46 results about "Arrhythmia detection" patented technology

The invention discloses a similarity measurement-based few-sample electrocardiosignal classification method. The method comprises the following steps of: acquiring an electrocardiosignal sequence from a public library and dividing the electrocardiosignal sequence into a support and query set according to a few-sample format; performing normalization processing and zero filling operation on the obtained sequence; inputting the sequence data with the uniform length into a parameter-shared one-dimensional convolutional neural network to extract a time sequence embedded feature vector; after the vectors are spliced and multiplied, weighting the vectors into weighted features through an attention network; inputting the weighted features into a multi-layer perceptron to calculate a similarity score, and training and fixing a neural network by using positive and negative sample pairs; and calculating the maximum similarity between the query sample and the support set, and outputting a prediction category. According to the method, the electrocardiosignal labeling cost can be remarkably reduced, the accuracy and efficiency of abnormal heart rhythm detection can be improved, dependence on large-scale labeling data is reduced, the practicability and expandability of electrocardiosignal classification are improved, and the method is applied to the field of medical signal processing and has important significance in the aspects of abnormal heart rhythm detection, disease diagnosis, wearable equipment application and the like.

The invention belongs to the field of non-contact vital sign detection, and particularly relates to a millimeter wave radar arrhythmia detection method based on particle swarm optimization variational mode decomposition and deep learning, and the method specifically comprises the steps: S1, obtaining a human chest micro-motion signal through a millimeter wave radar, performing static clutter filtering, phase extraction, unwrapping and detrending processing on the radar echo signal to obtain a chest displacement signal containing heartbeat and breathing information; s2, aiming at the thoracic cavity displacement signal, constructing a variational mode decomposition model, and carrying out adaptive optimization on a mode number and a penalty factor through a particle swarm optimization algorithm to obtain an optimal decomposition parameter and complete signal decomposition; s3, according to the center frequency and the energy distribution characteristics of each modal component, screening the modal components in the heartbeat frequency range and reconstructing the modal components to obtain heartbeat characteristic signals representing heart mechanical activities; s4, carrying out time sequence segmentation on the heartbeat characteristic signals, extracting local heart beat morphological characteristics by utilizing a convolutional neural network, and carrying out modeling on a long-time rhythm dependency relationship of the heartbeat signals in combination with a time sequence modeling network based on an attention mechanism to obtain heart rhythm depth characteristic representation; and S5, inputting the heart rhythm depth features into a heart rhythm discrimination model, analyzing the heart rhythm state of the detected person, and outputting an arrhythmia detection result. According to the method, adaptive selection of variational mode decomposition parameters is realized by introducing a particle swarm optimization mechanism, heartbeat signals and respiration and motion interference components are effectively separated, modeling is carried out on rhythm characteristics in combination with a deep learning model, and non-contact detection of arrhythmia is realized. The method does not need to wear an electrode or contact a human body, has the advantages of strong anti-interference capability, good adaptability and high detection precision, and has a good application prospect in the fields of heart rhythm health monitoring, disease screening and the like.

The invention discloses a form fidelity consistency constraint-based arrhythmia real-time detection method, which comprises the following steps of: S1, preprocessing continuous electrocardiosignals, and dividing the continuous electrocardiosignals into overlapped sliding time windows; s2, inputting each time window into an electrocardiosignal encoder obtained through joint training to obtain a potential representation vector of the time window; the joint training is specifically as follows: in a training stage, a supervised classification signal, a comparison consistency signal and a morphological fidelity signal are simultaneously utilized to optimize parameters of the encoder; s3, inputting the potential representation vector into a classifier, and outputting the category probability that the time window belongs to each arrhythmia category; and S4, based on the category probability, outputting a real-time detection result by adopting an online judgment strategy of threshold hysteresis and cross-window consistency. According to the method, stable identification under dynamic interference, precise reservation of clinical key waveform details and efficient real-time deployment of the edge end can be synchronously realized. The invention further provides an arrhythmia real-time detection system and device based on the morphological fidelity consistency constraint.

The invention discloses an arrhythmia detection method based on a pulse neural network. The method comprises the following steps: performing incremental modulation pulse coding and pooling operation on a target electrocardiosignal to obtain a corresponding pulse sequence; and inputting the pulse sequence into a trained multi-level pulse neural network classification model to obtain an arrhythmia detection result. Wherein the multi-level pulse neural network classification model comprises a weight sharing layer and a plurality of cascaded classifiers, the weight sharing layer is used for extracting electrocardiosignal feature information from the pulse sequence, and the plurality of classifiers are used for obtaining an arrhythmia detection result based on the electrocardiosignal feature information. According to the method, on the premise that the accuracy is equivalent, the classification result is more refined, and the requirements for hardware storage resources and computing resources are reduced.

