A flexible wearable patch device and system for ventricular arrhythmia risk warning

Abstract

The application discloses a flexible wearable patch device and system for ventricular arrhythmia risk warning, and relates to the technical field of arrhythmia warning, which comprises the following steps: contact impedance detection and polling sampling are performed on the multi-point intelligent electrodes of the flexible wearable patch device, a current working electrode pair and a fallback electrode pair are acquired, a switching trigger caliber is solidified, and a lead gating list is generated; according to the lead gating list, continuous wideband skin electrical signal acquisition and sliding window slicing are performed, baseline drift suppression, power frequency interference suppression and transient saturation segment rejection are executed, and a quality control sampling window is generated; and the quality control sampling window is input into a risk scale model, time sequence feature extraction and gated fusion processing are performed, and a risk scale item and a budget file switching instruction set are output. The application realizes key risk period data capture optimization, and shortens the warning response time.

Description

Technical Field

[0001] This invention relates to the field of arrhythmia early warning technology, and in particular to a flexible wearable patch device and system for early warning of ventricular arrhythmia risk. Background Technology

[0002] With the rapid evolution of medical electronics technology, flexible wearable devices have achieved breakthroughs in the field of real-time monitoring of cardiovascular diseases. Skin conductance signal acquisition technology based on intelligent sensors, through the integration of multi-point electrode arrays and wireless transmission modules, enables non-invasive and continuous monitoring of cardiac activity, significantly improving patient compliance and data acquisition efficiency in daily activities. In recent years, related research has focused on improving the robustness of signal processing algorithms and the biocompatibility of devices, driving a paradigm shift from traditional wired electrocardiograms to flexible wearable patch-based monitoring. The in-depth application of intelligent sensor technology, especially in the high-precision acquisition and low-power processing of broadband skin conductance signals, has laid the technological foundation for early warning of ventricular arrhythmias.

[0003] However, existing flexible wearable ECG monitoring technologies still have key bottlenecks: First, the dynamic fluctuations in electrode-skin contact impedance are not effectively quantified and compensated, leading to frequent baseline drift, power frequency interference, and transient saturation in the skin conductance signal, which severely weakens the signal-to-noise ratio of heart rhythm analysis and affects the accuracy of risk warning; Second, traditional systems use a fixed sampling strategy, which cannot dynamically optimize sampling density and window parameters according to real-time heart rhythm status, resulting in data redundancy and omission of key segments during high-risk periods, causing an increase in warning delay rate and making it difficult to meet the instantaneous response requirements for ventricular arrhythmia events. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a ventricular arrhythmia risk early warning system to solve the problems of signal distortion caused by electrode contact impedance fluctuations and early warning delay caused by fixed sampling strategies.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a ventricular arrhythmia risk early warning system, comprising: an electrode pair sampling module, which performs contact impedance detection and polling sampling on multi-point smart electrodes of a flexible wearable patch device to obtain the current working electrode pair and the retreating electrode pair, and solidifies the switching trigger caliber to generate a lead selection list; a sliding window slicing module, which performs continuous broadband skin conductance signal acquisition and sliding window slicing according to the lead selection list, and performs baseline drift suppression, power frequency interference suppression and transient saturation segment removal to generate a quality control sampling window; a feature extraction module, which inputs the quality control sampling window into a risk scale model, performs temporal feature extraction and gating fusion processing, and outputs risk scale entries and a budget level switching instruction set; a sampling window grading module, which determines the budget level according to the risk scale entries, and extracts the sliding step size, retained segment length and sampling density from the budget level switching instruction set to dynamically switch the sampling process and generate a graded sampling window; and a lead reconstruction module, which performs multi-electrode lead re-polling based on the graded sampling window, and performs lead reconstruction switching when an abnormal electrode state is detected, while recording key risk segments and generating an early warning event package.

[0008] As a preferred embodiment of the ventricular arrhythmia risk warning system of the present invention, the steps for performing contact impedance detection and polling sampling on the multi-point smart electrodes of the flexible wearable patch device are as follows:

[0009] Collect the electrode number, electrode position index and sampling channel mapping relationship of multi-point smart electrodes, and compile them into a polling queue list according to the polling order;

[0010] Based on the polling queue list, the contact impedance detection link of the multi-point smart electrode is activated, and short-term polling sampling is performed. At the same time, the interference occupied field and the missing measurement occupied field of the sampling window are retained to generate an impedance polling table.

[0011] By combining the impedance polling table with the electrode position index, electrode positions are paired and arranged to obtain a list of candidate electrode pairs. Paired polling sampling is then performed, and the stable contact impedance segment is locked to generate a candidate lead table.

[0012] As a preferred embodiment of the ventricular arrhythmia risk early warning system of the present invention, the steps for generating the lead gating list are as follows:

[0013] Based on the lead candidate list, the candidate electrode pair list is eliminated and prioritized to generate the current working electrode pair and the back-off electrode pair. The interference occupancy change, the missing measurement occupancy change and the contact impedance sudden change segment boundary are collected to form the switching trigger calibrator.

[0014] Based on the current working electrode pair and the back-off electrode pair, the switching trigger aperture is fixed as the electrode switching trigger aperture, and a lead selection list is generated.

[0015] As a preferred embodiment of the ventricular arrhythmia risk early warning system of the present invention, the step of performing continuous broadband electrodermal signal acquisition and sliding window slicing according to the lead gating list is as follows:

[0016] Based on the lead gating list, the sampling channel is verified, the gain and range are configured, and the sampling clock is aligned. Continuous broadband EKS signals are acquired to generate an EKS signal sampling stream.

[0017] The skin conduction signal sampling stream is sliced ​​according to a fixed sliding window length and sliding window step size, and the contact impedance detection state and body motion disturbance state are collected and integrated into an aligned sampling window.

[0018] As a preferred embodiment of the ventricular arrhythmia risk early warning system of the present invention, the steps for generating the quality control sampling window are as follows:

[0019] Slow-varying component tracking, slow-varying component subtraction, and window boundary continuity adjustment are performed on the aligned sampling window to generate a baseline-cleaned sampling window;

[0020] The baseline clean sampling window is suppressed in the power frequency center band, and transient saturation segments in the amplitude peak section are removed to generate a quality control sampling window.

[0021] As a preferred embodiment of the ventricular arrhythmia risk early warning system of the present invention, the steps for outputting risk scale entries and budget file switching instruction sets are as follows:

[0022] The quality control sampling window is checked for consistency between window number and time boundary, sampling point index range is aligned, range is unified and dynamic range is clipped, and missing measurement fields are extracted to generate a scale input package.

[0023] The risk scaling model includes an input embedding layer, a temporal feature extraction layer, a gated fusion layer, and a risk mapping layer;

[0024] In the input embedding layer, channel rearrangement and fragment alignment slicing are performed on the scale input packet, and the contact impedance detection state and the bulk motion disturbance state are spliced ​​into an evidence companion sequence to generate a window-by-window embedding sequence.

[0025] In the temporal feature extraction layer, local segment extraction, cross-window continuous segment splicing and mutation segment aggregation are performed on the window-by-window embedded sequence to obtain rhythmic unstable segments and morphological mutation segments, which are then integrated into a temporal representation sequence.

[0026] In the gated fusion layer, based on the contact impedance detection state, the volumetric disturbance state, and the missing measurement occupied field, the time series representation sequence is gated and weighted down to form a gated fusion representation sequence.

