M-mode Carotid Artery Image Classification and Risk Analysis Based on Machine Learning and Deep Learning Techniques

Abstract

AbstractIn India, Cardiovascular Disease (CVD) is one of the major reasons for death. The age-standard CVD death rate in India is 272 per 100,000 individuals based on Global Burden of Disease Study. In the proposed work, risk of heart failure is predicted based on arterial stiffness indices parameters from carotid artery image. This study was carried out in South Indian population in and around Chennai. Carotid artery image was taken using ultrasound machine for 165 subjects (55 normal, 55 CVD and 55 diabetic). From the carotid artery images, arterial stiffness indices parameters such as stiffness index, elastic modulus, distensibility, and compliance are measured for predicting the risk of heart failure. Biochemical parameters such as HDL, LDL, total cholesterol, fasting blood sugar, and postprandial blood sugar are also recorded for the 165 subjects. Estimated Framingham Risk Score (FRS) is based on biochemical parameters, systolic blood pressure and diastolic blood pressure, and smoking. FRS is 10-year risk score prediction for analyzing the risk of heart failure. All these parameters are analyzed using biostatistical methods, machine learning methods, and deep learning methods to determine the risk of heart failure. Results show that Pearson’s correlation gives significant results with p < 0.01 for biochemical parameters and FRS score. The performance was also determined by comparing various classification techniques in machine learning such as Naïve Bayes, K-Nearest neighbor, Random Forest, Support vector machine (SVM), and multilayer perceptron. Results show that multilayer perceptron and SVM produce good results in four different models. Further analysis is performed by extracting the features of M-mode carotid artery images using transfer learning techniques such as ResNet50 and InceptionV3. The results show the highest accuracy of 99% and 98% are achieved using ResNet50 and InceptionV3, respectively for normal and CVD subjects. Relative risk analysis are also performed using SPSS software for stiffness index and elastic modulus with FRS and the results are tabulated.KeywordsCardiovascular diseaseArterial stiffnessElastic modulusStiffness indexDistensibilityCompliance