MERAC: Multimodal fusion of ECG and clinical report with local-global encoding for arrhythmia classification.

Arrhythmia is a condition in which a person's heartbeat is irregular and can pose serious health risks. Effective arrhythmia detection is necessary to reduce associated risks. This study aims to develop a new deep learning architecture combining Convolutional Neural Networks (CNNs) to extract features, Long Short-Term Memory (LSTMs) to handle sequential data, and Gated Recurrent Units (GRUs) to reduce computational resources, leveraging the strengths of each to achieve better classification accuracy for diagnosing using the MIT-BIH Arrhythmia Database. The data is preprocessed by 0.5 Hz (low) and 50 Hz (high) to remove noise, then segmented into smaller, normalized signals to a unique scale, determine the higher point of the QRS complex in ECG, then gets labeled each segment, and finally converted ECG segment to (2D). The proposed model outperforms models on CNN, LSTM, and GRU if we apply them alone, with a precision of 92%, F1-score of 97%, recall of 100%, and accuracy of 97%; this study's notable discovery is that the suggested method may substantially decrease the duration when using RNN networks in conjunction with CNN. This paper presents a cost-effective approach to ECG signal reduction and a robust automatic scheme for arrhythmia detection, leveraging the strengths of CNN, LSTM, and GRU networks. The suggested model has achieved significant improvements in accuracy and is potentially a useful tool for real-time clinical practice. • A novel hybrid deep learning model combining CNN, LSTM, and GRU for ECG analysis. • Robust preprocessing pipeline improves ECG signal quality and arrhythmia classification. • Achieves 97% accuracy, 100% recall, and 97% F1-score on the MIT-BIH Arrhythmia dataset. • Outperforms standalone CNN, LSTM, and GRU models in efficiency and accuracy. • Demonstrates potential for real-time, cost-effective arrhythmia diagnosis in healthcare.

Exercise-associated cardiac arrhythmias may contribute to poor performance and sudden cardiac death in horses. Widespread monitoring is limited by device availability and suitability for field use. To (1) compare electrocardiographic (ECG) quality from a smart textile system (Myant Skiin Equine) with a reference device (Televet II) during exercise, and (2) evaluate whether arrhythmia detection with Myant Skiin Equine is comparable to the Televet II. Prospective blinded clinical study. Fifty racehorses (25 Thoroughbreds, 25 Standardbreds) underwent up to three high-speed exercise sessions, with simultaneous ECG recordings from both devices. Recordings were assessed for diagnostic quality and arrythmia detection. Simultaneous ECGs were recorded with both systems across 123 exercise sessions. Median (IQR) peak speed during exercise was 14.7 m/s (13.5-16.2). Twenty-nine (12%) recordings were non-diagnostic due to ≥10% artefact throughout the session (10 Myant; 19 Televet). The Myant Skiin Equine produced more artefact-free recordings (94/123; 76%) than the Televet II (45/123; 37%) (95% CI 28-46%; p < 0.001). Mean maximum heart rate (217.8 ± 10.1 bpm Myant; 217.6 ± 9.9 bpm Televet) showed strong agreement between devices (ICC = 0.995, mean bias = -0.153 bpm). In 79 paired diagnostic recordings assessed for arrhythmia detection, arrhythmias were detected in 85% of recordings, and in 43/50 horses (86%; 95% CI 73-94%). Agreement for arrhythmia classification was 97.5% (k = 0.96), with all discrepancies occurring because of artefact. Modest sample size and exclusion of some recordings due to artefact and horse withdrawal. Only the best-quality lead per recording was analysed, limiting multi-lead assessment and arrhythmia characterisation. Differences in device software and data handling may have influenced signal interpretation. The Myant smart textile system provides ECG recordings of comparable quality and arrhythmia detection to Televet II during high-speed exercise in racehorses. This technology offers a practical and alternative tool for field-based cardiac monitoring and may facilitate broader adoption.

