A Deep Learning Ensemble-Based Approach for Cine MRI Segmentation and Classification Pipeline

Cardiovascular Disease (CVD) pose one of the leading causes of death across the world. This demands for fast and accurate diagnosis of heart disease. As technology in scans of heart advances, there is a requirement of an efficient pipelined process for detection of the possibility of heart disease. This research primarily focuses on automated segmentation of heart regions (right ventricle, left ventricle and myocardium) from MRI scans and using the extracted feature maps to classify the patient to a heart disease. The benchmark Automated Cardiac Disease Challenge (ACDC) dataset is used in training and testing of the model metrics. Ensemble method in machine learning has been judiciously applied to models like UNET and Dilated-CNN to extract the most prominent masks which were used for classification using algorithms like Random Forest, SVM and KNN. The approach received an segmentation accuracy of $\mathbf{9 2 \%}$ and a classification accuracy of $90 \%$ which surpasses many of the previously done researches. Delving into the behaviours of segmentation induced classification this paper contributes to the ongoing efforts in leveraging advanced image processing and machine learning techniques for cardiovascular health.