[0027] In the risk mapping layer, the gating fusion representation sequence is mapped window by window to risk scale entries, and budget level mapping is arranged to generate a budget level switching instruction set.

[0028] As a preferred embodiment of the ventricular arrhythmia risk early warning system of the present invention, the steps for generating the graded sampling windows are as follows:

[0029] Pair the window number and start / end time boundaries of the risk scale entries and budget file switching instruction set, and fill in the adjacent window numbers for mismatched window numbers, and output the window alignment table.

[0030] The budget level is determined in the window alignment table according to the window number, adjacent and consistent levels are continuously merged and the fallback and back check is performed to generate the level effective table.

[0031] Based on the gear activation table, extract the sliding step size, retained segment length and sampling density from the budget gear switching instruction set, check the switching safety window, and generate a sampling switching configuration table;

[0032] Based on the sampling switching configuration table, continuous broadband electrodermal signals are acquired, and sliding step size replacement, retained segment length replacement, and sampling density replacement are performed. At the same time, boundary alignment verification is performed to generate graded sampling windows.

[0033] As a preferred embodiment of the ventricular arrhythmia risk early warning system of the present invention, the steps of multi-electrode lead re-polling based on graded sampling windows are as follows:

[0034] The sampling windows are filtered by window number to obtain the set of window numbers in the high-density range of the budget level, and then combined with the window start and end time boundaries and the current working electrode pair to form a multi-polling list.

[0035] Based on the multi-polling list, multiple smart electrodes are activated sequentially within the corresponding window start and end time boundaries, and contact impedance detection and short-time broadband skin conduction signal sampling are performed to generate a multi-polling observation table.

[0036] As a preferred embodiment of the ventricular arrhythmia risk early warning system of the present invention, the steps for generating the early warning event package are as follows:

[0037] Extract the boundaries of the stable and abrupt contact impedance segments from the multiple polling observation table, and screen out the saturation and missing measurement segments in the short-time broadband EKS signal sampling segments to generate an abnormal electrode status list.

[0038] The list of abnormal electrode states is reconstructed and switched, and broadband skin conductance segments are extracted from the graded sampling window, while key risk segments are cropped out.

[0039] Key risk segments, electrode pairs before lead reconfiguration switching, electrode pairs after lead reconfiguration switching, window sequence numbers, and window start and end time boundaries are aggregated and encapsulated into a warning event package.

[0040] In a second aspect, the present invention provides a flexible wearable patch device, comprising:

[0041] The host is the core control and processing unit of the device, integrating signal processing, risk scale model calculation, sampling strategy control, and early warning event generation functions;

[0042] Cable sockets are used to stably connect transmission cables, enabling reliable transmission of electrical signals, control commands, and power supply.

[0043] Cable connectors connect the transmission cables to the host unit, completing the connection of signal and power supply paths;

[0044] A hub is a signal relay and integration component, responsible for aggregating multiple electrical signals from the acquisition patch.

[0045] The sensor patch is a flexible wearable main component with a built-in multi-point intelligent electrode array. It fits the skin to complete broadband skin electrical signal acquisition and real-time contact impedance detection.

[0046] The beneficial effects of this invention are as follows: by using intelligent sensors to detect contact impedance and poll for sampling, real-time monitoring of electrode-skin contact status is achieved, baseline drift and transient saturation are suppressed, and signal quality is improved; by using risk scale to drive the switching of sampling parameters, data capture optimization during critical risk periods is achieved, and the early warning response time is shortened. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a schematic diagram of a ventricular arrhythmia risk warning system.

[0049] Figure 2 This is a schematic diagram of a flexible wearable patch device.

[0050] Figure 3 This is a time-series change chart and a magnified view of a specific area of ​​the quality control indicators.

[0051] Figure 4 This is a comparison chart of the early warning performance indicators of multiple schemes.

[0052] Attached image description: Figure 21. Main unit; 2. Cable socket; 3. Cable connector; 4. Hub; 5. Data acquisition patch. Detailed Implementation

[0053] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0054] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0055] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0056] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a ventricular arrhythmia risk early warning system, comprising the following steps:

[0057] The electrode pair sampling module performs contact impedance detection and polling sampling on the multi-point smart electrodes of the flexible wearable patch device, obtains the current working electrode pair and the retreating electrode pair, and solidifies the switching trigger aperture to generate a lead selection list.

[0058] Collect the electrode number, electrode position index and sampling channel mapping relationship of multi-point smart electrodes, and compile them into a polling queue list according to the polling order;

[0059] Specifically, the system performs a power-on self-test on the multi-point smart electrodes of the flexible wearable patch device and verifies the availability status of the electrode numbers of the multi-point smart electrodes, compiling the available electrode numbers into an electrode number list; according to the electrode layout table of the flexible wearable patch device, the available electrode numbers are located one by one and the corresponding electrode position index is locked, and the electrode number list and electrode position index are paired in a one-to-one correspondence to form an electrode position index table; the sampling channels of the sampling front end are verified for channel connectivity and the sampling channel mapping relationship is output, the sampling channel mapping relationship is key-matched with the electrode position index table and written into an electrode position index and sampling channel mapping lookup table, the electrode position index and sampling channel mapping lookup table are arranged sequentially and a polling sequence number is added to generate a polling queue list.

[0060] Figure 2The host unit 1 is the core control and processing unit of the device, integrating functions such as signal processing, risk scale model calculation, sampling strategy control, and early warning event generation. It is the core of the entire system's calculation and command issuance. The cable socket 2 is the physical connection interface on the host unit 1, used to stably connect the transmission cable to achieve reliable transmission of electrical signals, control commands, and power supply, ensuring signal transmission stability. The cable connector 3 is the connection end that matches the cable socket 2, connecting the transmission cable to the host unit 1 to complete the signal and power supply path connection, ensuring communication between the host unit 1 and the front-end acquisition component. The hub 4 is a signal relay and integration component, responsible for aggregating the multiple electrical signals from the acquisition patch 5, completing the initial signal regularization and transmission distribution, and reducing multi-electrode signal interference. The acquisition patch 5 is a flexible wearable main component with a built-in multi-point intelligent electrode array. It fits the skin to complete broadband skin electrical signal acquisition and real-time contact impedance detection, and is the core front-end of signal sensing.

[0061] Based on the polling queue list, the contact impedance detection link of the multi-point smart electrode is activated, and short-term polling sampling is performed. At the same time, the interference occupied field and the missing measurement occupied field of the sampling window are retained to generate an impedance polling table.

[0062] Specifically, the contact impedance detection link of the multi-point smart electrode is closed and the detection parameters (detection excitation amplitude, detection excitation frequency, and short-time polling sampling duration) are loaded. The detection excitation is output to the selected electrode and the readback voltage is collected simultaneously to form a short-time polling sampling segment. The short-time polling sampling segment is sliced ​​according to the sampling window boundary and the proportion of saturation segment and the proportion of power frequency change are counted to form the sampling window interference occupation field. At the same time, the proportion of missing sampling points is counted to form the missing measurement occupation field. The polling sequence number, electrode number, sampling channel mapping relationship, contact impedance sampling value, sampling window interference occupation field and missing measurement occupation field are collected into an impedance polling table.

[0063] The impedance polling table is combined with the electrode position index to perform electrode position pairing and arrangement, obtain a candidate electrode pair list, perform pairwise polling sampling, lock the contact impedance stable segment, and generate a lead candidate table.

