A multi-dimensional electric defibrillation heart rhythm rapid identification method

Abstract

The present application relates to the technical field of heart rhythm rapid identification, and discloses a multi-dimensional electric defibrillation heart rhythm rapid identification method, comprising: pre-processing an electrocardiosignal and intercepting a first time window segment; when it is determined that the first time window segment belongs to an abnormal heart rhythm, extracting a heart rhythm cycle parameter and a confidence factor; extracting multi-dimensional features of the first time window segment, performing mask reconstruction on the multi-dimensional features according to the confidence factor, and generating a first weighted vector; inputting the first weighted vector into a classification model for reasoning, and outputting a first state label; and through extraction of a quality confidence factor in a preliminary screening stage, and dynamic mask reconstruction of the extracted multi-dimensional features by using the factor, the system can automatically reduce the weight of time domain features susceptible to interference and compensate nonlinear features according to the confidence, so that a later-stage classification model can make a decision based on an adaptively reconstructed feature space.

Description

Technical Field

[0001] This invention relates to the field of rapid heart rhythm identification technology, specifically a multi-dimensional defibrillable rapid heart rhythm identification method. Background Technology

[0002] In the clinical emergency care of cardiac arrest and the application of automated external defibrillators (AEDs), accurate and rapid identification of shockable rhythms is a core prerequisite for improving patient survival rates. Due to the complex emergency scene environment and significant individual electrophysiological differences among patients, the collected electrocardiogram signals are often accompanied by strong electromyographic artifacts, baseline drift, and electrode contact noise, which places high demands on the robustness of the rhythm recognition algorithm.

[0003] Currently, conventional heart rhythm recognition technologies in this field generally adopt a sequentially cascaded unidirectional pipeline architecture. The typical processing flow usually involves first using heuristic algorithms such as Pan-Tompkins for preliminary QRS wave morphological screening, then extracting static time-domain, frequency-domain, or nonlinear features within a fixed time window, and feeding them into machine learning models such as support vector machines for binary classification output. In order to reduce the false positive rate caused by a single time window, existing solutions often directly introduce a sliding window with a fixed time step for secondary sampling and verification after the initial determination that defibrillation is possible.

[0004] However, this traditional decoupled pipeline architecture reveals deep logical flaws when dealing with extremely complex emergency situations. Throughout the data flow, the front-end morphological screening and the back-end machine learning feature extraction are isolated processes. The signal degradation perceived by the front-end when evaluating waveforms cannot be fed forward, causing the feature extraction module to blindly and statically calculate fixed-dimensional features regardless of whether the underlying ECG signal has suffered localized high-frequency interference. This results in the time-domain parameters, which are extremely sensitive to noise, contaminating the classifier's input space. Furthermore, the traditional secondary verification mechanism suffers from limitations in the time window. The shift strategy lacks synchronization with physiological rhythms. Because the cardiac cycle of patients is in a disordered or abnormally high-frequency state during ventricular fibrillation or ventricular tachycardia, fixed time step sliding will cut off the core physiological waveform, causing sample distortion and cycle fragmentation during secondary feature capture. Furthermore, at the final fault tolerance level, existing technologies usually rely only on the dead logical intersection of the two classification labels, completely stripping away the constraints of the underlying physical parameters. Once the secondary sampling window encounters occasional instantaneous poor contact noise, the classification model will output contradictory labels, thus directly rejecting the electric shock, causing fatal missed detections and misjudgments at critical rescue nodes. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a multi-dimensional method for rapid identification of defibrillable heart rhythms. It solves the problems of blind feature extraction and noise contamination in traditional unidirectional pipeline architectures, rigid time windows that easily truncate continuous physiological waveforms, and the lack of cross-dimensional physical energy fault tolerance arbitration that leads to missed detections due to occasional interference.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for rapid identification of multi-dimensional defibrillable cardiac rhythms, comprising: S1: Preprocess the electrocardiogram signal and extract the first time window segment; S2: When the preliminary morphological waveform detection determines that the first time window segment belongs to an abnormal heart rhythm, extract the heart rhythm cycle parameter and confidence factor; S3: Extract the multi-dimensional features of the first time window segment, and perform mask reconstruction on the multi-dimensional features according to the confidence factor to generate a first weighted vector; S4: Input the first weighted vector into the classification model for inference and output the first state label; S5: When the first status label indicates a defibrillable rhythm or when the first status label indicates a non-defibrillable rhythm and the confidence factor is less than a preset confidence threshold, the sliding step size is determined according to the heart rhythm cycle parameter, and a second time window segment is extracted from the electrocardiogram signal according to the sliding step size. S6: Extract the multi-dimensional features of the second time window segment, perform mask reconstruction based on the confidence factor to generate a second weighted vector, input the second weighted vector into the classification model for inference, and output the second state label; S7: Calculate the physical energy deviation between the first time window segment and the second time window segment, and output the defibrillation decision result by combining the first state label, the second state label and the physical energy deviation.

