Deep Learning-Based Classification of Echocardiographic Perspectives: Emphasis on Parasternal Long Axis and Apical 4 Chamber for Mitral Valve Assessment

Abstract

Mitral valve disorders necessitate precise diagnosis, often relying on accurate interpretation of echocardiographic views. This study investigates the potential of automating the classification of two crucial perspectives: Parasternal Long Axis (PLA) and Apical 4 Chamber (A4C). A dataset of 1773 echocardiograms containing these views is analyzed, and five pre-trained convolutional neural networks (CNNs) are applied for view differentiation. ResNet50 achieved the highest accuracy (98.65%), demonstrating exceptional performance. Other models like EfficientNetB0, InceptionV3, and MobileNetV2 also performed well, while Xception exhibited the weakest accuracy. Detailed analysis using confusion matrices revealed additional insights: ResNet50 showed the highest sensitivity for PLA views, and ResNet50 also achieved good A4C classification with less misidentification. These findings showcase the promising potential of deep learning for automated echocardiographic view classification. This novel approach has the potential to streamline workflow, reduce human error, and ultimately contribute to improved diagnosis and management of mitral valve disorders and other cardiac pathologies.