University of Tabriz Tabriz Journal of Electrical Engineering 2008-7799 2026 04 11 Acute Myocardial Infarction Detection Using ECG Signals from Wearable Sensors: a Comparison Between Machine Learning And Deep Learning Approaches Acute Myocardial Infarction Detection Using ECG Signals from Wearable Sensors: a Comparison Between Machine Learning And Deep Learning Approaches 21402 10.22034/tjee.2026.70183.5101 FA Amir Hossein Tajarrod Faculty of Biomedical Engineering, Sahand University of Technology, Tabriz, Iran Tania Hossein Khani Faculty of Biomedical Engineering, Sahand University of Technology, Tabriz, Iran Asghar Zarei Faculty of Biomedical Engineering, Sahand University of Technology, Tabriz, Iran Mousa Shamsi Faculty of Biomedical Engineering, Sahand University of Technology, Tabriz, Iran Journal Article 2025 11 11 Timely detection of acute myocardial infarction (AMI) is essential for reducing mortality rates and improving treatment outcomes. However, traditional wearable devices face limitations in capturing multi-lead ECG signals, especially from chest leads, which makes accurate AMI detection challenging. This study aimed to evaluate the performance of a wrist-based ECG (wECG) device equipped with three electrodes, comparing it to the reference ECG signals (12 leads). The proposed framework effectively distinguishes AMI patients from healthy individuals and those with other cardiovascular diseases (CVD). Initially, raw ECG and wECG signals were preprocessed to extract higher-order statistical features, Hjorth descriptors, and amplitude and phase components derived from the Fast Fourier Transform. Key features and leads were selected using mutual information (MI) and the F-test. These selected features were then used in various machine learning algorithms (SVM, DT, KNN, XGBoost) and deep learning models (CNN, ResNet, DenseNet, LSTM) to develop an accurate and interpretable model for AMI detection. The results indicated that the 12-lead ECG achieved an impressive average accuracy of 100% in distinguishing healthy subjects from patients (both AMI and CVD) when using XGBoost and CNN. However, the accuracy for differentiating AMI from CVD decreased to 99.4% (using ResNet) and 96.2% (using SVM). For wECG data, the V5-LA lead demonstrated the best performance, achieving an average accuracy of 98.1% in differentiating control subjects from CVD patients with XGBoost. Nevertheless, the accuracy for distinguishing AMI from CVD was limited to 93.2% (using CNN).These findings suggest that wECG has potential for the early detection of AMI. Timely detection of acute myocardial infarction (AMI) is essential for reducing mortality rates and improving treatment outcomes. However, traditional wearable devices face limitations in capturing multi-lead ECG signals, especially from chest leads, which makes accurate AMI detection challenging. This study aimed to evaluate the performance of a wrist-based ECG (wECG) device equipped with three electrodes, comparing it to the reference ECG signals (12 leads). The proposed framework effectively distinguishes AMI patients from healthy individuals and those with other cardiovascular diseases (CVD). Initially, raw ECG and wECG signals were preprocessed to extract higher-order statistical features, Hjorth descriptors, and amplitude and phase components derived from the Fast Fourier Transform. Key features and leads were selected using mutual information (MI) and the F-test. These selected features were then used in various machine learning algorithms (SVM, DT, KNN, XGBoost) and deep learning models (CNN, ResNet, DenseNet, LSTM) to develop an accurate and interpretable model for AMI detection. The results indicated that the 12-lead ECG achieved an impressive average accuracy of 100% in distinguishing healthy subjects from patients (both AMI and CVD) when using XGBoost and CNN. However, the accuracy for differentiating AMI from CVD decreased to 99.4% (using ResNet) and 96.2% (using SVM). For wECG data, the V5-LA lead demonstrated the best performance, achieving an average accuracy of 98.1% in differentiating control subjects from CVD patients with XGBoost. Nevertheless, the accuracy for distinguishing AMI from CVD was limited to 93.2% (using CNN).These findings suggest that wECG has potential for the early detection of AMI.