personalized behavior physiological atlas at least contains a set of baseline physiological features for representing the user in a natural relaxed state; It should be noted that the personalized behavior physiological atlas is a structured feature database that is unique to a specific user, and its core role is to provide a quantifiable, personalized reference dictionary for subsequent intent recognition in real-time monitoring. The atlas contains at least three key feature sets, corresponding to the specific physiological patterns induced by the three standardized behavior tasks.

[0026] In the atlas construction module, the personalized behavior physiological atlas is constructed: Based on the signals collected during the execution of the first task, the baseline physiological feature set is calculated; It should be noted that the baseline physiological feature set is calculated based on the signals collected during the execution of the first task (natural relaxation task). Its purpose is to establish the physiological baseline of the user in an unconscious control, relaxed state. The calculation process usually includes: Signal preprocessing: filter and denoise the original electrocardiogram, respiratory, and skin conductance signals to eliminate power frequency interference and motion artifacts.

[0027] Feature extraction: extract indicators with statistical or physiological significance from pre-processed signals. For example: calculate mean heart rate (HR) and standard deviation of heart rate (SDNN) from heart rate signals; calculate total power in frequency domain (TP) from heart rate variability signals; calculate mean respiration rate from respiration signals; calculate mean skin conductance level (SCL) from skin conductance signals; calculate mean motion energy of three-axis resultant vector from acceleration signals; the collection of these features constitutes the baseline physiological feature set, representing the unique resting physiological fingerprint of the user.

[0028] Based on the signals collected during the execution of the second task, calculate the intentional suppression feature set, wherein the intentional suppression feature set includes a first feature index extracted from the heart rate variability signal, and the first feature index is used to quantify the degree of autonomic nervous effort; The first feature index is the ratio of low-frequency power to high-frequency power in the frequency domain analysis of heart rate variability, and the calculation formula is: Among them, is the signal power in the preset low-frequency range, is the signal power in the preset high-frequency range; a significant increase in the ratio is characterized as a physiological manifestation of increased sympathetic nervous activity during suppression exercise.

[0029] It should be noted that the intentional suppression feature set is calculated based on the signals collected during the execution of the second task (intentional suppression task). The core is to capture the physiological cost paid by the user for the willful behavior of intentionally keeping still, especially the response of the autonomic nervous system. The first feature index, i.e. the LF / HF ratio, is a key discriminative feature in this set.

[0030] Physiological principle: in the frequency domain analysis of heart rate variability (HRV), low-frequency power (LF, usually 0.04-0.15Hz) is co-regulated by sympathetic and parasympathetic nerves, but is related to vascular tone and baroreflex; high-frequency power (HF, usually 0.15-0.4Hz) mainly reflects parasympathetic (vagus) nerve activity, which is synchronized with the respiratory cycle. LF / HF ratio is widely considered as an indicator of cardiac sympathetic-vagus nerve balance or sympathetic nerve excitability.

[0031] For example, when the user intentionally suppresses movement, even if he is outwardly still, his brain needs to pay cognitive effort to control the muscles, and this internal effort will activate the sympathetic nervous system, causing the LF component to relatively increase and the HF component to possibly decrease, thereby significantly increasing the LF / HF ratio. This change is in sharp contrast to the ratio in the natural relaxed state. The calculation formula and ​​By performing spectral analysis (e.g., fast Fourier transform or autoregressive model) on the heart rate interval sequence, and integrating the corresponding frequency band.

[0032] In addition to the LF / HF ratio, the deliberate suppression feature set can also include: the minimum variance of motion energy during the task (confirming "stillness"), the rising slope of skin conductance level (reflecting sympathetic activation), and the slight increase in heart rate itself, etc. These features together outline the physiological portrait of deliberate suppression.

[0033] Based on the signals collected during the execution of the third task, a deliberate breathing feature set is calculated, which includes a second feature index for quantifying the regularity of the breathing signal.

[0034] The second feature index for quantifying the regularity of the breathing signal is the sample entropy of the respiratory interval sequence, and its calculation method is as follows: for a respiratory interval sequence RR = {rr1, rr2,..., rrN} of length N, set the pattern length m and the tolerance r, and the calculation of sample entropy SampEn(m, r, N) follows the following steps: a. Form N-m+1 m-dimensional vectors ; b. Define as the number of vectors with a distance close to the vector within the tolerance r, and the ratio of to N-m; c. Calculate ; e. The sample entropy is defined as: ; Where a lower sample entropy value represents a highly regular breathing pattern.