The invention discloses a smart watch arrhythmia early warning method and system based on deep learning. The method comprises the steps that firstly, PPG signals are collected through a photoelectric volume pulse wave sensor, and denoising and standardization processing are conducted through an adaptive filtering algorithm; then, the heart beat position is recognized through a wavelet transform peak value detection algorithm, and a heart rate variability feature sequence is obtained; converting the one-dimensional heart rate variability feature sequence into a two-dimensional heart rhythm image by using an improved Gramian angular field transformation algorithm; and training an arrhythmia classification model by adopting a ResNet-50 convolutional neural network to realize automatic identification of different types of arrhythmia. And finally, continuous monitoring and timely alarming are realized through a sliding window technology and an early warning mechanism. The key technical problems of low detection precision, poor real-time performance, high equipment cost, poor user experience and the like in the existing arrhythmia detection technology are effectively solved, and an innovative solution is provided for heart rhythm health monitoring of the intelligent wearable equipment.

This invention discloses a real-time arrhythmia detection method based on morphological fidelity consistency constraints, comprising the following steps: S1, preprocessing continuous electrocardiogram (ECG) signals and dividing them into overlapping sliding time windows; S2, inputting each time window into an ECG signal encoder obtained through joint training to obtain a latent representation vector for that time window; the joint training is specifically defined as: simultaneously optimizing the encoder parameters using supervised classification signals, contrast consistency signals, and morphological fidelity signals during the training phase; S3, inputting the latent representation vector into a classifier and outputting the class probability of the time window belonging to each arrhythmia category; S4, based on the class probability, outputting real-time detection results using an online decision strategy of threshold hysteresis and cross-window consistency. This invention can simultaneously achieve robust identification under dynamic interference, accurate preservation of clinically critical waveform details, and efficient real-time deployment at the edge. This invention also provides a real-time arrhythmia detection system and device based on morphological fidelity consistency constraints.

Techniques are disclosed for using both feature delineation and machine learning to detect cardiac arrhythmia. A computing device receives cardiac electrogram data of a patient sensed by a medical device. The computing device obtains, via feature-based delineation of the cardiac electrogram data, a first classification of arrhythmia in the patient. The computing device applies a machine learning model to the received cardiac electrogram data to obtain a second classification of arrhythmia in the patient. As one example, the computing device uses the first and second classifications to determine whether an episode of arrhythmia has occurred in the patient. As another example, the computing device uses the second classification to verify the first classification of arrhythmia in the patient. The computing device outputs a report indicating that the episode of arrhythmia has occurred and one or more cardiac features that coincide with the episode of arrhythmia.

A cardiac treatment device, including:stimulation circuitry configured to generate a non-excitatory electrical signal which, when applied to ventricular tissue during a ventricular refractory period thereof improves a condition of heart failure in human patients;atrial arrhythmia detection circuitry; anddecision circuitry which controls the stimulation circuitry to delivery said signal, also when said atrial arrhythmia detection circuitry detects an atrial arrhythmia.

A wearable cardioverter defibrillator (WCD) comprises a plurality of electrocardiography (ECG) electrodes and a plurality of defibrillator electrodes to contact the patient's skin when the WCD is delivering therapy to the patient, a preamplifier coupled to the ECG electrodes to obtain ECG data from the patient. A processor to receive the ECG data from the preamplifier, and a high voltage subsystem to provide a defibrillation voltage to the patient through the plurality of defibrillator electrodes in response to a shock signal received from the processor. In a first power mode of a range of power modes, the preamplifier is configured to perform low-fidelity ECG acquisition and the processor is configured to perform simple arrythmia detection analysis, and in a second mode of the range of power modes, the preamplifier is configured to perform high-fidelity ECG acquisition and the processor is configured to perform complex arrythmia detection analysis.

Systems and methods are disclosed for improving arrhythmia detection to optimize resources of a medical device system, including detecting an arrhythmia episode using cardiac electrical information and analyzing cardiac mechanical information occurring over a final portion of a cardiac cycle of the detected arrhythmia episode to confirm the detected arrhythmia episode or to determine a type of the detected arrhythmia episode. The systems and methods can provide a control signal to control a mode or operation of a component of the medical device system based on the confirmation of the detected arrhythmia episode or the determined type of the detected arrhythmia episode to optimize resources of the medical device system.