Electrocardiography (ECG) is a widely adopted modality for monitoring cardiac rate and rhythm and identifying various abnormalities of the cardiac electrical system. Research on ECG preprocessing and intelligent classification is vital for the early diagnosis and clinical management of cardiovascular diseases and helps improve the accuracy and efficiency of cardiac diagnosis. Nevertheless, ECG signals captured by wearable acquisition devices frequently suffer from poor signal quality due to subject motion and uncontrolled acquisition environments. Furthermore, native ECG waveforms are faint and susceptible to diverse noise interference. To diagnose noise-corrupted arrhythmia, clinicians need to inspect lengthy ECG recordings relying on professional expertise, which is tedious, labor-intensive and prone to subjective bias. Another critical challenge lies in severe class imbalance within public ECG datasets, resulting in degraded classification performance for minority arrhythmia categories. Accordingly, developing robust automated arrhythmia detection algorithms is essential to ease the diagnostic workload of clinical practitioners. In this work, discrete wavelet transform (DWT) is adopted for noise elimination, and the SMOTE-Tomek hybrid resampling strategy is applied to mitigate class imbalance on the MIT-BIH and INCART datasets. We further propose MSCA-TNet, a hybrid architecture embedding the adaptive channel attention (ACA) module and Transformer to extract both local fine-grained features and global long-range contextual information from ECG sequences. The ACA module dynamically modulates channel-wise feature weights to emphasize clinically discriminative features while suppressing irrelevant components. On the two benchmark datasets, the proposed model yields macro-average accuracies of 93.97% and 98.79%, alongside overall accuracies of 98.88% and 99.47%.

BackgroundPreparticipation cardiovascular evaluation (PPE) is widely recommended to reduce the risk of sudden cardiac events in athletes; however, the optimal screening protocol remains debated. Although resting ECG and echocardiography are commonly used, intermittent arrhythmias may remain undetected.ObjectiveTo assess whether the addition of 24-hour ambulatory ECG (Holter) monitoring, including during sleep and training, increases the detection of potentially pathological cardiovascular findings.Design and settingsThis observational cross-sectional study included athletes (aged 14–42 years) from a Spanish elite multisport professional club undergoing pre-season PPE during the 2023–2024. All athletes completed a standardised evaluation comprising medical history and physical examination, resting 12-lead ECG, transthoracic echocardiography and 24-hour wireless Holter monitoring performed during daily activities including at least one training session. Detection rates of potentially pathological findings were compared across screening test combinations.ParticipantsOf 343 athletes recruited, 299 (79.6% male; mean age 22 years) were included in the final analysis.ResultsPotentially pathological findings prompting further evaluation were identified in 34 athletes (11.3%). The addition of any single test to medical history and physical examination significantly increased detection rates. Holter monitoring identified more intermittent arrhythmic events, particularly during exercise or sleep, whereas resting ECG more frequently detected repolarisation abnormalities. Echocardiography primarily contributed to the identification of structural abnormalities. Following a comprehensive evaluation, only one athlete was disqualified from competition.ConclusionsIn elite athletes, integrating Holter monitoring into PPE improved the detection of intermittent arrhythmias that may be missed by a resting ECG alone, and providing complementary diagnostic information and representing a useful adjunct to echocardiography in comprehensive cardiovascular screening strategies, particularly in selected high-risk athletic populations. Future longitudinal studies are warranted to determine the clinical significance of these findings.

Electrocardiogram (ECG) signals play a critical role in the early detection of cardiac arrhythmias, which remain a major cause of morbidity and mortality worldwide. While deep learning approaches have achieved high classification accuracy, their increasing complexity often limits interpretability and practical applicability. This study presents a systematic and interpretable framework for multi-class ECG arrhythmia classification, examining the effects of signal processing, feature extraction, feature selection, and evaluation strategies on classification performance. Experiments were conducted on the MIT-BIH Arrhythmia Database using two feature representations: (i) morphological and temporal features, and (ii) a compact wavelet-based representation. Sequential Forward Feature Selection (SFFS) revealed that classification performance saturates at approximately 15 features, indicating that most discriminative information is captured within a compact subset. Using this feature space, the support vector machine (SVM) achieved the best overall performance, reaching 98.54% accuracy with stable results across different configurations. The wavelet-based representation further improved performance balance, yielding lower Golden Distance (GD) values (as low as 0.0249), indicating more consistent behavior across evaluation metrics. Overall, the results demonstrate that carefully designed feature extraction and selection enable classical machine learning methods, particularly SVM, to achieve high and reliable performance, providing an interpretable alternative to more complex black-box models in ECG arrhythmia classification.

