Article: Heart disease detection using 1D transformer network: case of ECG signals and clinical data Journal: International Journal of Medical Engineering and Informatics (IJMEI) 2026 Vol.18 No.5 pp.1 - 18 Abstract: Early prediction of cardiovascular disease remains a critical public health challenge. This paper presents a 1D Transformer-based architecture for classifying patients as healthy or suffering from heart disease using ECG signals and clinical data. The model is evaluated on three benchmark databases: Cleveland Heart Disease (tabular data, 303 patients), PTB (ECG signals, 290 patients), and MIT-BIH (multi-class arrhythmia, 48 patients). Our approach achieves accuracies of 88.5% ± 1.2 (Cleveland), 94.2% ± 1.5 (PTB), and 89.2% ± 1.8 (MIT-BIH). The PTB dataset shows strong discriminative performance (AUC = 0.97), while Cleveland achieves AUC = 0.94. For MIT-BIH, class imbalance mitigation improves macro F1-score from 0.47 to 0.69. These results demonstrate the effectiveness of attention mechanisms for modelling biomedical time series, while highlighting the critical importance of proper validation protocols and imbalance mitigation for clinical applications. Inderscience Publishers - linking academia, business and industry through research

Title: Heart disease detection using 1D transformer network: case of ECG signals and clinical data

Authors: Amal Miloud Aouidate

Addresses: Department of Computer Sciences, Faculty of Science and Technology, Chadli Bendjedid El Tarf University, El Tarf, Algeria

Abstract: Early prediction of cardiovascular disease remains a critical public health challenge. This paper presents a 1D Transformer-based architecture for classifying patients as healthy or suffering from heart disease using ECG signals and clinical data. The model is evaluated on three benchmark databases: Cleveland Heart Disease (tabular data, 303 patients), PTB (ECG signals, 290 patients), and MIT-BIH (multi-class arrhythmia, 48 patients). Our approach achieves accuracies of 88.5% ± 1.2 (Cleveland), 94.2% ± 1.5 (PTB), and 89.2% ± 1.8 (MIT-BIH). The PTB dataset shows strong discriminative performance (AUC = 0.97), while Cleveland achieves AUC = 0.94. For MIT-BIH, class imbalance mitigation improves macro F1-score from 0.47 to 0.69. These results demonstrate the effectiveness of attention mechanisms for modelling biomedical time series, while highlighting the critical importance of proper validation protocols and imbalance mitigation for clinical applications.

Keywords: early prediction; cardiovascular disease; 1D transformer model; ECG classification; heart disease detection; Heart Cleveland database; PTB database; Mitbih database; supervised training; attention mechanisms; biomedical time series.

DOI: 10.1504/IJMEI.2026.153928

International Journal of Medical Engineering and Informatics, 2026 Vol.18 No.5, pp.1 - 18

Received: 07 Aug 2025
Accepted: 13 Apr 2026
Published online: 08 Jun 2026 *

Title: Heart disease detection using 1D transformer network: case of ECG signals and clinical data

Keep up-to-date