[0064] Furthermore, the impedance polling table is aligned with the electrode number and electrode position index to ensure a one-to-one correspondence between the sampling channel mapping relationship and the electrode position index. Based on the distance and lateral constraints of the electrode position pairing arrangement, the electrode position indexes are combined in pairs, and combinations with excessive spacing and conflicting same-side combinations are eliminated to form a candidate electrode pair list. The corresponding sampling channel mapping relationship is selected sequentially according to the candidate electrode pair list and paired polling sampling is performed to collect contact impedance sampling segments. The contact impedance sampling segments are sliced ​​according to the sampling window, and sampling windows that meet the upper limit of the sampling window interference occupancy field and the missing measurement occupancy field are retained (e.g., the upper limit of occupancy is 5%). The continuous sampling window interval where the contact impedance fluctuation amplitude does not exceed the stable threshold is locked as the contact impedance stable segment. The candidate electrode pair list, the contact impedance stable segment boundary and the contact impedance sampling segment index are compiled to form the lead candidate table.

[0065] It should be noted that the stability threshold (example range is 0.5% to 3% relative fluctuation of contact impedance) is set as follows: the lower limit is based on the noise floor and range resolution of the contact impedance detection link within the upper limit constraint, and is taken as a value slightly higher than the noise floor; the upper limit is based on the allowable contact impedance drift amplitude of the flexible wearable patch device and the requirement of the lead candidate table for the continuity of the contact impedance stable section, and is taken as the maximum value that does not affect the locking of the stable section.

[0066] Based on the lead candidate list, the candidate electrode pair list is eliminated and prioritized to generate the current working electrode pair and the back-off electrode pair. The interference occupancy change, the missing measurement occupancy change and the contact impedance sudden change segment boundary are collected to form the switching trigger calibrator.

[0067] Specifically, the candidate electrode pair list is expanded item by item according to the lead candidate table. The contact impedance stable segment length, stability threshold satisfaction, and upper limit satisfaction of the sampling window interference occupancy field and missing measurement occupancy field of each candidate electrode pair are checked. Candidate electrode pairs with too short a stable segment or exceeding the occupancy limit (e.g., stable segment length less than 200 milliseconds) are eliminated. The remaining candidate electrode pairs are sorted by key according to stable segment length, low interference occupancy field, low missing measurement occupancy field, and contact impedance level close to the median range. The first item in the sort is taken as the current working electrode pair, and the next item is taken as the fallback electrode pair. The increment of the interference occupancy change and missing measurement occupancy change of the current working electrode pair is extracted according to the sampling window sequence number, and the start and end sequence numbers of the contact impedance change segment boundary are collected to form the switching trigger caliber.

[0068] It should be noted that the priority ranking is based on the following criteria: the length of the stable contact impedance segment is prioritized to retain candidate electrode pairs with consistently stable contact states; the low value of the sampling window interference field is prioritized to retain candidate electrode pairs with less external interference; the low value of the missing measurement field is prioritized to retain candidate electrode pairs with good sampling continuity; and the small deviation of the contact impedance level from the median range is prioritized to retain candidate electrode pairs with contact impedance that is within a moderate range, neither too high nor too low.

[0069] Interference occupancy change refers to the increase or decrease in the proportion of the interference occupancy field between adjacent sampling windows, used to characterize the short-term fluctuation range of interference strength.

[0070] The change in missing data occupancy refers to the increase or decrease in the proportion of missing data occupancy between adjacent sampling windows, which is used to characterize the short-term fluctuation of the density of missing data points.

[0071] The boundary of the contact impedance abrupt change segment refers to the starting and ending sampling window numbers where the contact impedance changes between adjacent sampling windows in the contact impedance sampling segment. It is used to calibrate the time range of the abrupt change segment (e.g., starting number 12, ending number 15).

[0072] Based on the current working electrode pair and the back-off electrode pair, the switching trigger aperture is fixed as the electrode switching trigger aperture, and a lead selection list is generated.

[0073] Specifically, the current working electrode pair and the backoff electrode pair are associated with the sampling channel mapping relationship and the lead switching direction is locked. The switching trigger caliber is organized into a switching condition set, including the interference occupancy change amplitude, the missing measurement occupancy change amplitude, and the duration of the contact impedance abrupt change segment boundary. The interference occupancy change amplitude, the missing measurement occupancy change amplitude, and the duration of the contact impedance abrupt change segment boundary in the switching condition set are normalized to a reference value to generate a corresponding dimensionless trigger score. The dimensionless trigger score is then pruned and uniformly registered to solidify the value range as the electrode switching trigger caliber. A trigger hold duration and a backoff hold duration (e.g., 100 ms and 300 ms) are added to the electrode switching trigger caliber. The mapping relationship between the current working electrode pair, the backoff electrode pair, the electrode switching trigger caliber, and the corresponding sampling channel is collected and encapsulated to form a lead selection list.

[0074] The sliding window slicing module performs continuous broadband EEG signal acquisition and sliding window slicing based on the lead gating list, and performs baseline drift suppression, power frequency interference suppression and transient saturation segment removal to generate a quality control sampling window;

[0075] Based on the lead gating list, the sampling channel is verified, the gain and range are configured, and the sampling clock is aligned. Continuous broadband EKS signals are acquired to generate an EKS signal sampling stream.

[0076] Specifically, based on the lead selection list, the mapping relationship of the sampling channels corresponding to the current working electrode pair is locked. The connectivity self-check and channel number consistency check are performed for each sampling channel to complete the sampling channel verification. The gain and range levels are issued according to the gain and range parameter set, and the amplitude peak occupancy and zero drift amplitude are checked using short-time test waveforms to complete the gain and range configuration. The sampling clock reference source is aligned with the sampling start edge, and the cross-channel sampling phase difference is checked to complete the sampling clock alignment. The current working electrode pair is kept in the selection state to continuously acquire continuous broadband EKS signals and encapsulate them in the order of sampling point index and timestamp field to form an EKS signal sampling stream.

[0077] For the electrodermal signal sampling stream, the sliding window is sliced ​​according to a fixed sliding window length and sliding window step size, and the contact impedance detection state and body motion disturbance state are collected and integrated into an aligned sampling window;

[0078] Furthermore, a window start and end time boundary and a sampling point index range are generated on the timestamp field of the EDR signal sampling stream according to a fixed sliding window length and sliding window step size (e.g., sliding window length of 2 seconds and sliding window step size of 0.25 seconds). The window start and end time boundaries are continuously checked and over-boundary clipping is performed. The corresponding EDR signal sampling segment is extracted according to the sampling point index range of each window and a window number is attached. The contact impedance detection state and body motion disturbance state are synchronously collected within the window start and end time boundaries, and the same timestamp field is used for alignment mapping and missing segments occupying fields are filled in. The EDR signal sampling segment, contact impedance detection state and body motion disturbance state are collected and encapsulated into an aligned sampling window.

[0079] It should be noted that the contact impedance detection status refers to the skin contact electrical status information of the multi-point smart electrodes of the flexible wearable patch device during the current sampling period. It is collected by applying impedance detection signals to each electrode or electrode pair and synchronously reading the impedance response results.

[0080] Body motion disturbance state refers to the motion disturbance information caused by limb movements, changes in body position, or patch deformation of the wearer during the current sampling period. It is obtained by synchronously collecting the motion amplitude, posture changes, and disturbance duration through the inertial sensor built into the patch.