The invention discloses an arrhythmia detection algorithm based on deep active learning and complex feature fusion, and the algorithm comprises the steps: building a new signal evaluation index through the weighted fusion of the statistical characteristics and morphological indexes of electrocardiosignals, and the complexity and dynamic indexes; precise screening of high-value samples is completed by means of a signal screening strategy; and a heart rhythm detection model fusing a convolutional neural network, a bidirectional long-short-term memory network, a channel attention mechanism and a Transform is constructed and is used for identifying 15 heart rhythm types. According to the algorithm, the marking cost can be remarkably reduced, the model detection performance is optimized, a stable technical scheme is provided for intelligent diagnosis of arrhythmia, the clinical application prospect is wide, and the market value is prominent; according to the method, efficient and accurate detection of arrhythmia can be realized, the industrial pain point of label data scarcity is effectively solved, the diagnosis efficiency is improved while the data labeling cost is remarkably reduced, and the method has relatively strong clinical application prospects and engineering popularization values.

The present application relates to the technical field of non-contact vital sign monitoring, and specifically discloses a heart signal detection method based on feature decoupling millimeter wave radar based on contrast learning. The present application realizes the automatic separation of the rhythm feature and the morphological feature of the heart signal from the millimeter wave radar signal by constructing an adaptive filtering module, a sample generator, a multi-scale feature extraction module, a double-path feature decoupling module, a physical constraint module and a contrast learning loss function. The present application uses contrast learning to construct a positive sample pair, guides the model to learn a feature representation that is invariant to heart rate changes and waveform deformation, and combines physiological prior constraints to improve the robustness and interpretability of the features. The present application effectively solves the shortcomings of traditional methods in signal separation, noise suppression and generalization ability, and can be widely applied to heart rate variability analysis, arrhythmia detection and other heart health monitoring tasks.

The invention discloses a noise robust arrhythmia detection method and system based on multi-task learning, and the method comprises the following steps: S1, collecting an electrocardiosignal obtained by a wearable device, and carrying out the signal quality evaluation and preprocessing, thereby obtaining target electrocardiosignal data; s2, constructing a shared feature extraction network, and extracting high-dimensional feature representation of the electrocardiosignals; s3, simultaneously constructing an arrhythmia detection task, an electrocardiosignal denoising task and a feature representation distribution constraint task; s4, the optimization weights of different tasks are adjusted in a self-adaptive mode, and optimization conflicts among multiple tasks are relieved through a gradient correction mechanism; and S5, after model training is completed, a detection result is output only based on the arrhythmia detection task. According to the method, the arrhythmia detection accuracy and stability of the wearable electrocardiosignal in a complex noise environment can be effectively improved, the robustness of the model to noise interference is enhanced, the false detection and missing detection risks are reduced, and the method is widely applied to wearable electrocardiosignal monitoring, remote medical treatment and continuous health management systems.

The utility model discloses a dynamic electrocardio examination monitoring wearing device, which relates to the technical field of electrocardio medical sensors and comprises a recorder, the recorder comprises a recorder host and a sensor component, the recorder host comprises a shell, a first circuit board is arranged in an inner cavity of the shell, and a second circuit board is arranged in the inner cavity of the shell. The circuit board is provided with a connecting antenna and a light guide column, the sensor component comprises a connecting sleeve connected with the bottom of the shell in a buckled mode, a mounting groove is formed in the connecting sleeve, and five switching copper columns are arranged in the mounting groove; an adhesive tape is arranged at the bottom of the connecting sleeve, an electrocardio sensor is arranged on the adhesive tape, and PU waterproof films are arranged on the two sides of the adhesive tape; hydrogel is also arranged on the PU waterproof film; the utility model provides the arrhythmia detection device which is simple to operate and convenient to wear, can greatly improve the detection rate of arrhythmia, can be worn by a patient at any time, can carry out data transmission and checking, and simplifies the operation process of arrhythmia.

A medical device is configured to sense one or more cardiac electrical signals and sense a ventricular event signal from the one or more cardiac electrical signals. The medical device can determine that the one or more cardiac electrical signals satisfy a tachyarrhythmia detection criterion, and in response to satisfying the tachyarrhythmia detection criterion, apply a delay criterion to the one or more cardiac electrical signals sensed during each of a plurality of groups of a plurality of sensed ventricular event signals. In response to satisfying the detection delay criterion, the medical device can delay detection of a tachyarrhythmia.