[0035] It should be noted that the deliberate breathing feature set is calculated based on the signals collected during the execution of the third task (deliberate breathing task). Its goal is to quantify the physiological pattern of the user when performing consciously controlled regular deep breathing.

[0036] Sample entropy is an algorithm for measuring the complexity and regularity of time series. For the respiratory interval sequence RR, the core idea of steps (a-e) is to investigate the conditional probability of the repetition of patterns of length m and m+1 in the sequence. The more regular and predictable the sequence, the higher the conditional probability, and the lower the sample entropy value calculated.

[0037] For example, when a user performs a deliberate deep breath, he / she will try to follow the given rhythm, resulting in very uniform breath intervals and a very regular sequence, thus an abnormally low sample entropy value. In contrast, natural breathing has normal physiological variations, and the entropy value is higher; pathological breathing before or during a seizure is usually disordered and irregular, and the entropy value may also be abnormally high or show a specific pattern of fluctuations. Therefore, extremely low sample entropy is an effective indicator for identifying deliberate control of breathing state.

[0038] Parameter selection: the mode length m is usually 1 or 2, and the tolerance r is usually 0.1 to 0.25 times the standard deviation of the original sequence. These parameters can be determined empirically or by optimization.

[0039] Set construction: in addition to sample entropy, the deliberate breathing feature set may also include: the degree of coincidence of breathing frequency with the guide frequency, the regularity of the breathing waveform, and the degree of phase locking of heart rate and breathing (such as the specific pattern of respiratory sinus arrhythmia) and other features.

[0040] Through the above detailed calculation and construction process, the final personalized behavior physiological atlas contains the baseline physiological feature set, the deliberate suppression feature set and the deliberate breathing feature set. These three sets provide accurate and personalized mathematical scales for the system to distinguish between natural state, active behavior suppression state and quantify the degree of deviation in subsequent steps.

[0041] Feature extraction module: in the daily monitoring process of the intelligent wearable device, real-time feature vectors of the current time window are extracted based on the monitoring data stream; It should be noted that the monitoring data stream refers to the original physiological signal sequence arranged in chronological order continuously collected and output by various biological sensors (such as IMU, PPG, EDA sensors) of the intelligent wearable device during daily wearing. This is a continuous data stream.

[0042] Current time window: the system does not always process all historical data, but uses a sliding time window for real-time analysis. The current time window is a recent, fixed-length (for example, lasting for 30 seconds or 60 seconds) data buffer. The window will slide (e.g., move forward by 1 second) with the arrival of new data, thus realizing continuous, near real-time analysis. Selecting an appropriate window length requires a balance between feature stability and detection delay: a too short window will be susceptible to noise interference, and a too long window will cause a delay in response to state changes.

[0043] Real-time feature vector extraction: for the raw monitoring data stream within the current time window, the system performs the same set of feature computation procedures as constructing the personalized behavioral physiological profile in the profile construction module. This means: for the heart rate variability signal within the window, compute its LF / HF ratio and other frequency domain indicators; for the respiratory interval sequence within the window, compute its sample entropy; for the motion signal within the window, compute its variance or energy; for the skin conductance signal within the window, compute its mean or slope; combine these computed feature values into a multi-dimensional real-time feature vector, in the order and structure that is exactly the same as the feature set in the personalized behavioral physiological profile. This real-time feature vector is a digital summary of the user's physiological state within the current time window.

[0044] State determination module: match the real-time feature vector with the reference physiological feature set in the personalized behavioral physiological profile, and generate a state determination result about the user's current behavioral intention; the state determination result is used to distinguish between the user's natural physiological state and the active behavioral inhibition state controlled by consciousness; In the state determination module, matching the real-time feature vector with the personalized behavioral physiological profile for analysis includes: calculating the first similarity between the real-time feature vector and the intentional inhibition feature set ; calculating the second similarity between the real-time feature vector and the intentional breathing feature set ; calculating the third similarity between the real-time feature vector and the reference physiological feature set ; If >θ1or >θ1, and <θ2, then determine that the state determination result is an active behavioral inhibition state; otherwise, determine that it is a natural physiological state; wherein θ1 and θ2 are preset decision thresholds.

[0045] The first similarity , the second similarity , and the third similarity are calculated using the cosine similarity method. For two feature vectors A and B, the similarity is calculated as: , where A·B represents the dot product of the vectors, and ||A|| and ||B|| represent the length of the vectors.

