Detection of Fibrosis in LGE-Cardiac MRI using Kernel DL-based Clustering

Abstract

In this paper we address the problem of fibrosis detection in patients with Hypertrophic cardiomyopathy (HCM) by using a sparse-based clustering approach and Dictionary learning. HCM, as a common cardiovascular disease, is characterized by the abnormal thickening, architectural disorganization and the presence of fibrosis in the left ventricular myocardium. Myocardial fibrosis in HCM leads to both systolic and diastolic dysfunction. It can be detected in Late Gadolinium Enhanced (LGE) cardiac magnetic resonance imaging. We present the use of a Dictionary Learning (DL)-based clustering technique for the detection of fibrosis in LGE-Short axis (SAX) images. The DL-based detection approach consists in two stages: the construction of one dictionary with samples from 2 clusters (LGE and Non-LGE regions) and the use of sparse coefficients of the input data obtained with a kernel-based DL approach to train a K-Nearest Neighbor (K-NN) classifier. The label of a test patch is obtained with its respective sparse coefficients obtained over the learned dictionary and using the trained K-NN classifier. The method has been applied on 11 patients with HCM providing good results.