Aims: Myocardial infarction (MI) is a severe case of coronary artery disease (CAD) and ultimately, its detection is substantial to prevent progressive damage to the myocardium. Recently, there has been a growing interest in leveraging machine learning (ML) to develop fast, accessible, and cost-effective assistive tools in echocardiography analysis. In this study, we propose a novel view-fusion model named self-attention fusion network (SAF-Net) to detect MI from multi-view echocardiography recordings.
Methods: The proposed network uses apical 2-chamber (A2C) and apical 4-chamber (A4C) view echocardiography recordings to perform the classification task. From each echocardiography recording, we extract three reference frames for both A2C and A4C views and deploy pre-trained state-of-the-art deep networks to extract highly representative features. The extracted features of A2C and A4C views are then stacked and given as the input to the proposed SAF-Net model. Accordingly, the SAF-Net model utilizes a self-attention mechanism which is widely used in vision transformer models that can exploit the relations between non-local patches. In the proposed model, the self-attention mechanism is used to learn dependencies between each echocardiography view and corresponding extracted feature vectors. This fusion mechanism so-called view pooling is trained end-to-end to boost the detection accuracy. The proposed model is computationally efficient thanks to its compact architecture having three main parts: a feature embedding to reduce dimensionality, self-attention for view-pooling, and dense layers for the classification.
Results: Experimental evaluation is performed using HMC-QU-TAU dataset which consists of 130 patients with A2C and A4C view echocardiography recordings. The proposed SAF-Net model achieves a high-performance level with 88.26% precision, 77.64% sensitivity, and 79% specificity with 81.57% F1-score and 78.13% accuracy.
Conclusion: It has been observed that the proposed SAF-Net model with a novel view-pooling technique and joint optimization of different parts provides a successful MI detection over multi-view echocardiography recordings.