[0007] Preferably, the preprocessing of the electrocardiogram signal and the extraction of a first time window segment includes: The acquired raw ECG signal was sequentially subjected to bandpass filtering and power frequency notch filtering, and baseline drift components were removed to generate a standardized ECG signal. Based on a preset time window length, continuous discrete sample data are extracted from the standardized electrocardiogram signal to construct the first time window segment.

[0008] Preferably, determining that the first time window segment belongs to an abnormal heart rhythm includes: Morphological waveform detection is performed on the first time window segment; If no wave group matching the preset normal heart rhythm index is detected within the first time window segment, the first time window segment is determined to belong to the abnormal heart rhythm.

[0009] Preferably, the extraction of cardiac rhythm cycle parameters and confidence factors includes: Calculate the expected time interval between adjacent characteristic peak values ​​within the first time window segment, or extract the reciprocal of the dominant frequency of the spectrum of the first time window segment as the cardiac rhythm parameter; The confidence factor characterizing signal quality is calculated based on the moving integral peak and the root mean square of the background noise within the first time window segment.

[0010] Preferably, the step of extracting multi-dimensional features of the first time window segment and performing mask reconstruction on the multi-dimensional features based on the confidence factor to generate a first weighted vector includes: Extract the time-domain features, frequency-domain features, nonlinear features, and time-frequency-domain features of the first time window segment, and construct an initial feature set after dimensionality reduction; Generate a corresponding set of feature mask weights based on the confidence factor; The first weighted vector is obtained by performing element-wise multiplication between the initial feature set and the feature mask weight set.

[0011] Preferably, generating the corresponding feature mask weight set based on the confidence factor includes: when the confidence factor indicates a decrease in signal quality, reducing the mask weight corresponding to the time-domain feature in the feature mask weight set, and increasing the mask weight corresponding to the nonlinear feature.

[0012] Preferably, the classification model is a support vector machine model, which uses radial basis functions to map the input features to a high-dimensional space to define the boundary between defibrillable and non-defibrillable heart rhythms.

[0013] Preferably, the step of determining the sliding step size based on the cardiac rhythm parameter and extracting a second time window segment from the electrocardiogram signal according to the sliding step size includes: The sliding step size is obtained by multiplying the heart rhythm cycle parameter by a preset step size ratio coefficient and adding a time compensation constant. The sliding step size is converted into a discrete sampling point shift amount, and a segment with a length equal to that of the first time window segment is extracted from the electrocardiogram signal by shifting it backward, which is then used as the second time window segment.

[0014] Preferably, calculating the physical energy deviation between the first time window segment and the second time window segment includes: Calculate the sum of squares of the amplitudes of all sampling points within the first time window segment and the sum of squares of the amplitudes of all sampling points within the second time window segment, respectively; The physical energy deviation is obtained by subtracting the sum of squares of the second time window segment from the sum of squares of the first time window segment and taking the absolute value.

