Development of computational tool for early detection of cardiovascular diseases through analysis of data from otherwise healthy individuals

Abstract

Cardiovascular Disease (CVD) is one of the leading Non- newlineCommunicable Diseases (NCD) and contributes 31% towards global death. newlineCVDs refer to the disorder in the heart and blood vessels. A blood vessel newlinecirculates blood to all parts of the body and is affected by plaque on artery newlinewall. As a result, the arteries narrow down and restrict the flow of blood newlineleading to heart attack and stroke. Intima-media thickness (IMT) is a marker newlineto detect the presence of plaque in the arterial walls for diagnosis of CVDs. newlineThe proposed research focuses on diagnosis of CVDs by analysing carotid newlineartery ultrasound (US) images. 650 carotid ultrasound images are collected newlinefrom Apollo Hospitals, Chennai. The collected images are pre-processed for newlinethe removal of speckle noise. The main work of the research focuses on (i) newlinesegmentation of Intima Media Complex (IMC) and (ii) thickness newlinemeasurement of IMC.IMT is an important marker showing the onset of CVDs. Other risk newlinefactors of CVDs include age, gender, Body Mass Index (BMI), blood newlinepressure, cholesterol and sugar. State-of-art deep learning architectures are newlineproposed for segmentation of IMC, measurement of IMT, and image newlineclassification. Four different deep learning architectures are developed under newlinetwo subdivisions: pipeline architecture and end-to-end architecture. newlinePipeline architecture with Convolution Neural Network (CNN) is used newlinefor classifying the region containing IMC as Region of Interest (RoI) and newlinenon-IMC as Region of Non Interest (RoNI). The boundaries of lumen-intima newline(LI) and Media-Adventitia (MA) region are extracted from IMC using newlinethresholding technique for IMT measurement. The architecture has limitation newlineon thresholding method. Hence an end-to-end architecture is proposed newline newline