[0081] Slow-varying component tracking, slow-varying component subtraction, and window boundary continuity adjustment are performed on the aligned sampling window to generate a baseline-cleaned sampling window;

[0082] Furthermore, the aligned sampling window is expanded according to the window number, the start and end time boundaries of the window and the sampling point index range are locked, and the skin electrodermal signal sampling segments are extracted. Sliding smoothing, trend fitting and endpoint extension are performed on the skin electrodermal signal sampling segments to generate a slow-varying component sequence, and amplitude limiting and abrupt point masking are performed on the slow-varying component sequence. The slow-varying component sequence and the skin electrodermal signal sampling segments are subtracted point by point according to the sampling point index to perform slow-varying component subtraction. The skin electrodermal signal sampling segments after slow-varying component subtraction are subjected to overlapping area weighted splicing, endpoint transition reduction and amplitude jump clipping at the start and end time boundaries of the window to perform window boundary continuity trimming (e.g., overlapping area 50 ms), and encapsulated together with the contact impedance detection state and body motion disturbance state to generate a baseline clean sampling window.

[0083] The baseline clean sampling window is suppressed in the power frequency center band, and transient saturation segments in the amplitude peak section are removed to generate a quality control sampling window.

[0084] Furthermore, the baseline purification sampling window is expanded according to the window number, skin signal sampling segments are extracted and the start and end time boundaries of the window are retained. Band-stop suppression is performed on the skin signal sampling segments around the power frequency center band and the amplitude range before and after suppression is checked. The amplitude peak sampling points that reach the upper boundary, lower boundary or continuously close to the boundary of the sampling range are scanned along the sampling point index and continuous peak intervals are aggregated. Signal segments composed of amplitude peak sampling points and whose duration is in a short interval are registered as transient saturation segments. The buffer is extended to both ends of the interval to form a list of elimination intervals. The corresponding sampling points are shielded according to the list of elimination intervals and connected by interpolation with adjacent non-peak sampling points. Boundary smoothing convergence is performed. The processed skin signal sampling segments, together with the contact impedance detection state and the body motion disturbance state, are encapsulated to generate a quality control sampling window.

[0085] It should be noted that transient saturation segments refer to segments in the electrodermal signal sampling where the signal amplitude reaches the sampling range boundary within a short time and forms a continuous peak due to instantaneous electrode impact, body traction, short-term overload of the amplification link, or sudden change in contact state.

[0086] Figure 3This diagram uses a combined upper and lower view to illustrate the temporal variations of baseline drift amplitude, interference occupancy ratio, saturation occupancy ratio, and missing measurement occupancy ratio during continuous monitoring under the complete solution. The upper overview diagram reflects the overall fluctuation of each quality control indicator throughout the monitoring period, and a local magnified analysis window is marked with a red dashed rectangle. The lower local magnified diagram further presents the detailed changes of each indicator within this window, allowing a clear view of the synchronous fluctuations and phased convergence phenomena among different quality control indicators during local disturbance phases. This diagram illustrates that the present invention, through contact impedance detection, polling sampling, and quality control sampling window generation processing, can perform fine-grained identification and constraint of baseline drift, interference intrusion, transient saturation, and missing measurement occupancy, providing a higher-quality input basis for the subsequent generation of risk scale entries, thereby supporting subsequent sampling parameter switching and key risk segment capture.

[0087] Baseline drift amplitude represents the degree of low-frequency slow-varying deviation of the continuous broadband electrodermal signal within the current time period, that is, the amplitude of the signal baseline drifting up and down relative to its stable position. This line mainly reflects the impact of changes in electrode-skin contact, loosening of the electrode, or chronic interference on the stability of the original signal. The larger the value, the more unstable the signal, and the more difficult it is to interpret the subsequent waveform.

[0088] Interference Occupancy Ratio: This represents the proportion of data length within the current sampling window affected by external noise, motion disturbances, or other abnormal interference to the total sampling window length. This line primarily reflects the degree to which invalid disturbances intrude into the signal; a higher value indicates fewer effective signal components that can be directly used for analysis within the current window.

[0089] Saturation Occupancy Ratio: This represents the proportion of the current sampling window containing transient saturation, amplitude truncation, or excessive front-end amplification links, relative to the total sampling window length. This line primarily reflects whether the signal experiences short-term distortion or amplification anomalies. A higher value indicates a larger area within the current window affected by saturation, resulting in more severe loss of waveform details.

[0090] Missing Data Occupancy Rate: This represents the proportion of data length within the current sampling window that failed to form a valid record due to poor contact, abnormal acquisition, or data loss, out of the total sampling window length. This line primarily reflects the data continuity during the current monitoring process; a higher value indicates poorer sampling link integrity, which will also affect the reliability of subsequent risk scale entries.

[0091] The feature extraction module inputs the quality control sampling window into the risk scale model, performs time-series feature extraction and gating fusion processing, and outputs risk scale entries and budget file switching instruction sets;

[0092] The quality control sampling window is checked for consistency between window number and time boundary, sampling point index range is aligned, range is unified and dynamic range is clipped, and missing measurement fields are extracted to generate a scale input package.

[0093] Furthermore, the quality control sampling windows are expanded according to window numbers, and the incremental relationship of window numbers and the continuity of window start and end time boundaries are checked. If window start and end time boundaries overlap or are discontinuous, the window numbers are rearranged and the window start and end time boundaries are trimmed and padded. The sampling point index range is aligned to a fixed length according to a unified window length. Overly long segments are truncated at the end, and insufficient segments are padded by interpolation at the missing sampling positions indicated by the missing measurement occupancy field, while retaining the missing measurement occupancy field. The range is unified for the EEG signal sampling segments, mapping the amplitude of different range levels to a unified amplitude unit, and amplitude is trimmed according to a unified dynamic range to suppress outlier spikes (e.g., trimmed to ±3 mV). The EEG signal sampling segments after range unification and dynamic range trimming, window numbers, window start and end time boundaries, contact impedance detection status, body motion disturbance status, and missing measurement occupancy fields are collected and packaged to generate a calibration input package.

[0094] The risk scaling model includes an input embedding layer, a temporal feature extraction layer, a gated fusion layer, and a risk mapping layer;

[0095] Furthermore, the risk scaling model uses the scaling input package as the entry point. The input embedding layer, the temporal feature extraction layer, the gated fusion layer, and the risk mapping layer establish a unified reference chain along the window sequence number, the window start and end time boundaries, and the sampling point index range. The window-by-window embedding sequence output by the input embedding layer serves as the processing object of the temporal feature extraction layer. The temporal feature extraction layer forms a temporal representation sequence on the same reference chain. The gated fusion layer aligns the contact impedance detection state, body motion disturbance state, and missing measurement occupied field to the temporal representation sequence according to the window sequence number to form a gated fusion representation sequence. The risk mapping layer outputs risk scaling entries based on the gated fusion representation sequence and simultaneously outputs the budget file switching instruction set.