[0015] Preferably, the step of combining the first state label, the second state label, and the physical energy deviation to output a defibrillation decision includes: When both the first status label and the second status label indicate a defibrillable rhythm, the defibrillation decision result recommending defibrillation is output. When both the first status label and the second status label indicate a non-shockable rhythm, the defibrillation decision result of refusing defibrillation is output. When the indications of the first status label and the second status label conflict, the physical energy deviation is compared with a preset energy tolerance threshold. If the physical energy deviation is less than or equal to the preset energy tolerance threshold, the determination of the first state label is maintained when the first state label is a shockable defibrillable rhythm, and the determination of the second state label is adopted when the first state label is a non-shockable defibrillable rhythm, and the defibrillation decision result of the recommended defibrillation is output. If the physical energy deviation is greater than the preset energy tolerance threshold, the defibrillation decision result of rejecting defibrillation is output.

[0016] This invention provides a multi-dimensional method for rapid identification of defibrillable heart rhythms. It has the following beneficial effects: 1. This invention extracts a quality confidence factor in the initial screening stage and uses this factor to dynamically reconstruct the extracted multi-dimensional features using a mask. This breaks the limitation of traditional classification models that statically receive fixed feature vectors. When the underlying signal is severely affected by electromyography artifacts or baseline disturbances, the system can automatically reduce the weight of easily disturbed time-domain features and compensate for nonlinear features based on the confidence level. This allows the subsequent classification model to make decisions based on the adaptively reconstructed feature space, thus eliminating the interference of high-frequency noise on sensitive feature components from a mechanism perspective.

[0017] 2. This invention extracts the cardiac rhythm cycle parameter of the initial ECG segment and uses this parameter to dynamically calculate the sliding step size of the time window for secondary verification according to a linear affine relationship. This eliminates the physical defect of conventional fixed step size truncation, which easily cuts off the core cardiac cycle. Through a non-equal length windowing mechanism driven by underlying physiological characteristics, it ensures that secondary sampling can accurately capture continuous waveforms with complete morphological significance, whether in extremely high-frequency ventricular fibrillation or relatively regular ventricular tachycardia, thus guaranteeing the physical validity of the secondary inference data source.

[0018] 3. This invention constructs a collaborative state machine and introduces physical energy deviation as an arbitration condition. When there is a conflict between the inference labels of the two models, it compares the degree of energy fluctuation of the underlying signal segments of the two time windows across dimensions. This can identify the classification jump of the algorithm layer caused by occasional local noise. When it is confirmed that the physical energy has not actually changed, the latching mechanism is activated to maintain the original judgment, eliminating the hidden danger of false negatives caused by simply relying on the mathematical classifier, and establishing a dual verification mechanism across the algorithm layer and the physical layer.

[0019] 4. This invention separates and feeds forward the periodic parameters and confidence factors generated in the pre-process morphological detection, so that the heuristic rules, which were originally only used as independent screening tools, are substantially intertwined with the post-process machine learning model. By directly reusing the byproducts of the front-end signal evaluation process, it is used as the driving parameters to control the back-end feature mask and time axis cutting. This not only avoids the repetitive calculations caused by stacking independent modules, but also breaks the drawbacks of the linear pipeline with unidirectional decoupling of each step.

[0020] 5. This invention extracts candidate features in multiple time and frequency domains and implements dynamic dimensionality reduction. Combined with the above-mentioned adaptive windowing and state machine backtracking mechanism, it constructs a recognition system with deep mutual verification. In the entire decision flow, the system does not blindly follow the data input of a single dimension or the output result of a single model. Instead, it forces the front-end physical parameters, the mid-end multi-domain masking mechanism and the back-end joint state machine to constrain each other, so that the overall system architecture has logical cohesion and can output real and reliable defibrillation suggestions in complex and ever-changing clinical electrophysiological emergency environments. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the identification method architecture of the present invention; Figure 2 This is one of the schematic diagrams of the identification method of the present invention; Figure 3 This is a second schematic diagram of the identification method of the present invention; Figure 4 This is the third schematic diagram of the identification method of the present invention; Figure 5 This is the fourth schematic diagram of the identification method of the present invention. Detailed Implementation

