Key result
Enhanced AlexNet classifies four-category ECG images with ~99% accuracy, outperforming benchmark models.
Why the study?
Current diagnosis of myocardial infarction relies on manual analysis of ECG traces, which is time-consuming and prone to human error and inter-observer variation. This study introduced an improved AlexNet-based convolutional neural network for automated multi-class classification of ECG images.
Does an enhanced AlexNet-based convolutional neural network with a 13-lead contour-extraction and PCA preprocessing pipeline improve multi-class ECG image classification accuracy for myocardial infarction compared to benchmark machine learning models?
Does an enhanced AlexNet-based convolutional neural network with a 13-lead contour-extraction and PCA preprocessing pipeline improve multi-class ECG image classification accuracy for myocardial infarction compared to benchmark machine learning models?
An enhanced AlexNet CNN combined with 13-lead contour extraction and PCA dimensionality reduction achieves 99.44% accuracy in classifying myocardial infarction and abnormal heartbeats from ECG images, demonstrating potential for integration into automated clinical decision support systems.
Answers are grounded in this paper and cite their sources.
Hypothesis-generating for CNN-based ECG classification; prospective clinical validation required before any adoption.
Cardiovascular disease (CVD) is still the greatest cause of death around the world, myocardial infarction (MI) making up a large number of the deaths related to CVD. Currently, diagnosis depends on manual analysis of electrocardiogram (ECG) traces, a time-consuming process prone to human error and variation between observers. The present study introduces an improved AlexNet-based convolutional neural network (CNN) which enables automated multi-class classification of ECG images into four categories: Normal Heartbeat (NH), Myocardial Infarction (MI), Previous Myocardial Infarction (PMI), and Abnormal Heartbeat (HB). The dataset consisted of 929 ECG images taken from a public Mendeley repository. We used a thorough preprocessing pipeline involving denoising, digitisation, augmentation, segmentation into thirteen ECG leads, contour-based signal extraction, Min–Max normalisation, and Principal Component Analysis (PCA), which reduced the 3,060 contour features to 400 principal components while retaining about 99% of the total variance. To modify the AlexNet architecture, we replaced the three-channel RGB input with a single-channel grayscale input; this required changing the convolutional kernel sizes and redesigning the output layer to include four Softmax neurons and a 0.5 dropout rate. The model was trained using the Adam optimiser with a learning rate of 0.001 and with a train–test split of 80:20. The results obtained were 99.44% accuracy, a macro-averaged F1-score of 99.2%, and an AUC-ROC of 96.06%, thereby exceeding the performance of a lightweight CNN, a SMOTE-enhanced CNN, and a Particle Swarm Optimisation-based neural network (PSO-NN). The fact that modified CNN architectures are able to accurately detect heart attacks from ECG data shows that they have the potential to be incorporated into clinical decision support systems in order to allow efficient and clinically relevant automated cardiac screening.
No takes yet. Share an insight, caveat, or question.
Asabia et al. (2026) studied Myocardial Infarction and Cardiac Abnormalities (n=929). Enhanced Convolutional AlexNet Model with 13-lead contour extraction and PCA preprocessing vs. Lightweight CNN, CNN with Correlation-Aware SMOTE, and Particle Swarm Optimisation-based Neural Network (PSO-NN) was evaluated on Four-class ECG image classification accuracy (NH, MI, PMI, HB). The enhanced AlexNet model achieved 99.44% accuracy, 99.2% F1-score, and 96.06% AUC-ROC in classifying four-category ECG images, outperforming benchmark CNN and PSO-NN models.
Synapse has enriched 10 related papers. Consider them for comparative context: