Local Binary Patterns used on Cardiac MRI to classify high and low risk patient groups

In patients having suffered myocardial infarction, the myocardium does not function properly due to scarring. These patients are divided into high and low risk of getting arrhythmia using recognized risk markers like Left Ventricular Ejection Fraction (LVEF) and scar size. In Cardiac Magnetic Resonance (CMR) imaging, the scarred tissue in the myocardium is studied by increasing the intensity of scar area with the help of contrast agents. In this work, we have explored if a group of patients with high risk of getting arrhythmias (HAG) can be distinguished from a group of patients with low risk of getting arrhythmias (LAG) using the texture differences present in the scar tissue as inputs to a classifier. In this work, the textural differences of scarred myocardium tissue in HAG and LAG are captured using Local Binary Patterns (LBP). Automatic classification of HAG and LAG is important as patients with high risk of arrhythmia are identified and implanted with Implantable Cardioverter-Defibrillator (ICD). A non-parametric classification method is used to classify the LBP and contrast measure distributions of HAG and LAG. This is a preliminary work on the classification of HAG patients and LAG patients that has to be explored further. Even with a limited dataset, experiments show that HAG and LAG can be distinguished with a sensitivity of 75% and specificity of 83.33% using LBP.