[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] See attached document Figure 1- Appendix Figure 5 This invention provides a method for rapid identification of multi-dimensional defibrillable heart rhythms, including: S1: Preprocess the ECG signal and extract a first time window segment, including: sequentially performing bandpass filtering and power frequency notch filtering on the acquired raw ECG signal, and removing baseline drift components to generate a standardized ECG signal; extracting continuous discrete sample data from the standardized ECG signal according to a preset time window length to construct the first time window segment. Specifically, the acquired raw ECG signal is sequentially subjected to bandpass filtering and power frequency notch filtering to remove interference components outside the core frequency band. Then, median filtering or polynomial fitting algorithms are used to calculate the low-frequency baseline drift component, which is then subtracted from the raw ECG signal to generate a standardized ECG signal. Based on a pre-set time window length and system sampling rate Determine the number of discrete sample points included within the time window. Continuous discrete sample data are extracted from standardized electrocardiogram signals in time sequence to construct the first time window segment. Its time-domain performance satisfies ,in .

[0024] S2: When the first time window segment is determined to be an abnormal heart rhythm, extract the heart rhythm cycle parameter and confidence factor; determining that the first time window segment is an abnormal heart rhythm includes: performing morphological waveform detection on the first time window segment, and when no wave group conforming to the preset normal heart rhythm index is detected in the first time window segment, the first time window segment is determined to be an abnormal heart rhythm; extracting the heart rhythm cycle parameter and confidence factor includes: calculating the expected value of the time interval between adjacent characteristic wave peaks in the first time window segment, or extracting the reciprocal of the dominant frequency of the spectrum of the first time window segment as the heart rhythm cycle parameter, and calculating and generating a confidence factor characterizing the signal quality based on the moving integral peak and the root mean square of the background noise in the first time window segment; Specifically, after obtaining the first time window segment, the system enters the initial waveform morphology screening stage, which involves processing the first time window segment... Morphological waveform detection is performed. When a waveform cluster with frequency, amplitude, and width all within the preset normal sinus rhythm threshold range is detected in the current segment, the current ECG rhythm is directly determined to be non-defibrillable, and the current detection cycle is terminated. If no waveform cluster meeting the above normal rhythm indicators is detected in the first time window segment, the first time window segment is determined to be an abnormal rhythm, and the system triggers the state parameter extraction process. The extracted rhythm cycle parameter is denoted as... The confidence factor is denoted as The expected value of the time interval between adjacent characteristic waves (such as the equivalent R-wave peak) within the first time window segment is calculated as... If the signal within a segment exhibits an extremely high-frequency disordered pattern, then the reciprocal of the dominant frequency of the segment's spectrum is extracted as... Simultaneously, based on the signal motion integral peak and the root mean square of the background noise within the segment, a confidence factor characterizing the fundamental physical quality of the signal is calculated and generated. The value is normalized and constrained to the range of [0,1]. The lower the value, the more the current segment is affected by electromyographic artifacts or high-frequency baseline perturbations.

[0025] S3: Extract multi-dimensional features from the first time window segment, and perform mask reconstruction on the multi-dimensional features based on the confidence factor to generate the first weighted vector. This includes: extracting the time-domain features, frequency-domain features, nonlinear features, and time-frequency-domain features of the first time window segment, reducing their dimensionality to form an initial feature set, generating a corresponding feature mask weight set based on the confidence factor, and performing element-wise multiplication of the initial feature set and the feature mask weight set to obtain the first weighted vector. Generating the corresponding feature mask weight set based on the confidence factor includes: when the confidence factor indicates a decrease in signal quality, reducing the mask weights corresponding to the time-domain features in the feature mask weight set and increasing the mask weights corresponding to the nonlinear features.

