Diagnosis of Atherosclerotic Plaques in Carotid Artery using Transfer Learning

The presence of plaque in the carotid artery can help accurately predict cardiovascular infirmities like strokes and heart attacks in people despite having no history of cardiovascular disease. To avoid the use of an invasive method to determine the amount of plaque in the carotid artery to predict the heart risk, proposed to use deep learning methods such as transfer learning to perform computational image analysis on ultrasound images to accurately predict the presence of plaque in the artery, thus helping prevent a life-threatening heart attack. Feature extraction was done on the images to create the dataset. Various mathematical parameters are used in the preprocessing of the images such as skewness, kurtosis etc. The dataset and the images were loaded into the model for descriptive analysis. Two approaches have been used for developing our model for predicting stroke risks. The first approach was to design a simple CNN to understand our purpose and attain preliminary results for the classification and the second approach is to retrain a famous image classifier known as MobileNet to attune to our specific requirements and have seen the promising result of attaining a validation accuracy of 95%