[0096] It should be noted that the risk scale model training uses the scale input package as the training sample carrier. The quality control sampling window is aligned with key risk segments in the historically labeled risk segment library according to the window number and the window start and end time boundaries. Risk scale entry labels are generated for each window according to the labeling specifications, and budget file switching instruction set labels are generated, which are then compiled to form the training sample set. The historically labeled risk segment library consists of offline collected broadband electrodermal signal segments, corresponding contact impedance detection states, body motion disturbance states, and risk level results after manual labeling or clinical review. The training sample set undergoes standardized measurement ranges, dynamic range pruning, retention of missing measurement fields, and balanced category sampling, and is divided into training subsets and verification subsets proportionally. The risk scale model reads the training subsets in batches to complete forward mapping and generates loss quantities based on label differences, executing parameter feedback to update the risk scale model parameters. After each round of parameter updates, the verification subset is used to check the consistency of risk scale entries and budget file switching instruction sets, and the risk scale model parameter version with the highest consistency is selected and fixed as the online version.

[0097] In the input embedding layer, channel rearrangement and fragment alignment slicing are performed on the scale input packet, and the contact impedance detection state and the bulk motion disturbance state are spliced ​​into an evidence companion sequence to generate a window-by-window embedding sequence.

[0098] Furthermore, the input embedding layer expands the window sequence of the scaled input packet one by one, locks the channel order according to the sampling channel mapping relationship, performs channel rearrangement on the skin electrodermal signal sampling segments, and splices the channel data into multi-channel segments according to a unified order; performs segment alignment slicing on the multi-channel segments according to the window start and end time boundaries and the sampling point index range, truncates the tail segments that exceed the unified window length, and interpolates the gap positions that are less than the unified window length according to the missing measurement occupied field to form fixed-length slices; simultaneously obtains the contact impedance detection state and body motion disturbance state and aligns them according to the window sequence, extends the contact impedance detection state and body motion disturbance state to the same time sequence as the fixed-length slice and splices them into an evidence companion sequence, and encapsulates the fixed-length slice and evidence companion sequence according to the time sequence to generate a window-by-window embedding sequence.

[0099] In the temporal feature extraction layer, local segment extraction, cross-window continuous segment splicing and mutation segment aggregation are performed on the window-by-window embedded sequence to obtain rhythmic unstable segments and morphological mutation segments, which are then integrated into a temporal representation sequence.

[0100] Furthermore, the temporal feature extraction layer unfolds the window-by-window embedding sequence in order of window number. For each window, the fixed-length slice is divided according to the sub-window step size and local fragment vectors are extracted. At the same time, the start and end indices of the sub-windows are recorded. The temporal feature extraction layer performs amplitude change comparison and morphological similarity comparison on the local fragment vectors of adjacent sub-windows, and locks the intervals with concentrated changes as candidate intervals for mutation fragments. The overlapping area index of adjacent two windows is aligned at the boundary, and the candidate intervals for mutation fragments are extended and spliced ​​at the window boundary to form cross-window continuous fragments. The cross-window continuous fragments with adjacent positions are aggregated and merged and divided into rhythmically unstable fragments and morphological mutation fragments according to the change type. The fragment index and window number are collected and encapsulated to form a temporal representation sequence.

[0101] In the gated fusion layer, based on the contact impedance detection state, the volumetric disturbance state, and the missing measurement occupied field, the time series representation sequence is gated and weighted down to form a gated fusion representation sequence.

[0102] Furthermore, the gated fusion layer expands the time-series representation sequence according to the window number, and aligns the contact impedance detection state, body motion disturbance state, and missing measurement occupied field to the segment index of each rhythmic unstable segment and morphological change segment according to the start and end time boundaries of the window, forming a gated alignment table; based on the gated alignment table, gated pruning is performed on the coverage interval of the missing measurement occupied field and the occupied interval of the body motion disturbance state, deleting the corresponding segment index and retaining the pruning boundary; for the retained segment index, weight reduction is performed according to the coverage ratio of the stable segment of the contact impedance detection state, mapping the coverage ratio to a weight coefficient and multiplying it with the segment vector to obtain the reduced segment vector; the reduced segment vector is collected and encapsulated according to the window number to form a gated fusion representation sequence.

[0103] It should be noted that gating and weight reduction are used to remove or reduce the participation of segments that are heavily affected by missing measurements, body motion disturbances, or poor electrode contact stability from the time-series representation sequence, so that the subsequent risk scale can more effectively reflect the ventricular risk changes corresponding to the effective ECG segments.

[0104] In the risk mapping layer, the gating fusion representation sequence is mapped window by window to risk scale entries, and budget level mapping is arranged to generate a budget level switching instruction set.

[0105] The risk mapping layer expands the gated fusion representation sequence according to the window number. For each window, the suppressed segment vector is aggregated into a window-level representation vector according to the segment index. The risk scale value is calculated on the window-level representation vector according to the risk scale value relationship, and encapsulated to form a risk scale entry bound to the window number and the window start and end time boundary. Candidate budget levels are obtained according to the budget level mapping and are compared with the effective budget level of the previous window window window by window. If the candidate budget level is higher than the effective budget level of the previous window window, the switch is performed directly. If the candidate budget level is lower than the effective budget level of the previous window window window, a fallback hysteresis and rollback check are performed. The number of windows that keep the candidate budget level at the lower level is continuously counted and the missing test occupied field is checked to ensure that it has not increased. If the number of windows that need to be kept is met, the fallback is allowed. If the number of windows that need to be kept is not met, the fallback is to the effective budget level of the previous window window window. The sliding step size, retained segment length and sampling density fields are filled according to the effective budget level to generate a budget level switching instruction set.

[0106] It should be noted that the example range for the number of windows is 2 to 5. The number of windows is defined based on the shortest risk stabilization time calculated from the fixed sliding window length and the sliding window step size. The value of the number of windows is to ensure that the shortest risk stabilization time covers continuous window segments and does not exceed the acceptable response delay for budget level switching.

[0107] The formula for calculating the risk scale value is:

[0108] ;

[0109] in, The window number is The corresponding risk scale value at that time Indicates the window number. This represents a monotonically bounded normalization mapping function used to map weighted convergent quantities to a uniform scale interval. The dimension index representing the window-level representation component. The number of dimensions representing the window-level representation components. Indicates the window sequence number of the gated fusion representation sequence. The first convergence point obtained Window-level representation components Indicates the first The weighting coefficients corresponding to the window-level representation components. Indicates window number The corresponding proportion of body motion disturbances. Indicates window number The corresponding percentage of missing tests, Indicates window number The corresponding coverage ratio of the stable contact impedance section This represents the volumetric disturbance compression coefficient. Indicates the low coefficient of the missing pressure measurement. This indicates insufficient stability and a reduced coefficient.

[0110] The weight coefficients corresponding to the characterization components (example range: 0-1) are updated based on the risk scale model parameters driven by the training sample set and normalization constraints are applied. At the same time, the consistency of risk scale entries in the verification subset and the consistency of budget file switching instruction set are used as the selection criteria.

[0111] The volumetric disturbance reduction factor (example range is 0.5-6) ​​is determined by comparing the reduction of the risk scale value when the volumetric disturbance occupancy ratio increases, and the boundary value is set as the budget file switching instruction set in the verification subset does not jitter and the key risk segments can still be covered.

[0112] The missing test suppression coefficient (example range is 0.5-8) is determined by comparing the reduction of the risk scale value when the missing test occupancy ratio increases, and the boundary value is set by the fact that the window segment with higher missing test occupancy in the verification subset does not trigger the high budget level.