[0026] Specifically, multi-dimensional candidate features are extracted from the first time window segment, including time domain, frequency domain, nonlinearity, and time-frequency domain (specifically, these may include QRS wave average amplitude, zero-crossing rate, spectral entropy, power ratio of each frequency band, sample entropy, and variance of wavelet decomposition coefficients). After dimensionality reduction using a feature selection algorithm, these multi-dimensional features constitute a subset of the desired features. The initial feature set of each dimension is denoted as a column vector. The system is based on the confidence factor input in step S2. Dynamically generate a set of feature mask weights to match it, denoted as a column vector. The specific weight modulation rule is as follows: when When the signal quality deteriorates, the system reduces the mask weight components corresponding to time-domain features with weak anti-interference capabilities in the feature mask weight set, and correspondingly increases the mask weight components corresponding to nonlinear features with strong anti-noise capabilities. Then, the initial feature set... With feature mask weight set Element-wise multiplication (i.e., Hadamard product) is performed to reconstruct the feature space using a mask, resulting in the first weighted vector. ; S4: Input the first weighted vector into the classification model for inference and output the first state label. The classification model is a support vector machine model. The support vector machine model uses radial basis functions to map the input features to a high-dimensional space to divide the judgment boundary between shockable and non-shockable heart rhythms. Specifically, the classification model employs a Support Vector Machine (SVM) model pre-trained based on an ECG database. This model uses a Radial Basis Function (RBF) as the kernel function to map the input features to a high-dimensional space to delineate the decision boundaries. The first weighted vector... Input the classification model for forward inference and output the first state label. If the model determines that defibrillation is not possible, If the heart rhythm is determined to be shockable, .

[0027] S5: When the first status label indicates a defibrillable rhythm or when the first status label indicates a non-defibrillable rhythm and the confidence factor is less than the preset confidence threshold, the sliding step size is determined according to the rhythm cycle parameter, and a second time window segment is extracted from the ECG signal according to the sliding step size, including: multiplying the rhythm cycle parameter by a preset step size ratio coefficient and adding a time compensation constant to obtain the sliding step size, converting the sliding step size into a discrete sampling point translation amount, and translating backward from the ECG signal to extract a segment with a length equal to that of the first time window segment as the second time window segment; Specifically, set a preset reliability threshold. When the first state label At time, or first state label And the extracted confidence factor At that time, the adaptive sliding check mechanism is triggered, and the system retrieves the heart rate cycle parameter extracted in step S2. Calculate the sliding step size according to the linear affine relation. ; In the formula, This is the preset step size ratio coefficient. The calculated sliding step size is set to a preset time compensation constant. Multiply by the system sampling rate Then round down to the nearest integer to convert it into a discrete sampling point shift. From standardized electrocardiogram signals Translate backward from center For each sampling point, a continuous segment of length equal to that of the first time window segment is extracted and used as the second time window segment. ,satisfy ,in This ensures that the secondary sampling action dynamically scales up and down based on the underlying cardiac cycle, thus guaranteeing the physiological continuity of the captured waveform.

[0028] S6: Extract multi-dimensional features of the second time window segment, reconstruct the mask based on the confidence factor to generate the second weighted vector, input the second weighted vector into the classification model for inference, and output the second state label; Specifically, for the acquired second time window segment Repeatedly perform feature reconstruction and inference operations, that is, extract multi-dimensional features of the second segment, based on the same confidence factor. The same mask weights are generated and multiplied element-wise to obtain a second weighted vector. This second weighted vector is then input into the same classification model for secondary inference, and a second state label is output. .

