Using deep learning for calculation detection in coronary artery disease intravascular ultrasound image

Abstract

Coronary artery calcification is also part of atherosclerosis in cardiovascular systems. In this study, we present an automatics deep learning systems to detect the presence and the absent of calcification in intravascular ultrasound (IVUS) images. The standard practice for radiologists and clinicians to detect calcification are using visual inspection. The proposed system used Convolutional Neural Networks (CNNs), named AlexNet model, with six types of classifiers (Support Vector Machine, Discriminant Analysis, Ensembles, Decision Tree, K-Nearest neighbour and Naïve Bayes). The dataset B from MICCAI challenge 2011 consists of 1643 with calcification absent and 530 images with calcification present is used to demonstrate the effectiveness of our proposed automatic deep learning approach. The performance measures recorded are Accuracy, Sensitivity, Specificity, Positive predictive value and Negative predictive value. The performance is compared to the ground truth provided by the MICCAI challenge 2011. For testing and training, the cross-validation k-fold used are 2, 3, 5 and 10. The accuracy of one is obtained using four classifiers namely Support Vector Machine, Discriminant Analysis, Ensembles and k-Nearest neighbour using a Cartesian coordinate image when taken with k-fold=10. This is followed by Decision Tree classifier with an accuracy of 0.9890 using Polar Reconstructed Warp Coordinate images and Naïve Bayes classifier with an accuracy of 0.8464 when using Cartesian Warp coordinate images.