Atrial fibrillation (AF) is a common arrhythmia with a high incidence, and patients with chronic obstructive pulmonary disease (COPD) are at particularly high risk. However, there are currently no tools available for early risk stratification of AF in this population. To develop and validate a neural network diagnostic model based on transthoracic echocardiography to address two clinical challenges in patients with COPD: risk stratification for AF; and detection of occult supraventricular arrhythmias (including "micro-AF") based on 24 h ECG monitoring data. The study consisted of three consecutive stages: development of a neural network (NN) based on transthoracic echocardiography (TTE) parameters, validation of the model's predictive ability in patients (n = 311, including 99 with COPD), and assessment of the ability to detect occult atrial arrhythmias (n=207) in patients with COPD. The model architecture consists of a fully connected multilayer perceptron (MLP) with 13 inputs, 4 hidden layers of 130 neurons each, and 2 output neurons. Training was performed on 684 TTE scans (292 without AF, 392 with AF). The echocardiographic parameters were validated on an independent test set (n = 100). Statistical analysis included pairwise and multiple comparisons, logistic regression analysis, and ROC analysis with assessment of the area under the ROC curve (AUC). The median follow-up period for study participants was 18 months. The neural network demonstrated high classification metrics for AF on the test set (AUC = 0.80). A threshold value of the first output layer neuron > 0.75 allowed for the identification of a high-risk subgroup, in which the incidence of AF in patients with COPD was 14.8% versus 0% in the low-risk subgroup (p = 0.0073). Logistic regression models of the relationship between AF development and the neural network output value were statistically significant in both patients with COPD and patients without COPD (p < 0.0001). In patients with COPD without a history of AF, the neural network identified a high-risk group. In this group, 24 h ECG monitoring more frequently recorded episodes of AF, group supraventricular extrasystoles, and the combined endpoint (AF + GSE) compared to the low-risk group (55.32% vs. 17.5%; p < 0.0001). The area under the ROC curve for detecting latent AF in patients with sinus rhythm based on the neural network prediction was 0.93. The developed neural network model, which integrates a set of TTE parameters into a single quantitative measure of the severity of myocardial remodeling, is an effective tool for risk stratification for AF. The model may help identify COPD patients who could benefit from intensified rhythm monitoring; however, external validation is required before clinical implementation.

Self-tuned healthy homogeneous core: Addressing heterogeneities in biomedical datasets.

Cardiovascular disease (CVD) continues to claim more lives than any other condition worldwide. Traditional clinic visits capture only brief snapshots of health, leaving many opportunities for prevention and early intervention unmet. Wearable technologies now extend cardiovascular care into daily life, delivering continuous physiologic and behavioral data that could transform how we detect, treat, and ultimately prevent CVD. This review highlights how wearable devices, ranging from consumer-grade wristbands to advanced sensor-embedded textiles, are reshaping cardiovascular medicine. We discuss mechanical, optical, and electrochemical sensing modalities and their applications across the spectrum of care: risk assessment, arrhythmia and ischemia detection, heart failure monitoring, and cardiac rehabilitation. When paired with artificial intelligence, these devices generate predictive insights that anticipate clinical deterioration before symptoms appear. Landmark studies demonstrate real-world potential, but persistent challenges, including measurement accuracy, patient adherence, data overload, and limited integration into health systems, temper their current impact. Equally important, gaps in affordability and digital literacy risk widening disparities in cardiovascular outcomes if not urgently addressed. Wearables are moving cardiovascular medicine beyond the hospital and into the home, offering an unprecedented opportunity to shift from reactive treatment to proactive, personalized, and equitable care. Advances in multimodal sensing, artificial intelligence, and seamless health system integration could position wearables as cornerstone tools in the fight against CVD, provided that innovation is matched with rigorous validation, thoughtful regulation, and a commitment to health equity.

Cardiovascular diseases are the leading cause of death in the world, requiring the accurate and timely detection of arrhythmias to prevent sudden cardiac death. In this work, ScaHybNet, a deep learning ensemble model is proposed for multi-class arrhythmia classification using the widely adopted ECG Heartbeat Categorization Dataset. The dataset comprises 109,446 samples across five heartbeat classes (N, S, V, F, Q), enabling comprehensive arrhythmia analysis. The proposed method first transforms the ECG signals to 224 × 224 RGB-scalogram images using CWT with the Morlet wavelet. Then, a hybrid model is developed, which is composed of (1) a residual block-based CNN with skip connections to learn spatial features, (2) a BiLSTM layer for learning temporal features from the CNN feature maps and (3) a Transformer encoder layer with a custom-built multi-head self-attention mechanism to capture long-term dependencies. Thus, to address the extreme class imbalance within the data, stratified balancing of the data among normal beat, supraventricular ectopic beat, ventricular ectopic beat, fusion beat, and unknown beat, and inverse-frequency class weighting were performed. They assessed model robustness using fivefold cross-validation. Hyperparameters set to final values included a batch size of 2, 150 epochs, and an Adam optimizer. Ensemble train accuracy 99.81% and the mean accuracy on the fivefold cross validation set was 90.42% ± 1.26 (std) for ScaHybNet. On the test set (unseen data), it showed a total ensemble test accuracy of 94.73%, precision of 76.51%, recall of 82.93%, and F1-score of 77.40%. The ablation test proved the joint efficacy of each part of the model, and state-of-the-art analysis revealed better or equal results on current standards regarding ECG data with noise and imbalance. ScaHybNet appears to offer the potential to act as a more patient-centric tool that could offer considerable benefits to the medical field.

