Medical image-based diagnostics for cardiovascular diseases using machine learning

Abstract

With cardiac imaging’s important role in the diagnosis of cardiovascular diseases, along with the dawn of big data and machine learning (ML), there are emergent opportunities to build artificial intelligence (AI) tools that will directly assist physicians in heart failure (HF) diagnostics. An important application in biomedical engineering, as HF is very difficult to diagnose because of its complex symptoms, circumstances, and comorbidities. This study aims to: (1) perform accurate and precise cardiac segmentation and quantification of key left ventricle functional indices from CMR; and (2) build a ML tool using decision trees for image-based HF diagnosis. Quantification of left ventricular end-diastolic, end-systolic volumes and ejection fraction were achieved using Heron’s formula and the area-length method. One-sample T tests revealed there were no statistical significance between the obtained mean values and the comparative mean values in each quantified variable. Statistical results show the quantified values closely resemble those established in the Sunnybrook Cardiac Data. Finally, a Machine Learning tool using decision trees for imagebased heart failure diagnosis was successfully built, as every tested patient was classified correctly using the trained ML model.