[0029] S7: Calculate the physical energy deviation between the first time window segment and the second time window segment, and output the defibrillation decision result by combining the first state label, the second state label and the physical energy deviation; The physical energy deviation between the first time window segment and the second time window segment is calculated by: calculating the sum of squares of the amplitudes of all sampling points in the first time window segment and the sum of squares of the amplitudes of all sampling points in the second time window segment respectively; and taking the absolute value of the difference between the sum of squares of the second time window segment and the sum of squares of the first time window segment to obtain the physical energy deviation. Combining the first state label, the second state label, and the physical energy deviation, the system outputs a defibrillation decision, including: when both the first and second state labels indicate a defibrillable rhythm, a defibrillation decision recommending defibrillation is output; when both the first and second state labels indicate a non-defibrillable rhythm, a defibrillation decision refusing defibrillation is output; when the indications of the first and second state labels conflict, the physical energy deviation is compared with a preset energy tolerance threshold; if the physical energy deviation is less than or equal to the preset energy tolerance threshold, the first state label's determination is maintained when the first state label indicates a defibrillable rhythm; the second state label's determination is adopted when the first state label indicates a non-defibrillable rhythm, and a defibrillation decision recommending defibrillation is output; if the physical energy deviation is greater than the preset energy tolerance threshold, a defibrillation decision refusing defibrillation is output. Specifically, a collaborative state machine is constructed to handle the final arbitration between the joint labeling results and the measured physical energy: when the first state label... With the second state label All were indicated as defibrillable rhythms (i.e., and When this happens, the state machine outputs a defibrillation decision suggesting defibrillation. When the first state label With the second state label All were indicated as non-defibrillable rhythms (i.e.) and When this happens, the state machine outputs a defibrillation decision that refuses to defibrillate; When there is a conflict between the preceding and following tags (i.e.) and ,or and When the energy cross-validation logic is triggered, the first time window segment is calculated respectively. With the second time window segment The physical energy deviation is obtained by summing the squares of the amplitudes of all discrete sampling points within the range, taking the difference between the two, and then taking the absolute value. : ; Physical energy deviation Compared with the preset energy tolerance threshold Perform a comparison, if This indicates that the underlying ECG physical energy did not change substantially between the two time windows. The system determines that the variation in the second state label is caused by occasional local noise. At this point, the state machine activates the live-state latch mechanism, maintains the conclusion of the first state label, and outputs a defibrillation decision recommending defibrillation. If the conflict type is... , The system determines The missed reports are caused by low confidence (i.e., strong artifact interference), and the state machine adopts the data obtained under the interference-free state. Both conclusions output defibrillation recommendations. like This indicates a substantial change in the electrocardiogram signal, and the state machine rejects the defibrillation recommendation, outputting a defibrillation decision of refusal.

[0030] In summary, this invention breaks away from the isolated operation mode of linear decoupling of each step in traditional ECG recognition algorithms. It constructs a deeply collaborative dynamic closed-loop detection architecture. By using the physiological cycle and quality confidence extracted from the previous morphological detection as forward feedback variables, it performs adaptive mask reconstruction on the multi-dimensional feature space, enabling the classification model to dynamically adjust the feature perception weights in complex noise environments. At the same time, based on the underlying cardiac cycle parameters driving non-equal length sliding sampling, it ensures the complete capture of physiological waveforms in the secondary confirmation stage. In the system decision closing stage, the collaborative state machine breaks away from the logical classification at the pure algorithm level. When encountering conflicts between front and back labels, it backtracks across dimensions and compares the physical energy deviation of the original signal segments, realizing dual cross-validation of algorithm state + underlying energy. Through the deep interweaving of parameters and the mutual nesting of logic, heuristic rules, machine learning models and physical features are tightly integrated. Under the premise of adapting to the computing power of embedded hardware, it constructs an extremely high architectural robustness and error prevention barrier.

[0031] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for rapid identification of multi-dimensional defibrillable cardiac rhythms, characterized in that, include: S1: Preprocess the electrocardiogram signal and extract the first time window segment; S2: When the preliminary morphological waveform detection determines that the first time window segment belongs to an abnormal heart rhythm, extract the heart rhythm cycle parameter and confidence factor; S3: Extract the multi-dimensional features of the first time window segment, and perform mask reconstruction on the multi-dimensional features according to the confidence factor to generate a first weighted vector; S4: Input the first weighted vector into the classification model for inference and output the first state label; S5: When the first status label indicates a defibrillable rhythm or when the first status label indicates a non-defibrillable rhythm and the confidence factor is less than a preset confidence threshold, the sliding step size is determined according to the heart rhythm cycle parameter, and a second time window segment is extracted from the electrocardiogram signal according to the sliding step size. S6: Extract the multi-dimensional features of the second time window segment, perform mask reconstruction based on the confidence factor to generate a second weighted vector, input the second weighted vector into the classification model for inference, and output the second state label; S7: Calculate the physical energy deviation between the first time window segment and the second time window segment, and output the defibrillation decision result by combining the first state label, the second state label and the physical energy deviation.

2. The method for rapid identification of multi-dimensional defibrillable cardiac rhythm according to claim 1, characterized in that, The preprocessing of the electrocardiogram signal and the extraction of a first time window segment include: The acquired raw ECG signal was sequentially subjected to bandpass filtering and power frequency notch filtering, and baseline drift components were removed to generate a standardized ECG signal. Based on a preset time window length, continuous discrete sample data are extracted from the standardized electrocardiogram signal to construct the first time window segment.