[0046] It should be noted that the cosine similarity is used to measure the similarity between two feature vectors. Its calculation formula Cosine similarity measures the directional consistency of vectors A and B in multidimensional space, and is insensitive to the absolute length (magnitude) of the vectors. This is well-suited for comparing physiological feature vectors because we need to focus on the similarity of physiological patterns (the relative proportions between features) rather than absolute signal strength. For example, regardless of whether a user's resting heart rate is 60 or 70, as long as relative patterns such as an increased LF / HF ratio and decreased exercise variance appear in their "deliberately suppressed" state, cosine similarity can effectively capture this pattern matching.

[0047] Among them, the first similarity : Calculate the cosine similarity between the real-time feature vector and the deliberately suppressed feature set vector in the graph. The higher the value, the closer the current physiological pattern is to the state when the user is deliberately keeping still.

[0048] Second similarity Calculate the cosine similarity between the real-time feature vector and the deliberate breathing feature set vector in the graph. The higher the value, the closer the current physiological pattern is to the state when the user consciously engages in regular deep breathing.

[0049] Third similarity : Calculate the cosine similarity between the real-time feature vector and the baseline physiological feature set vector in the map. The higher the value, the closer the current physiological pattern is to the user's baseline state when naturally relaxed.

[0050] It should be noted that, >θ1 or >θ1: Indicates that the current state is highly similar to at least one deliberate behavioral pattern. This triggers suspicion that the state may be under active control.

[0051] <θ2: Indicates that the current state is not similar to the baseline pattern of natural relaxation. This is a key condition used to exclude a situation where the user is already in a natural resting state (in which case...). (It should be very high), and this natural state may happen to have some similarity to a certain deliberate behavior pattern in a certain characteristic. Only when it is both like deliberate behavior and unlike a natural state, is the confidence in judging it as a state of active behavior inhibition strengthened.

[0052] Threshold (θ1, θ2) setting: The threshold can be determined through initial data analysis or model training. For example, collect multiple test data for the user, calculate the distribution of each similarity, and set θ1 as... and At the lower percentile (e.g., the 10th percentile) of the similarity distribution under the voluntary task to ensure sufficient sensitivity to the voluntary mode; set θ2 as the higher percentile (e.g., the 90th percentile) of the similarity distribution under the natural task to ensure sufficient specificity to the deviation from the natural state.

[0053] It should be noted that through the above matching analysis and logical judgment, the system finally generates a discrete state judgment result. This result interprets the continuous physiological signal stream into a higher level, clearly meaningful behavior intention label: natural physiological state or active behavior inhibition state. This judgment result is the direct basis for differentiated data management (such as whether to isolate data) in the subsequent strategy execution module, and is the key bridge connecting low-level signal processing and high-level application decision.

[0054] Strategy execution module: based on the state judgment result, execute a preset data management strategy on the monitoring data of the current time window; when the state judgment result is the active behavior inhibition state, the data management strategy includes marking and isolating the corresponding monitoring data so that it does not participate in the training and optimization of the core model for epilepsy event identification.

[0055] It should be noted that the data management strategy is a preset, rule-driven data processing logic. Once the state judgment module generates a state judgment result for the current time window, the system automatically triggers and executes the strategy corresponding to the result.

[0056] When judged as a natural physiological state: the corresponding strategy is usually to route the monitoring data of this time window and its extracted real-time feature vector to the main analysis channel. In this channel, the data will be input into the core epilepsy event detection or prediction model on the one hand, for real-time risk assessment and alarm judgment; on the other hand, in the system's set model incremental learning or updating period, these data can be used as high-quality, behavior-pollution-free positive or negative samples to optimize the model parameters, so that they better adapt to the user's current physiological mode.

[0057] When judged as an active behavior inhibition state: the corresponding strategy includes a key operation - marking and isolating. Specifically: Marking: the system will mark the metadata of the monitoring data of this time window with a specific label. This label records the time point, duration, and judgment basis (such as high or ) of the data being judged as being controlled by the user's consciousness.

[0058] Isolation: the original monitoring data and its feature vectors marked with this label will be stored in an independent, access-controlled isolated data buffer. This buffer is logically separated from the main database used for model training.

[0059] It's important to note that the core purpose of implementing isolation is to strictly cut off the interference of behavioral inhibition data on the training process of the core epilepsy monitoring model. Therefore, when the system performs incremental model learning, parameter optimization, or retraining, its data sampler actively excludes all data with behavioral inhibition labels from the isolation buffer. This fundamentally prevents the model from learning the erroneous association that deliberate inhibition equals safety.

[0060] This isolated data is not permanently discarded. After obtaining explicit user authorization and anonymizing, it can be used for: a. User behavior analysis: helps users or doctors understand their behavioral patterns and frequency in dealing with early warning signs.

