Background: This study investigates the application of machine learning models to classify electrocardiogram signals, addressing challenges such as class imbalances and inter-class overlap. In this study, “normal” and “abnormal” refer to electrocardiogram findings that either align with or deviate from a standard electrocardiogram, warranting further evaluation. “Borderline” indicates an electrocardiogram that requires additional assessment to distinguish benign variations from pathology. Methods: A hierarchical framework reformulated the multi-class problem into two binary classification tasks—distinguishing “Abnormal” from “Non-Abnormal” and “Normal” from “Non-Normal”—to enhance performance and interpretability. Convolutional neural networks, deep neural networks, and tree-based models, including Gradient Boosting Classifier and Random Forest, were trained and evaluated using standard metrics (accuracy, precision, recall, and F1 score) and learning curve convergence analysis. Results: Results showed that convolutional neural networks achieved the best balance between generalization and performance, effectively adapting to unseen data and variations without overfitting. They exhibit strong convergence and robust feature importance rankings, with ventricular rate, QRS duration, and P-R interval identified as key predictors. Tree-based models, despite their high performance metrics, demonstrated poor convergence, raising concerns about their reliability on unseen data. Deep neural networks achieved high sensitivity but suffered from overfitting, limiting their generalizability. Conclusions: The hierarchical binary classification approach demonstrated clinical relevance, enabling nuanced diagnostic insights. Furthermore, the study emphasizes the critical role of learning curve analysis in evaluating model reliability, beyond performance metrics alone. Future work should focus on optimizing model convergence and exploring hybrid approaches to improve clinical applicability in electrocardiogram signal classification.

Background/Objectives: Electrocardiogram data are widely used to diagnose cardiovascular diseases, a leading cause of death globally. Traditional interpretation methods are manual, time-consuming, and prone to error. Machine learning offers a promising alternative for automating the classification of electrocardiogram abnormalities. This study explores the use of machine learning models to classify electrocardiogram abnormalities using a dataset that combines clinical features (e.g., age, weight, smoking status) with key electrocardiogram measurements, without relying on time-series data. Methods: The dataset included demographic and electrocardiogram-related biometric data. Preprocessing steps addressed class imbalance, outliers, feature scaling, and the encoding of categorical variables. Five machine learning models—Gaussian Naive Bayes, support vector machines, random forest trees, extremely randomized trees, gradient boosted trees, and an ensemble of top-performing classifiers—were trained and optimized using stratified k-fold cross-validation. Model performance was evaluated on a reserved testing set using metrics such as accuracy, precision, recall, and F1-score. Results: The extremely randomized trees model achieved the best performance, with a testing accuracy of 66.79%, recall of 66.79%, and F1-score of 62.93%. Ventricular rate, QRS duration, and QTC (Bezet) were identified as the most important features. Challenges in classifying borderline cases were noted due to class imbalance and overlapping features. Conclusions: This study demonstrates the potential of machine learning models, particularly extremely randomized trees, in classifying electrocardiogram abnormalities using demographic and biometric data. While promising, the absence of time-series data limits diagnostic accuracy. Future work incorporating time-series signals and advanced deep learning techniques could further improve performance and clinical relevance.

Stimulation of sensory vagal pathways is typically delivered via invasive, cervical vagus nerve stimulation (cVNS) or noninvasive, trans-auricular nerve stimulation (taNS). While both methods are investigated therapeutically, their effects on brain physiology remain poorly understood, hindering mechanistic insights and stimulus optimization. In 6 awake nonhuman primates, we recorded cortical vagal-evoked potentials (VEPs) from subdural electrodes placed in prefrontal, sensorimotor and parietal cortical areas, in response to cVNS or taNS. Across 478 different taNS and cVNS protocols, we varied stimulation side, intensity, frequency, pulse count, and pulse width and assessed independent effects on amplitude and latency of early (EC; 30-100 ms), intermediate (IC; 101-200 ms) and late components (LC; 201-500 ms) of VEPs. Fixed and random effects of stimulation parameters and subjects, respectively, on VEPs were assessed using a linear mixed-effects model. Overall, cVNS elicits more robust VEPs than taNS, with larger EC, IC and LC amplitudes, in both hemispheres. cVNS-elicited ECs and LCs are largest in PFC and PC areas, whereas ICs are largest in SM areas. On the other hand, taNS generally does not elicit area-specific responses. cVNS-elicited ECs have slower latency than ta-NS elicited ECs. Higher stimulation frequencies and intensities and a longer pulse width elicit larger ECs and ICs for cVNS, and to some extent for taNS. Both short and long cVNS trains elicit stronger ECs, and long trains elicit slower ICs. Earlobe stimulation elicits VEPs that partially overlap with those from taNS. In conclusion, cVNS and taNS elicit cortical VEPs in a manner consistent with distinct engagement of ascending vagal pathways, with both similarities and differences in the effects of stimulation parameters on evoked responses.