[0113] The stability deficiency suppression factor (example range is 0.5-8) is determined by comparing the reduction of the risk scale value when the coverage ratio of the stable contact impedance section decreases, and the boundary value is taken as the fact that the window segment with insufficient coverage of the stable contact impedance section in the verification subset does not enter the high budget level.

[0114] The sampling window grading module determines the budget grading level based on the risk scale items, and extracts the sliding step size, retained segment length and sampling density from the budget grading switching instruction set to dynamically switch the sampling process and generate graded sampling windows.

[0115] Pair the window number and start / end time boundaries of the risk scale entries and budget file switching instruction set, and fill in the adjacent window numbers for mismatched window numbers, and output the window alignment table.

[0116] Furthermore, the risk scale entries and budget file switching instruction sets are sorted separately by window number. The window number and window start and end time boundaries of each entry are extracted to form a pairing key. One-to-one matching is performed on the pairing key. Entries that match successfully are grouped together by the same window number, and the risk scale entry field and budget file switching instruction set field are retained to form an aligned row. For window numbers that fail to match, adjacent window numbers are used for padding. The previous window number or the next window number is selected as the padding window number based on the overlap and time interval between the window start and end time boundaries and the adjacent window start and end time boundaries. The coverage relationship of the window start and end time boundaries corresponding to the padding window number is checked to ensure consistency. The aligned row and the padding row are encapsulated and output as a window alignment table in window number order.

[0117] The budget level is determined in the window alignment table according to the window number, adjacent and consistent levels are continuously merged and the fallback and back check is performed to generate the level effective table.

[0118] Furthermore, the window alignment table is expanded row by row according to the window number, and the risk scale items and budget level fields in the budget level switching instruction set are extracted to form candidate budget levels. The candidate budget levels are then compared with the effective budget level of the previous window to determine the budget level. When the candidate budget level is consistent with the effective budget level of the previous window, the same effective window segment is extended and the start and end time boundaries of the window are merged. When the candidate budget level is higher than the effective budget level of the previous window, a new effective window segment is opened. When the candidate budget level is lower than the effective budget level of the previous window, a fallback and rollback check are performed. The number of windows that keep the candidate budget level at a lower level is continuously counted and the missing test occupied field is checked to ensure that it has not increased. If the number of windows that need to be kept is met, the candidate budget level is switched. If the number of windows that need to be kept is not met, the effective budget level of the previous window is maintained. The effective window segments are collected and packaged to form a level effective table.

[0119] It should be noted that the budget level determination refers to comparing the candidate budget levels in the same window alignment table with the effective budget level of the previous window window by window according to the window number, and determining the effective budget level of the current window based on the following criteria: if they are equal, the effective window segment is extended; if they increase, the effective window segment is switched; if they decrease, a fallback and rollback verification are performed.

[0120] Based on the gear activation table, extract the sliding step size, retained segment length and sampling density from the budget gear switching instruction set, check the switching safety window, and generate a sampling switching configuration table;

[0121] Furthermore, the effective window segments are expanded segment by segment according to the effective window table, and the window number range and window start and end time boundaries of the effective window segments are extracted. The window number is used to match each budget switch instruction set and verify that the window start and end time boundaries are consistent. The sliding step size, retained segment length and sampling density are extracted to form candidate switch entries. The switch safety window is checked for the candidate switch entries. The switch window number is checked to be located in the middle of the effective window segment and to maintain a safe interval with the window start and end time boundaries. The change range of the sliding step size between two adjacent levels is checked to be within the allowable range. The retained segment length is checked to cover the overlapping area on both sides of the switch window and to check that the change of sampling density does not cause an increase in the field occupied by missing measurements. If the switch safety window check fails, the switch window number is moved to the next available window number and the previous level configuration is maintained. The candidate switch entries that pass the switch safety window check are collected and packaged according to the window number to generate a sampling switch configuration table.

[0122] It should be noted that the allowable range of amplitude (example range: 1 to 3 sliding window steps) is defined based on the boundary overlap area length and the coverage capability of the retained segment length obtained by converting the fixed sliding window length and the sliding window step length, which satisfies the continuity of boundary alignment verification and does not cause an increase in the field occupied by missing tests.

[0123] Based on the sampling switching configuration table, continuous broadband electrodermal signals are acquired, and sliding step size replacement, retained segment length replacement, and sampling density replacement are performed. At the same time, boundary alignment verification is performed to generate graded sampling windows.

[0124] Furthermore, the current working electrode pair is selected to acquire continuous broadband electrodermal signals according to the window number range of the sampling switching configuration table, and the sampling point index and timestamp fields are numbered consecutively. When the switching window number indicated by the sampling switching configuration table is encountered, the sliding step size replacement is performed, and the retention segment length replacement and sampling density replacement are performed simultaneously. The sampling density replacement is performed by extracting sampling points or interpolating supplementary points to adjust the sampling point interval. Boundary alignment verification is performed, and the tail segment of the previous window is truncated according to the retention segment length and spliced ​​with the first segment of the new window. The timestamps at the splicing point are checked to be continuous, the sampling point index does not backtrack, and the window start and end time boundaries cover the sliding step size. The spliced ​​window segments are collected according to the window number to generate graded sampling windows.

[0125] The lead reconstruction module performs multi-electrode lead re-polling based on graded sampling windows, and performs lead reconstruction switching when an abnormal electrode state is detected. At the same time, it records key risk segments and generates early warning event packages.

[0126] The sampling windows are filtered by window number to obtain the set of window numbers in the high-density range of the budget level, and then combined with the window start and end time boundaries and the current working electrode pair to form a multi-polling list.

[0127] Furthermore, the tiered sampling windows are expanded sequentially by window number, and the sampling density field corresponding to each window is extracted. The sampling density field is then checked against the corresponding sampling density level registered in the budget level switching instruction set window by window. When the sampling density field corresponds to a high-density sampling level interval, the corresponding window number is determined to belong to the high-density sampling level interval. Window numbers that meet the high-density interval are filtered out, and duplicate window numbers are removed to form a window number set. The window number set is matched against the start and end time boundaries of the tiered sampling windows one by one, and the continuity of the start and end time boundaries is checked and window numbers with missing start and end time boundaries are removed. The window number set is bound to the corresponding start and end time boundaries one by one, and the current working electrode pair is attached to each record as an association field. The results are then collected and encapsulated in the order of window number to form a multi-polling list.

[0128] It should be noted that the high-density interval filtering reads the sampling density field of the graded sampling window one by one according to the window number, and aligns the sampling density field of the graded sampling window with the sampling density field of the corresponding window number of the budget grade switching instruction set. If the sampling density field falls into the highest sampling density grade of the budget grade switching instruction set, the budget grade is determined to be in the high-density interval and included in the window number set. If the sampling density field does not fall into the highest sampling density grade, it is removed.

[0129] High-density range (example range is sampling density that is 1.3 to 2.0 times that of normal sampling density). The high-density range is defined according to the sampling density field of the budget file switching instruction set. The value is set to meet the requirements of improving the coverage of key risk segments without increasing the number of missing test fields occupied by the tiered sampling window.

[0130] Based on the multi-polling list, multiple smart electrodes are activated sequentially within the corresponding window start and end time boundaries, and contact impedance detection and short-time broadband skin conduction signal sampling are performed to generate a multi-polling observation table.