Techniques are disclosed for using both feature delineation and machine learning to detect cardiac arrhythmia. A computing device receives cardiac electrogram data of a patient sensed by a medical device. The computing device obtains, via feature-based delineation of the cardiac electrogram data, a first classification of arrhythmia in the patient. The computing device applies a machine learning model to the received cardiac electrogram data to obtain a second classification of arrhythmia in the patient. As one example, the computing device uses the first and second classifications to determine whether an episode of arrhythmia has occurred in the patient. As another example, the computing device uses the second classification to verify the first classification of arrhythmia in the patient. The computing device outputs a report indicating that the episode of arrhythmia has occurred and one or more cardiac features that coincide with the episode of arrhythmia.

The invention relates to an arrhythmia detection method and device, a terminal device and a computer program product, and the method comprises the steps: receiving an abnormal PPG signal segment sent by a wearable device, the abnormal PPG signal segment being obtained by the wearable device through screening from a plurality of collected PPG signal segments based on a PPG signal template; converting the abnormal PPG signal fragment into a time-frequency spectrogram; the abnormal PPG signal segments and the time-frequency spectrogram are input into a trained arrhythmia detection model to obtain an arrhythmia classification result, and the arrhythmia detection model is used for carrying out feature extraction on the abnormal PPG signal segments and the time-frequency spectrogram, fusing the extracted features and outputting the arrhythmia classification result based on the fused features; and sending the arrhythmia classification result to the wearable device. According to the invention, the problem that an ECG signal-based arrhythmia detection method in the prior art cannot be suitable for long-term health monitoring is solved.

The invention discloses an AI-integrated remote ischemia adaptation training and blood pressure management integrated system, and belongs to the technical field of intelligent medical treatment. The system adopts a four-layer distributed architecture and comprises an intelligent sensing layer, an AI analysis layer, a clinical decision-making layer and an application interaction layer. The intelligent sensing layer acquires physiological data through equipment such as an intelligent ischemia training instrument and a multi-guide physiological monitor; the AI analysis layer performs intelligent analysis through five modules, namely a blood pressure analysis module, an arrhythmia detection module, an arteriosclerosis evaluation module, a training effect quantification module and a tolerance screening module; the clinical decision-making layer generates a personalized training scheme based on the analysis result and performs dynamic optimization; the application interaction layer provides multi-terminal services for patients, doctors, family members and management mechanisms. The problems that in the prior art, the function is single, parameters are fixed, and intelligent evaluation is lacked are solved, intellectualization, individuation and closed loop of remote ischemia adaptation training and blood pressure management are achieved, and the treatment effect and safety are remarkably improved.

A health monitoring system for the autonomous support of elderly people, the system comprising the following: a variety of Internet of Medical Things (IoMT) sensors, including at least one wearable physiological sensor, one or more environmental sensor units, a mobility tracking device and a medication intake module, wherein the IoMT sensors are configured to continuously collect physiological, environmental, behavioral and medication use data; an edge processing device that includes an embedded processor and a lightweight machine learning inference unit configured to perform real-time anomaly detection and generate local alerts in the absence of cloud connectivity; a cloud-based analytics engine comprising one or more deep learning models configured to perform longitudinal health modeling, population-level clustering, and predictive risk forecasting using historical and real-time data; and a multi-channel communication unit configured to securely transmit notifications, reports, and emergency triggers to nursing staff, clinical staff, family members, and emergency services, wherein the edge processing device includes a dedicated hardware accelerator selected from a field-programmable gate array (FPGA), the accelerator being optimized to perform low-latency computations, including fall detection, arrhythmia detection, and respiratory abnormality monitoring, within a processing window of less than 200 milliseconds;and wherein the wearable physiological sensor is designed as a wrist-worn or patch-based device comprising a set of sensor elements, including an electrocardiography (ECG) electrode, a photoplethysmography (PPG) sensor for SpO2 measurement, a three-axis accelerometer, a gyroscope, and a skin temperature probe, the device further being configured to integrate raw signal data streams through a multimodal fusion technique performed locally on the edge device to minimize noise, compensate for motion artifacts, and improve the specificity of anomaly detection across multiple concurrent parameters.

