Data Science Day @ Columbia University

Aiming to establish the relationship between ultrastructure and endomyocardial tissue types, especially diseased tissue, we develop an automated algorithm to segment and classify tissue types from intracardiac OCT images. We segmented the OCT image volumes using a graph searching method. Features are extracted and compared in each segmented region. A probabilistic model of relevance vector machine is developed to classify multiple tissue types such as scar, fibrotic myocardium, normal myocardium, endocardium, and adipose tissue. The algorithm is validated from OCT images obtaining from human heart. The tissue types are classified with a good accuracy and are visualized in three dimensions.