Atrial cardiomyopathy as a multidomain disease: longitudinal evidence for autonomic remodelling.

Accurate and efficient analysis of electrocardiogram (ECG) signals is essential for early detection of cardiac arrhythmias. However, many deep learning models suffer from limited generalization to unseen patients, reduced interpretability, and high computational demands. To address these challenges, we propose HAT-ECG, a hybrid Autoencoder-Transformer architecture that integrates unsupervised feature learning with attention-based temporal modeling. The convolutional autoencoder extracts compact, noise-robust latent representations of ECG beats, while the Transformer's Multi-Head Attention mechanism adaptively focuses on diagnostically relevant waveform segments. The model was evaluated on three public datasets: MIT-BIH, INCART, and the independent PTB Diagnostic ECG Dataset (290 subjects, entirely outside the original datasets). Under standard beat-wise splitting for benchmarking, HAT-ECG achieved state-of-the-art accuracies of 99.91% (MIT-BIH 5-class), 99.69% (MIT-BIH AAMI), 99.15% (INCART 3-class AAMI), and 98.45% (PTB 2-class). Critically, under strict patient-wise splitting on MIT-BIH (completely disjoint training and test patients), the model maintained a realistic accuracy of 90.81% (F1-score 92.61%), with strong generalization to new patients. With only 0.021 GFLOPs, HAT-ECG offers an excellent balance of high performance, interpretability, and efficiency, making it highly suitable for real-time wearable and edge-device cardiac monitoring. This work advances deep learning for intelligent and deployable arrhythmia classification by combining accuracy, cross-patient generalization, and computational minimalism.

Novel ECG signal classification based on Minkowski distance to enhance intelligent arrhythmia detection systems

ECGyolo: An end-to-end model combining transformer and 1D YOLO for arrhythmia detection

Cardiac arrhythmia is a disorder caused by disruptions in the regular heart rhythm. Arrhythmias are categorized into two classes: sinus and non-sinus rhythms. Whereas sinus rhythms are generally low-risk, non-sinus rhythms are associated with higher risks of morbidity and mortality, including stroke and death. This research proposes a novel method, Optimized Hierarchical Fused Fuzzy Deep Reinforcement Learning OHFFDRL, which incorporates three steps for arrhythmia prediction: data preprocessing, reinforcement learning, and fuzzy deep learning. We evaluated the proposed method using a 12-lead electrocardiogram dataset comprising 10,646 patients. Our approach leverages recent advances in machine learning and medical science to achieve an accuracy of 94% in predicting non-sinus rhythms. Furthermore, the area under the ROC curve for OHFFDRL was 0.91, and the empirical ROC area was 0.90. In addition, the interpretability of the model has been analyzed with SHAP, LIME, Calibration Curve, Adversarial vulnerability, and Integrated Gradients. Our experimental results, among other findings, indicate that the most important feature for distinguishing heart rhythms is TAxis (the movement range in ventricular repolarization). These results demonstrate the potential of machine learning for the early prevention of heart disease through non-sinus rhythm prediction. The source code is available at (https://github.com/arman-daliri/OHFFDRL).

Fetal arrhythmias, including tachycardias, bradycardias and irregular rhythms, may require close surveillance and management to prevent fetal heart failure, premature delivery and stillbirth. With early detection and appropriate management of pathological fetal rhythms, 90-95% of fetuses with even high-risk arrhythmias are live-born and can be delivered at term gestation. Fetal heart rate monitoring (FHRM) is a feasible method for ambulatory detection of new-onset or recurrent arrhythmias. When indications for FHRM are met, prescribed monitoring, guided and supported by the healthcare team, can empower pregnant individuals in the management of their pregnancies. Despite several studies showing utility of ambulatory FHRM, fetal cardiologists and perinatologists face challenges in successfully implementing FHRM programs. Under the auspices of the Fetal Heart Society, the Fetal Cardiology Program Leaders Committee formed a writing group to develop a consensus-based ambulatory FHRM process to be used as a guide for providers caring for pregnancies complicated by or at risk of manifesting fetal arrhythmias.

