Machine Learning Models for the Prediction of Ejection Fraction of Left Ventricle in Coronary Heart Disease

Abstract

Coronary heart disease includes the formation of plaque around the walls of arteries of heart. That causes the arteries to be tightened and restricts the flow of blood inside the artery. The left ventricle is the main powerhouse that pumps and supply the blood to other organs via arteries. The ejection fraction is the proportion of blood that leaves heart each time it squeezes (contracts). American Heart Association describes the ejection fraction threshold of left ventricle is 50 to 70 percent for normal and anything below 50 percent is alarming. Ejection fraction can be measured by Echocardiography, MRI, CT scans, Nuclear Medicine Scan and Cardiac Catheterization. However, these tests are time demanding and might led to death due to wrong or delayed medication. Therefore, in present study machine learning models were developed for the prediction of ejection fraction of left ventricle for the diagnostics and economically personalized treatments. In present study, patient cardiac records have been used to train a model and predicted the ejection fraction by clinical dataset of 1343 which includes angiography and pathology reports having clinical features like heart rate, age, Troponin-I, low density lipoproteins, diabetes, systolic and diastolic blood pressures along with target variable having ejection fraction classes of normal, mild, moderate, and severe patients. Several models like Decision Trees, Gradient Boosted Trees, K-Nearest Neighbors, Artificial Neural Networks, Naïve Bayes, Support Vector Machine have been developed. However, out of these models, nonlinear support vector machines using polynomial kernel achieved 97 percent accuracy, which shows the robustness of the model, and that model can be used to predict ejection fraction for quick and accurate diagnostics and personalized therapies for coronary heart disease patients.