Hierarchical conditional random fields for myocardium infarction detection

Abstract

Accurate detection and delineation of myocardium infarction is important for treatment planning in patients with heart disease. Delayed contrast enhanced magnetic resonance imaging (DE-MRI) is a well established technique for the assessment of myocardial infarction. However, manual delineation of myocardium infarction in DE-MRI is both time consuming and prone to intra and inter rater variability. In this paper, we present an automatic, probabilistic framework for segmentation of myocardium infarction using Hierarchical Conditional Random Fields (HCRFs). In each level, a CRF classifier with up to triplet clique potentials is learnt. Furthermore, incorporation of spin image features in the second level allows for better learning the neighbourhood characteristics. The performance of the HCRF classifier on 5 animal scans and 5 human scans shows promising results. © 2013 Springer-Verlag.