3. The method for rapid identification of multi-dimensional defibrillable cardiac rhythm according to claim 1, characterized in that, The determination that the first time window segment belongs to an abnormal heart rhythm includes: Morphological waveform detection is performed on the first time window segment; If no wave group matching the preset normal heart rhythm index is detected within the first time window segment, the first time window segment is determined to belong to the abnormal heart rhythm.

4. The method for rapid identification of multi-dimensional defibrillable heart rhythm according to claim 3, characterized in that, The extraction of cardiac rhythm parameters and confidence factors includes: Calculate the expected time interval between adjacent characteristic peak values ​​within the first time window segment, or extract the reciprocal of the dominant frequency of the spectrum of the first time window segment as the cardiac rhythm parameter; The confidence factor characterizing signal quality is calculated based on the moving integral peak and the root mean square of the background noise within the first time window segment.

5. The method for rapid identification of multi-dimensional defibrillable cardiac rhythm according to claim 1, characterized in that, The step of extracting multi-dimensional features from the first time window segment and performing mask reconstruction on the multi-dimensional features based on the confidence factor to generate a first weighted vector includes: Extract the time-domain features, frequency-domain features, nonlinear features, and time-frequency-domain features of the first time window segment, and construct an initial feature set after dimensionality reduction; Generate a corresponding set of feature mask weights based on the confidence factor; The first weighted vector is obtained by performing element-wise multiplication between the initial feature set and the feature mask weight set.

6. The method for rapid identification of multi-dimensional defibrillable cardiac rhythm according to claim 5, characterized in that, The step of generating a corresponding feature mask weight set based on the confidence factor includes: when the confidence factor indicates a decrease in signal quality, reducing the mask weight corresponding to the time-domain feature in the feature mask weight set, and increasing the mask weight corresponding to the nonlinear feature.

7. The method for rapid identification of multi-dimensional defibrillable cardiac rhythm according to claim 1, characterized in that, The classification model is a support vector machine model, which uses radial basis functions to map the input features to a high-dimensional space to define the boundary between defibrillable and non-defibrillable heart rhythms.

8. The method for rapid identification of multi-dimensional defibrillable cardiac rhythm according to claim 1, characterized in that, The step of determining the sliding step size based on the cardiac rhythm parameters and extracting a second time window segment from the electrocardiogram signal according to the sliding step size includes: The sliding step size is obtained by multiplying the heart rhythm cycle parameter by a preset step size ratio coefficient and adding a time compensation constant. The sliding step size is converted into a discrete sampling point shift amount, and a segment with a length equal to that of the first time window segment is extracted from the electrocardiogram signal by shifting it backward, which is then used as the second time window segment.

9. The method for rapid identification of multi-dimensional defibrillable cardiac rhythm according to claim 1, characterized in that, The calculation of the physical energy deviation between the first time window segment and the second time window segment includes: Calculate the sum of squares of the amplitudes of all sampling points within the first time window segment and the sum of squares of the amplitudes of all sampling points within the second time window segment, respectively; The physical energy deviation is obtained by subtracting the sum of squares of the second time window segment from the sum of squares of the first time window segment and taking the absolute value.

10. The method for rapid identification of multi-dimensional defibrillable cardiac rhythm according to claim 9, characterized in that, The step of combining the first state label, the second state label, and the physical energy deviation to output a defibrillation decision includes: When both the first status label and the second status label indicate a defibrillable rhythm, the defibrillation decision result recommending defibrillation is output. When both the first status label and the second status label indicate a non-shockable rhythm, the defibrillation decision result of refusing defibrillation is output. When the indications of the first status label and the second status label conflict, the physical energy deviation is compared with a preset energy tolerance threshold. If the physical energy deviation is less than or equal to the preset energy tolerance threshold, the determination of the first state label is maintained when the first state label is a shockable defibrillable rhythm, and the determination of the second state label is adopted when the first state label is a non-shockable defibrillable rhythm, and the defibrillation decision result of the recommended defibrillation is output. If the physical energy deviation is greater than the preset energy tolerance threshold, the defibrillation decision result of rejecting defibrillation is output.

Images