A medical device is configured to obtain cardiac signal fragments from cardiac signals sensed by sensing circuitry of the medical device, and perform morphological analysis of the cardiac signal fragments for classifying the cardiac signal fragments as one of ventricular tachyarrhythmia, non-ventricular tachyarrhythmia, or cardiac arrest. A medical device may cover a denial rule for suppressing detection of tachyarrhythmia by detecting a ventricular tachyarrhythmia in response to detecting a desired number of tachyarrhythmia intervals when the denial rule is satisfied and a cardiac signal segment is classified as the ventricular tachyarrhythmia based on a morphological analysis.

The invention discloses application of a reagent for detecting SCN2B and / or SCN4B gene mutation in preparation of a hereditary arrhythmia detection product. According to the present invention, through the dual detection system combining whole exon sequencing with Sanger sequencing verification, the accurate capture of the pathogenic mutation such as SCN2B-R28Q / Y69H / P210L and SCN4B-T211M is achieved, and the problem that only SCN5A is determined as the main pathogenic gene in the clinical diagnosis of J wave syndrome (JWS) at present, and the gene detection of about 70-80% of patients is negative is solved. Wherein the three missense mutations of the SCN2B and the SCN4B-T211M variation are reported in the JWS for the first time, and the SCN4B is clear as the JWS pathogenic gene for the first time, so that the pathogenic gene spectrum of the JWS is obviously expanded. A detection product developed on the basis of the method can realize early accurate diagnosis, provide personalized risk assessment (such as sudden death early warning induced by fever) for patients, guide genetic screening of family members and remarkably improve clinical diagnosis and treatment efficiency.

The invention discloses an arrhythmia detection model modeling method based on multi-scale spatial-temporal feature fusion, and the method comprises the steps: designing a multi-scale selection fusion embedding module, and achieving the efficient fusion of multi-scale time features in an ECG signal; an arrhythmia detection model is designed, a space-time agency attention module is introduced into one branch, space-time features of ECG signals in input leads are extracted, fused and enhanced, and in the other branch, the cosine similarity of approximation coefficients between all lead pairs is calculated and a threshold value is applied, so that the time-space characteristics of the ECG signals in the leads are extracted, fused and enhanced. Constructing a graph structure of the correlation between the leads to quantify a feature dependency relationship between the leads, and performing operation depth extraction on spatial topological features between the leads through a multi-layer graph convolutional network module; in-lead multi-scale spatial-temporal features and inter-lead spatial features obtained by two branches of the arrhythmia detection model are fused cooperatively, so that the network can understand ECG signals more comprehensively, and obtained mixed features are input into a full-connection layer for prediction and classification.

Techniques are described for discriminating SVT and, in particular, rapidly conducting AF. The techniques include detecting an onset of a fast rate of ventricular events sensed from a cardiac electrical signal and detecting a pause in the fast rate of ventricular sensed events. A threshold number of ventricular event intervals required to detect a ventricular tachyarrhythmia is detected with each of the threshold number of ventricular event intervals being less than a tachyarrhythmia detection interval. Detection of the ventricular tachyarrhythmia and an electrical stimulation therapy for treating the ventricular tachyarrhythmia are withheld in response to at least the pause being detected.

The application discloses application of a reagent for detecting SCN2B and / or SCN4B gene mutation in preparation of a genetic arrhythmia detection product. Through a double detection system of whole exon sequencing combined with Sanger sequencing verification, the application realizes accurate capture of pathogenic mutations such as SCN2B-R28Q / Y69H / P210L and SCN4B-T211M, and solves the difficulty that only SCN5A is confirmed as a main pathogenic gene in the current clinical diagnosis of J wave syndrome (JWS), and about 70%-80% of patients have gene detection negative. The three missense mutations of SCN2B and the SCN4B-T211M variation are reported for the first time in JWS, and SCN4B is confirmed as a pathogenic gene of JWS for the first time, which significantly expands the pathogenic gene spectrum of JWS. The detection product developed based on this can realize early accurate diagnosis, provide personalized risk assessment (such as early warning of sudden death induced by fever) for patients, and guide genetic screening of family members, and significantly improves the clinical diagnosis and treatment efficiency.

