Texture classification of scarred and non-scarred myocardium in cardiac MRI using learned dictionaries

Abstract

The late gadolinium enhancement in Cardiac Magnetic Resonance (CMR) imaging is used to increase the intensity of scar area in myocardium for thorough examination. The results in our previous work [1] arises the hypothesis that there are textural differences between the non-scarred myocardium and the scarred areas. This paper presents our work of testing the hypothesis further by applying dictionary learning techniques and sparse representation on CMR images (manually segmented by cardiologists) in order to find textural differences in the myocardium and to classify texture in the non-scarred myocardium and the scarred areas. After myocardial infarction, cardiac patients considered to have high risk of ventricular arrhythmia are implanted with Implantable Cardioverter-Defibrillator (ICD). Our ultimate goal is to accurately identify the patients with highest risk of arrhythmia, who are to be implanted with ICD by exploring the textural properties in the scarred region of late gadolinium enhanced CMR images.