[0131] Furthermore, the window sequence number and window start and end time boundaries are expanded one by one according to the multi-polling list, keeping the current working electrode pair selected and locking the multi-polling execution interval corresponding to the window start and end time boundaries; within the multi-polling execution interval, the sampling channel mapping relationship of the multi-point smart electrodes is selected sequentially according to the polling sequence number of the polling queue list, the contact impedance detection link is closed and contact impedance sampling segments are collected, and at the same time, short-time broadband skin signal sampling segments are collected for the selected electrodes within the same window start and end time boundaries; window sequence number, timestamp field and electrode number are added to the contact impedance sampling segments and short-time broadband skin signal sampling segments, and the continuity of the sampling point index is checked, and the contact impedance sampling segments and short-time broadband skin signal sampling segments formed by each selection are collected and packaged to generate a multi-polling observation table.

[0132] It should be noted that the multiple polling observation table is used to centrally record the contact impedance sampling segments and short-time broadband skin conductance signal sampling segments obtained by multiple polling of multiple smart electrodes within a high-density sampling window, thereby providing a corresponding reference within the window for subsequent identification of abnormal electrode states, execution of lead reconstruction switching, and extraction of key risk segments.

[0133] Extract the boundaries of the stable and abrupt contact impedance segments from the multiple polling observation table, and screen out the saturation and missing measurement segments in the short-time broadband EKS signal sampling segments to generate an abnormal electrode status list.

[0134] Furthermore, the observation table is expanded sequentially according to the window number and electrode number. Contact impedance sampling segments are extracted and sliced ​​according to the sampling window. The start and end window numbers of the sampling window intervals where the contact impedance fluctuation amplitude continuously meets the stable threshold are solidified as the contact impedance stable segment boundary. The start and end window numbers of the sampling window intervals where the contact impedance transitions of adjacent sampling windows are concentrated are solidified as the abrupt change segment boundary. Short-time broadband skin conductance signal sampling segments corresponding to the same window number and electrode number are extracted. The amplitude peak sampling points are scanned and continuous peak intervals are aggregated to obtain saturation occupancy. The missing sampling points are scanned and continuous missing intervals are aggregated to obtain missing measurement occupancy. The contact impedance stable segment boundary, abrupt change segment boundary, saturation occupancy and missing measurement occupancy are aggregated according to the electrode number. Electrode numbers with a high abrupt change segment boundary coverage ratio, high saturation occupancy or high missing measurement occupancy are included in the electrode status anomaly list.

[0135] The list of abnormal electrode states is reconstructed and switched, and broadband skin conductance segments are extracted from the graded sampling window, while key risk segments are cropped out.

[0136] Furthermore, the electrode number and corresponding window number are expanded item by item according to the list of abnormal electrode states. It is checked whether the current working electrode pair and the backoff electrode pair contain electrode numbers. If the current working electrode pair contains electrode numbers, the selected electrode pair is switched to the backoff electrode pair. If the backoff electrode pair still contains electrode numbers, the candidate electrode pair that does not contain electrode numbers is selected according to the priority of the lead candidate table to complete the lead reconstruction switch. According to the window number, the start and end time boundaries of the window are located in the graded sampling window and continuous broadband skin electric field segments are extracted by expanding to both sides according to the length of the retained segment. At the same time, according to the start and end time boundaries of the window corresponding to the risk scale entry, the window segment with continuously increasing risk scale is locked in the broadband skin electric field segment and clipped into the critical risk segment.

[0137] It should be noted that the key risk segment refers to the broadband skin conductance segment formed by locking the window segment with continuously increasing risk scale according to the risk scale items within the graded sampling window, which is used for the encapsulation and playback verification of early warning events.

[0138] Key risk segments, electrode pairs before lead reconfiguration switching, electrode pairs after lead reconfiguration switching, window sequence numbers, and window start and end time boundaries are aggregated and encapsulated into a warning event package.

[0139] Furthermore, the continuity of the timestamp field and the sampling point index of key risk segments are verified, and the start and end time boundaries of key risk segments are locked. The key risk segments are matched with their corresponding window numbers and window start and end time boundaries to form segment index entries. The electrode pairs before and after lead reconstruction switching are matched with the field integrity according to the mapping relationship between electrode number and sampling channel, and the lead reconstruction switching window number is added. The segment index entries, electrode pairs before and after lead reconstruction switching, window numbers and window start and end time boundaries are aggregated according to the same window number key, and missing fields are filled in. Event numbers are generated for the aggregated content and the warning event package is encapsulated and output according to the event number.

[0140] Figure 4 The differences in the average percentage of effective sampling window, average coverage of key risk segments, and average integrity of key risk segments are illustrated using a comparative bar chart for fixed electrode pairs, fixed sampling strategies, contact impedance detection and polling sampling schemes, and the complete scheme. As shown in the figure, while the fixed sampling strategy can complete basic monitoring, it is insufficient in terms of coverage and integrity retention of key risk segments. Introducing contact impedance detection and polling sampling increases the percentage of effective sampling window, indicating that electrode-skin contact status monitoring improves the stability of the front-end signal. Further superimposing risk scale-driven sampling parameter switching increases both the coverage and integrity of key risk segments, demonstrating that this invention not only improves signal quality but also more effectively retains key data during periods of rising risk, thus providing a more complete and continuous evidentiary basis for early warning of ventricular arrhythmias.

[0141] Mean percentage of effective sampling windows: This represents the average proportion of sampling windows that meet quality control requirements and can be used normally by the subsequent risk scaling model throughout the entire monitoring process. It primarily reflects whether the front-end signal quality has improved; a higher value indicates that more effective data is available for analysis after contact impedance detection, polling sampling, and quality control sampling window processing.

[0142] Mean coverage of critical risk segments: This represents the average proportion of the length of critical risk segments actually successfully captured by the tiered sampling windows within a known critical risk period, relative to the total length of critical risk segments that should be captured. It primarily reflects the invention's ability to capture critical risk periods; a higher value indicates more sufficient data retention for high-risk phases after the risk scale-driven sampling parameter switching.

[0143] Mean Completeness of Key Risk Fragments: This represents the average proportion of the total length of captured key risk fragments that remain continuous and can be used for subsequent early warning determination and event packet encapsulation. It primarily reflects whether key risk fragments are sufficiently complete after capture. A higher value indicates that not only were risk fragments captured, but there were also fewer missing, saturated, and interrupted segments within them, making them more suitable as a basis for early warning of ventricular arrhythmias.

[0144] This embodiment also provides a flexible wearable patch device, including:

[0145] The host is the core control and processing unit of the device, integrating signal processing, risk scale model calculation, sampling strategy control, and early warning event generation functions;

[0146] Cable sockets are used to stably connect transmission cables, enabling reliable transmission of electrical signals, control commands, and power supply.

[0147] Cable connectors connect the transmission cables to the host unit, completing the connection of signal and power supply paths;

[0148] A hub is a signal relay and integration component, responsible for aggregating multiple electrical signals from the acquisition patch.

[0149] The sensor patch is a flexible wearable main component with a built-in multi-point intelligent electrode array. It fits the skin to complete broadband skin electrical signal acquisition and real-time contact impedance detection.

[0150] In summary, this invention achieves real-time monitoring of electrode-skin contact status by using intelligent sensors for contact impedance detection and polling sampling, thereby suppressing baseline drift and transient saturation and improving signal quality. Furthermore, by using risk scale-driven sampling parameter switching, it optimizes data capture during critical risk periods and shortens the early warning response time.

