An assessment of machine learning algorithms in diagnosing cardiovascular disease from right ventricle segmentation of cardiac magnetic resonance images

Abstract

Healthcare research shows early detection of cardiovascular disease (CVD) is the key to saving the lives of patients suffering from CVDs. Furthermore, recent studies show the Right Ventricle (RV) is equally important as the Left Ventricle (LV) in diagnosing CVD. The research also shows it is difficult to assess the functioning of the RV in terms of eco-cardiograph because of the complexity of the RV position and structure of the heart. Assessing the RV for the accurate identification of echocardiographic parameters still remains the challenge for diagnosing CVDs. Machine learning (ML) and deep learning (DL) methods can produce accurate results from the images by segmenting them to reveal the clearer parts of the image for better identification or classification. This research aims to develop three algorithms: Vanilla Convolutional Neural Networks (Vanilla-CNN), Fourier-CNN (FCNN), and Residual Networks (ResNet) for the image segmentation of the images obtained from Magnetic Resonance Imaging (MRI) scan to extract the RV position. This helps in the identification of anomalies in heart and problems related to CVDs. The proposed algorithms are compared using standard performance metrics to evaluate their performance and viability. The results of DL algorithms reveal that all the flavours of CNN such as Vanilla CNN, FCNN and ResNet are capable to produce accurate results in image segmentation to identify CVDs.