[0061] b. Algorithm research: Under the supervision of doctors or researchers, it can be used to develop more advanced behavior recognition algorithms, or under very strict control to study "suppression failure" cases (i.e., situations where an attack occurs despite being identified as being in a suppressed state).

[0062] c. System audit: As a record, it is used to trace the rationality of the system's decision-making logic.

[0063] The benchmark calibration module periodically or when preset conditions are met re-executes the first task in the active behavior testing protocol to generate iterative benchmark physiological signals and iterates the set of benchmark physiological features in the personalized behavioral physiological atlas based on the iterative benchmark physiological signals.

[0064] Periodically or under preset conditions, including at least one of the following situations: a. Reaching a preset fixed time period, wherein the fixed time period is daily, weekly, or monthly; b. The statistical difference between the natural state features extracted from the current monitoring data and the baseline physiological feature set stored in the personalized behavioral physiological atlas exceeds a preset drift threshold.

[0065] It should be noted that the main problem addressed is the natural drift of the user's physiological baseline over time, ensuring that the reference benchmark in the personalized behavioral physiological map remains accurate and effective, thereby maintaining the long-term accuracy of intent recognition.

[0066] Iteration triggering conditions: Iteration is not arbitrary, but triggered by explicit conditions, mainly including two categories: (a) Fixed-time periodic triggering: This is the simplest and most reliable periodic maintenance mechanism. The system can be set to automatically guide the user to perform a simplified proactive behavior test protocol (usually only including the first task: the natural relaxation task) at fixed times such as every morning (when the user has just woken up and is in a stable state), every Sunday, or the first day of every month. This method can prevent baseline failure caused by slow, cumulative drift.

[0067] (b) Data-driven drift detection triggering: This is a more intelligent adaptive mechanism. Its implementation logic is as follows: The system continuously monitors daily data in the background for time windows judged as natural physiological states (i.e., high-confidence natural state samples). Features extracted from these windows are periodically (e.g., daily) statistically compared with the currently stored set of baseline physiological features in the atlas. For example, the Mahalanobis distance between the two sets of features in multidimensional space is calculated, or hypothesis testing (e.g., t-test) is performed on each key feature. If the calculated distance or the significance (p-value) of the difference exceeds a preset drift threshold, it indicates that the user's true physiological baseline has changed significantly, and the system immediately triggers a baseline update test.

[0068] New execution process: When any triggering condition is met, the system guides the user to re-execute the first task, collecting a new iterative baseline physiological signal. Subsequently: Using the same feature calculation method as in the atlas construction module, an iterative set of baseline physiological features (candidates) is generated from the new signal.

[0069] The update strategy can be direct replacement, that is, to completely cover the old baseline physiological feature set with the new feature set.

[0070] A smoother strategy could be weighted fusion, for example: new benchmark set = α * old benchmark set + (1-α) * candidate benchmark set, where α is a forgetting factor between 0 and 1, used to control the update speed and avoid drastic fluctuations in the benchmark due to noise from a single measurement.

[0071] Through the strategy execution module and the benchmark calibration module, this invention not only achieves real-time defensive isolation of behavioral contamination data (strategy execution module), but also establishes a proactive maintenance and calibration mechanism for the reference benchmark (benchmark calibration module). Together, they form a dynamic, adaptive system that effectively combats challenges arising from long-term user behavioral adaptation and changes in physiological state, thereby ensuring the stability and reliability of the core functions of epilepsy auxiliary diagnosis and treatment throughout the entire product lifecycle.

[0072] In summary, this patent application constructs a preventative defense system through a series of steps including proactive baseline establishment, real-time intent recognition, and contaminated data isolation. This ensures the long-term robustness, accuracy, and reliability of the core algorithm for epilepsy auxiliary diagnosis and treatment, effectively enhancing the safety value for patients in long-term use.

[0073] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0074] In the embodiments provided by this invention, it should be understood that the disclosed system or method can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for instance, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.

[0075] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0076] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0077] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the basic characteristics of the present invention.

[0078] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An epilepsy auxiliary diagnosis and treatment system based on intelligent wearable devices, characterized in that, include: Data acquisition module: By executing a preset active behavior testing protocol, which includes at least one standardized behavioral task, the module collects multimodal physiological signals of the user when performing at least one standardized behavioral task and generates a test dataset. Atlas construction module: Based on the test dataset, construct a personalized behavioral physiological atlas; the personalized behavioral physiological atlas contains at least one set of baseline physiological features to characterize the user in a natural relaxed state; Feature extraction module: During the daily monitoring process of the smart wearable device, the monitoring data stream is acquired in real time, and the real-time feature vector of the current time window is extracted based on the monitoring data stream; State determination module: Matches and analyzes the real-time feature vector with the set of baseline physiological features in the personalized behavioral physiological map to generate a state determination result about the user's current behavioral intention; Strategy execution module: Based on the status determination result, executes a preset data management strategy on the monitoring data of the current time window; When the state determination result is an active behavior inhibition state, the data management strategy includes marking and isolating the corresponding monitoring data.