[0151] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A ventricular arrhythmia risk early warning system, characterized in that, include: The electrode pair sampling module performs contact impedance detection and polling sampling on the multi-point smart electrodes of the flexible wearable patch device, obtains the current working electrode pair and the retreating electrode pair, and solidifies the switching trigger aperture to generate a lead selection list. The sliding window slicing module performs continuous broadband EEG signal acquisition and sliding window slicing based on the lead gating list, and performs baseline drift suppression, power frequency interference suppression and transient saturation segment removal to generate a quality control sampling window; The feature extraction module inputs the quality control sampling window into the risk scale model, performs time-series feature extraction and gating fusion processing, and outputs risk scale entries and budget file switching instruction sets; The sampling window grading module determines the budget grading level based on the risk scale items, and extracts the sliding step size, retained segment length and sampling density from the budget grading switching instruction set to dynamically switch the sampling process and generate graded sampling windows. The lead reconstruction module performs multi-electrode lead re-polling based on graded sampling windows, and performs lead reconstruction switching when an abnormal electrode state is detected. At the same time, it records key risk segments and generates early warning event packages.

2. The ventricular arrhythmia risk early warning system as described in claim 1, characterized in that, The steps for contact impedance detection and polling sampling of the multi-point smart electrodes of the flexible wearable patch device are as follows: Collect the electrode number, electrode position index and sampling channel mapping relationship of multi-point smart electrodes, and compile them into a polling queue list according to the polling order; Based on the polling queue list, the contact impedance detection link of the multi-point smart electrode is activated, and short-term polling sampling is performed. At the same time, the interference occupied field and the missing measurement occupied field of the sampling window are retained to generate an impedance polling table. By combining the impedance polling table with the electrode position index, electrode positions are paired and arranged to obtain a list of candidate electrode pairs. Paired polling sampling is then performed, and the stable contact impedance segment is locked to generate a candidate lead table.

3. The ventricular arrhythmia risk early warning system as described in claim 1, characterized in that, The steps for generating the lead gating list are as follows: Based on the lead candidate list, the candidate electrode pair list is eliminated and prioritized to generate the current working electrode pair and the back-off electrode pair. The interference occupancy change, the missing measurement occupancy change and the contact impedance sudden change segment boundary are collected to form the switching trigger calibrator. Based on the current working electrode pair and the back-off electrode pair, the switching trigger aperture is fixed as the electrode switching trigger aperture, and a lead selection list is generated.

4. The ventricular arrhythmia risk early warning system as described in claim 1, characterized in that, The steps for acquiring continuous broadband electrodermal signal and performing sliding window slicing based on the lead gating list are as follows: Based on the lead gating list, the sampling channel is verified, the gain and range are configured, and the sampling clock is aligned. Continuous broadband EKS signals are acquired to generate an EKS signal sampling stream. The skin conduction signal sampling stream is sliced ​​according to a fixed sliding window length and sliding window step size, and the contact impedance detection state and body motion disturbance state are collected and integrated into an aligned sampling window.

5. The ventricular arrhythmia risk early warning system as described in claim 4, characterized in that, The steps for generating the quality control sampling window are as follows: Slow-varying component tracking, slow-varying component subtraction, and window boundary continuity adjustment are performed on the aligned sampling window to generate a baseline-cleaned sampling window; The baseline clean sampling window is suppressed in the power frequency center band, and transient saturation segments in the amplitude peak section are removed to generate a quality control sampling window.

6. The ventricular arrhythmia risk early warning system as described in claim 1, characterized in that, The steps for outputting risk scale entries and budget file switching instructions are as follows: The quality control sampling window is checked for consistency between window number and time boundary, sampling point index range is aligned, range is unified and dynamic range is clipped, and missing measurement fields are extracted to generate a scale input package. The risk scaling model includes an input embedding layer, a temporal feature extraction layer, a gated fusion layer, and a risk mapping layer; In the input embedding layer, channel rearrangement and fragment alignment slicing are performed on the scale input packet, and the contact impedance detection state and the bulk motion disturbance state are spliced ​​into an evidence companion sequence to generate a window-by-window embedding sequence. In the temporal feature extraction layer, local segment extraction, cross-window continuous segment splicing and mutation segment aggregation are performed on the window-by-window embedded sequence to obtain rhythmic unstable segments and morphological mutation segments, which are then integrated into a temporal representation sequence. In the gated fusion layer, based on the contact impedance detection state, the volumetric disturbance state, and the missing measurement occupied field, the time series representation sequence is gated and weighted down to form a gated fusion representation sequence. In the risk mapping layer, the gating fusion representation sequence is mapped window by window to risk scale entries, and budget level mapping is arranged to generate a budget level switching instruction set.

7. The ventricular arrhythmia risk early warning system as described in claim 1, characterized in that, The steps for generating the graded sampling window are as follows: Pair the window number and start / end time boundaries of the risk scale entries and budget file switching instruction set, and fill in the adjacent window numbers for mismatched window numbers, and output the window alignment table. The budget level is determined in the window alignment table according to the window number, adjacent and consistent levels are continuously merged and the fallback and back check is performed to generate the level effective table. Based on the gear activation table, extract the sliding step size, retained segment length and sampling density from the budget gear switching instruction set, check the switching safety window, and generate a sampling switching configuration table; Based on the sampling switching configuration table, continuous broadband electrodermal signals are acquired, and sliding step size replacement, retained segment length replacement, and sampling density replacement are performed. At the same time, boundary alignment verification is performed to generate graded sampling windows.

8. The ventricular arrhythmia risk early warning system as described in claim 7, characterized in that, The steps for multi-electrode lead re-polling based on graded sampling windows are as follows: The sampling windows are filtered by window number to obtain the set of window numbers in the high-density range of the budget level, and then combined with the window start and end time boundaries and the current working electrode pair to form a multi-polling list. Based on the multi-polling list, multiple smart electrodes are activated sequentially within the corresponding window start and end time boundaries, and contact impedance detection and short-time broadband skin conduction signal sampling are performed to generate a multi-polling observation table.

9. The ventricular arrhythmia risk early warning system as described in claim 8, characterized in that, The steps for generating the early warning event package are as follows: Extract the boundaries of the stable and abrupt contact impedance segments from the multiple polling observation table, and screen out the saturation and missing measurement segments in the short-time broadband EKS signal sampling segments to generate an abnormal electrode status list. The list of abnormal electrode states is reconstructed and switched, and broadband skin conductance segments are extracted from the graded sampling window, while key risk segments are cropped out. Key risk segments, electrode pairs before lead reconfiguration switching, electrode pairs after lead reconfiguration switching, window sequence numbers, and window start and end time boundaries are aggregated and encapsulated into a warning event package.

10. A flexible wearable patch device for a ventricular arrhythmia risk early warning system as described in any one of claims 1 to 9, characterized in that, include: The host is the core control and processing unit of the device, integrating signal processing, risk scale model calculation, sampling strategy control, and early warning event generation functions; Cable sockets are used to stably connect transmission cables, enabling reliable transmission of electrical signals, control commands, and power supply. Cable connectors connect the transmission cables to the host unit, completing the connection of signal and power supply paths; A hub is a signal relay and integration component, responsible for aggregating multiple electrical signals from the acquisition patch. The sensor patch is a flexible wearable main component with a built-in multi-point intelligent electrode array. It fits the skin to complete broadband skin electrical signal acquisition and real-time contact impedance detection.

Images