Electrocardiogram Abnormality Detection Using Machine Learning on Summary Data and Biometric Features

Diagnostics, Journal Year: 2025, Volume and Issue: 15(7), P. 903 - 903

Published: April 1, 2025

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

Language: Английский

Evaluating the Performance of DenseNet in ECG Report Automation

Electronics, Journal Year: 2025, Volume and Issue: 14(9), P. 1837 - 1837

Published: April 30, 2025

Ongoing advancements in machine learning show great promise for automating medical data interpretation, potentially saving valuable time life-threatening situations. One such area is the analysis of electrocardiograms (ECGs). In this study, we investigate effectiveness using a DenseNet121 encoder with three decoder architectures: Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), and Transformer-based approach. We utilize these models to generate automated ECG reports from publicly available PTB-XL dataset. Our results that paired GRU yields higher performance than previously achieved. It achieves METEOR (Metric Evaluation Translation Explicit Ordering) score 72.19%, outperforming previous best result 55.53% ResNet34-based model used LSTM Transformer components. also discuss several important design choices, as how initialize decoders, use attention mechanisms, apply augmentation. These findings offer insights into creating more robust reliable deep tools interpretation.

Language: Английский