2. The epilepsy auxiliary diagnosis and treatment system based on intelligent wearable devices as described in claim 1, characterized in that: In the data acquisition module, the proactive behavior testing protocol includes the following standardized behavioral tasks executed sequentially: The primary task is to guide users to maintain a natural and relaxed state. The second task is to guide users to consciously suppress the movement of the limbs wearing the device. The third task is to guide users to perform regular deep breathing.

3. The epilepsy auxiliary diagnosis and treatment system based on intelligent wearable devices as described in claim 2, characterized in that: In the map construction module, personalized behavioral physiological maps are constructed: The baseline physiological feature set is calculated based on the signals collected during the execution of the first task. Based on the signals collected during the execution of the second task, a set of deliberate inhibition features is calculated, wherein the set of deliberate inhibition features includes a first feature index extracted from the heart rate variability signal, and the first feature index is used to quantify the degree of autonomic neural effort. Based on the signals collected during the execution of the third task, a set of deliberate breathing features is calculated, which includes a second feature index for quantifying the regularity of breathing signals.

4. The epilepsy auxiliary diagnosis and treatment system based on intelligent wearable devices as described in claim 3, characterized in that: The first characteristic index is the ratio of low-frequency power to high-frequency power in the frequency domain analysis of heart rate variability, and its calculation formula is as follows: ， in, To preset the signal power in the low-frequency range, The signal power is set within the preset high-frequency range.

5. The epilepsy auxiliary diagnosis and treatment system based on intelligent wearable devices as described in claim 3, characterized in that: The second characteristic index used to quantify the regularity of respiratory signals is the sample entropy of the respiratory interval sequence, which is calculated as follows: For a respiratory interval sequence RR={rr1,rr2,...,rrN} of length N, with the pattern length set to m and the tolerance to r, the sample entropy SampEn(m,r,N) is calculated following these steps: a, forming N-m+1 m-dimensional vectors ; b, definition To find the vector within the tolerance r Vectors that are close to each other The ratio of the quantity to Nm; c, Calculation ; d, calculates the vector of dimension m+1. ; e, the sample entropy is defined as: ; Among them, lower sample entropy values ​​indicate that the breathing pattern is highly regular.

6. The epilepsy auxiliary diagnosis and treatment system based on intelligent wearable devices as described in claim 3, characterized in that: In the state determination module, the real-time feature vector is matched and analyzed with the personalized behavioral physiological map, including: Calculate the first similarity between the real-time feature vector and the deliberately suppressed feature set. ; Calculate the second similarity between the real-time feature vector and the deliberate breathing feature set. ; Calculate the third similarity between the real-time feature vector and the baseline physiological feature set. ; If satisfied >θ1 or >θ1, and If θ1 < θ2, the state determination result is determined to be an active behavior inhibition state; otherwise, it is determined to be a natural physiological state; where θ1 and θ2 are preset decision thresholds.

7. The epilepsy auxiliary diagnosis and treatment system based on intelligent wearable devices as described in claim 6, characterized in that: First similarity Second similarity and third similarity The cosine similarity method is used for calculation. For two feature vectors A and B, the similarity is calculated as follows: ， Where A·B represents the dot product of vectors, and ||A|| and ||B|| represent the magnitudes of the vectors.

8. The epilepsy auxiliary diagnosis and treatment system based on intelligent wearable devices as described in claim 6, characterized in that: It also includes a benchmark calibration module: periodically or when preset conditions are met, re-executes the first task in the active behavior testing protocol to generate iterative benchmark physiological signals, and iterates the set of benchmark physiological features in the personalized behavioral physiological atlas based on the iterative benchmark physiological signals.

9. The epilepsy auxiliary diagnosis and treatment system based on intelligent wearable devices as described in claim 8, characterized in that: Periodically or under preset conditions, including at least one of the following situations: a. Reaching a preset fixed time period, wherein the fixed time period is daily, weekly, or monthly; b. The statistical difference between the natural state features extracted from the current monitoring data and the baseline physiological feature set stored in the personalized behavioral physiological atlas exceeds a preset drift threshold.

Images