Deep learning techniques have shown significant promise for the automated diagnosis of CVD using ECG analysis. Nevertheless, several critical challenges persist with current approaches: severe class imbalance, intricate temporal dependencies, and poor modeling of inter-beat relationships ultimately limit clinical applicability. This work presents a novel hybrid deep learning framework that combines a CNN for time-domain feature extraction, k-nearest neighbor-based hypergraph construction with cosine similarity to model inter-beat dependencies, and a residual-connection-enhanced hypergraph neural network (EHGNN) for robust classification. To address class imbalance, focal loss is implemented with adaptive class weighting. For evaluation, our model was tested on two benchmark datasets: the MIT-BIH Arrhythmia Database and the St. Petersburg Institute of Cardiological Technics (INCART) 12-lead Arrhythmia Database. In the case of the MITBIH dataset, a classification accuracy of 98.66% was achieved using the AAMI five-class classification system, whereas the results indicated a three-class arrhythmia detection accuracy of 95.46% for the INCART database. The model demonstrated consistent performance on two separate data sets, thus emphasizing its ability to generalize well. These results confirm that the proposed framework effectively addresses class imbalance through focal loss and SMOTE, temporal dependencies through CNN-based feature extraction, and inter-beat relationships through EHGNN, demonstrating great potential for clinical diagnostic systems by capturing complex heartbeat patterns that are otherwise overlooked by standard models.

Cardiovascular diseases, especially arrhythmias, are a leading cause of death worldwide. This highlights the need for automated systems that can detect and diagnose these conditions early. This research introduces a deep learning model that identifies arrhythmias using electrocardiogram (ECG) signals. The model focuses on five main types of heartbeats: Normal (N), Left Bundle Branch Block (L), Right Bundle Branch Block (R), Atrial Premature Beat (A), and Premature Ventricular Contraction (V). The system uses Lead I signals from several databases, including MIT-BIH Arrhythmia, Supraventricular, INCART 12-lead, and Sudden Cardiac Death Holter. This provides more than 3.9 million training segments and 112,575 testing segments. The data is preprocessed by dividing it into fixed windows of 180 samples, scaling it using Min-Max normalization, and balancing the classes with the Synthetic Minority Over-sampling Technique. The model combines 1D Convolutional Neural Networks to extract spatial features and transformer layers to capture time-based patterns. It uses the Adam optimizer and includes dropout and batch normalization to enhance performance. The system achieves 99.99% accuracy, precision, and F1-score across all classes, which is better than the TN4 model and other top-performing models. The use of Convolutional Neural Networks and deep hybrid architectures improves the robustness of features. This model shows great potential for scalable and real-time arrhythmia detection and contributes to the advancement of AI-driven, personalized digital healthcare.

Non-ischemic cardiomyopathies (NICMs) are a heterogeneous group of myocardial disorders whose shared phenotypes complicate diagnosis and risk stratification. This complexity has driven growing interest in computational approaches that can integrate high-dimensional data and capture patterns not evident through conventional analyses. In this review, we synthesize studies published between 2020 and 2026 that apply machine learning and deep learning models to NICMs across three interconnected domains: phenotype classification, mechanism discovery, and clinical decision support. We find that imaging and ECG-based models help improve subtype discrimination, arrhythmia detection, and early disease identification; genomic and multi-omics approaches advance variant interpretation and biomarker discovery; and emerging multimodal frameworks extend these efforts toward outcome prediction and individualized management. Challenges remain in generalizability, interpretability, and prospective validation, as well as in integrating modalities into patient-level risk models. Continued development of multimodal and longitudinal approaches will be essential for translating these advances into precision care for NICMs.

The detection of cardiac arrhythmias from electrocardiogram (ECG) signals is essential for preventing sudden cardiac deaths. However, model performance on minority classes, such as supraventricular (S) and fusion (F) beats, which make up less than 3% of samples, is severely hampered by severe class imbalance in benchmark datasets like MIT-BIH. Conventional methods, such as basic GAN-based augmentation, oversampling, and undersampling, frequently result in low-fidelity synthetic samples or fail to preserve important ECG morphological features, leading to less-than-ideal F1-scores for uncommon arrhythmias. In this paper, a novel framework for high-fidelity minority class augmentation using conditional Wasserstein GAN with gradient penalty (cWGAN-GP) is proposed. It is integrated with a hierarchical multi-stream ResNet34 classifier that combines raw, Parzen-filtered, and similarity map features. In order to generate realistic ECG beats, the cWGAN-GP conditions generation on class labels. Mode collapse is prevented by enforcing gradient penalties and Wasserstein distance. Extensive experiments on the MIT-BIH arrhythmia database (5 AAMI classes) demonstrate significant improvements: minority S and F classes achieve F1-score increases of up to 28%, while the macro F1-score improves from 0.78 (baseline) to 0.94. The approach is computationally efficient for possible wearable device deployment and outperforms state-of-the-art GAN-augmented models in minority class detection.