The present application relates to a kind of R-wave peak detection and arrhythmia detection combined method based on end-to-end framework, comprising obtaining single-channel original ECG signal;Original ECG signal is directly input 1D diffusion model as original heartbeat label, generates reconstructed heartbeat label by forward noise and reverse denoising;DilSelfONN self-generation heterogenous neuron is constructed, adaptive nonlinear transformation is realized by Taylor series approximation, long-range dependence is captured by combining expansion mechanism;DilSelfONN network is built, and end-to-end joint framework is formed with 1D diffusion model, generates the joint output for R-wave peak detection and arrhythmia detection;Loss function is constructed, and the training and update of DilSelfONN network are carried out.The present application fuses diffusion model generation principle and self-generation heterogenous neuron characteristics, can fully represent the intrinsic characteristics of ECG data, and the generalization ability is excellent, can be efficiently deployed in portable, edge medical equipment, provides technical support for cardiovascular disease early accurate detection, with wide industrial application prospect.

This invention discloses a method and system for early warning of arrhythmias in smartwatches based on deep learning. The method first acquires PPG signals using a photoplethysmography (PPG) sensor and performs noise reduction and standardization using an adaptive filtering algorithm. Then, it uses a wavelet transform peak detection algorithm to identify heartbeat locations and obtain a heart rate variability (HRV) feature sequence. Next, it uses an improved Gramian angular field transform algorithm to convert the one-dimensional HRV feature sequence into a two-dimensional heart rhythm image. Then, it uses a ResNet-50 convolutional neural network to train an arrhythmia classification model, achieving automatic identification of different types of arrhythmias. Finally, it uses sliding window technology and an early warning mechanism to achieve continuous monitoring and timely alarms. This invention effectively solves key technical problems in existing arrhythmia detection technologies, such as low detection accuracy, poor real-time performance, high equipment costs, and unsatisfactory user experience, providing an innovative solution for heart rate health monitoring in smart wearable devices.

A cardiac therapy device includes a stimulation circuit configured to generate a non-excitatory electrical signal that improves a condition of heart failure in a human patient when the non-excitatory electrical signal is applied to ventricular tissue during a ventricular refractory period of the ventricular tissue; an atrial arrhythmia detection circuit; and a decision circuit that controls the stimulation circuit to deliver the signal when the atrial arrhythmia detection circuit detects an atrial arrhythmia.

A medical device is configured to sense event signals from a cardiac electrical signal and determine maximum amplitudes of cardiac electrical signal segments associated with sensed event signals. The medical device is configured to determine at least one tachyarrhythmia metric based on at least a greatest one of the determined maximum amplitudes. The medical device may determine when the at least one tachyarrhythmia metric does not meet true tachyarrhythmia evidence and, in response, determine when the maximum amplitudes meet suspected noise criteria. The medical device may withhold a tachyarrhythmia detection and tachyarrhythmia therapy when suspected noise criteria are met.

Systems and methods for detecting cardiac arrhythmia are discussed. A medical-device system includes an arrhythmia detector circuit and an event prioritizer circuit. The arrhythmia detector circuit can detect an atrial activation event from a cardiac electrical signal sensed from the patient, and determine an atrial fibrillation (AF) confidence indicator based on a signal characteristic of the atrial activation event. The event prioritizer circuit can generate an event priority for the AF event based on the AF confidence indicator. Multiple AF events may be prioritized in a specific order and presented to a user or a process.

The application discloses an arrhythmia detection method based on cross-modal data enhancement. In view of the problems of serious imbalance of class distribution of existing ECG data sets and limited effect of multi-modal feature fusion, the method comprises the following steps: db6 wavelet denoising and heartbeat segmentation are performed on the original electrocardiogram signal, and categories are merged according to the AAMI standard; a one-dimensional heartbeat time sequence signal is converted into a two-dimensional polyline waveform image with a pixel size of 224*224, and a metadata CSV file containing signal indexes, image paths and category labels is constructed; signal data and images are sequentially matched sample by sample, and a multi-modal paired data set is constructed; an intra-class multi-modal reorganization (ICMR) enhancement strategy is proposed, signal and image are independently and randomly sampled from the same category sample pool under the constraint of maintaining category consistency, and are re-paired, and the imbalance problem of categories is relieved through differential amplification rate; a double-flow multi-modal fusion classification network composed of a one-dimensional CNN-bidirectional LSTM signal encoder, a ResNet18 image encoder, a gating fusion module and a classification head is constructed, and two-way features of signal and image are adaptively integrated; cross-entropy loss function and Adam optimizer are used for end-to-end training and evaluation. The application effectively improves the recognition performance of the minority class arrhythmia.