Identifying High Risk of Atherosclerosis Using Deep Learning and Ensemble Learning

Abstract

Atherosclerosis refers to the buildup of plaque on the artery walls. As the disease advances in its further stages, its burden could lead to stroke or heart attack. Atherosclerosis develops gradually, and mild stages of the condition are usually symptomless. Diagnosing patients in their early stages of the disease can facilitate timely clinical interventions enhancing patient’s quality of life by altering the course of the disease. The work presented in this paper is focused on classifying patients who are at high risk of Atherosclerosis using simple diagnosis tools available in every clinic. The final system is a prescreening tool providing the medical practitioners with recommendations regarding the disease. High risk patients can be referred to a cardiologist for further assessments. A dataset of 44 patients was collected including 17 low-risk and 27 high-risk patients. Two different approaches were taken, 1. using deep learning and time series data (ECG signals) 2. using traditional machine learning algorithms and tabular data. In the first approach, a Conv-GRU model was trained using ECG signals collected from patients. This method resulted in an average accuracy of 77% which was computed over 4 folds using cross validation. In the second approach, Stacking, an ensemble learning technique in which the final prediction is obtained by combining the prediction of different machine learning models trained on several attributes readily collected in the clinic, was used. An average accuracy of 81% was achieved using this method.