Tissues classification of the cardiovascular system using texture descriptors

Abstract

In this paper, we present an approach to automatically classify tissues of the cardiovascular system using texture information. Additionally, this process makes possible to identify some cardiovascular organs, since some tissues belong to muscles associated to those, i.e. identifying the tissue makes possible to identify the organ. We have assessed rotation invariant Local Binary Patterns (LBPri) and Haralick features to describe the content of histological images. We also assessed Random Forest (RF) and Linear Discriminant Analysis (LDA) for the classification of these descriptors. The tissues were classified into four classes: (i) cardiac muscle of the heart, (ii) smooth muscle of the elastic artery, (iii) loose connective tissue, and (iv) smooth muscle of the large vein and the elastic artery. The experimental validation is conducted with a set of 2400 blocks of 100 × 100 pixels each. The classifier was assessed using a 10-fold cross-validation. The best AUCs (0.9875, 0.9994 and 0.9711 for the cardiac muscle of the heart, the smooth muscle of muscular artery, the smooth muscle of the large vein and the elastic artery classes, respectively) are achieved by LBPri and RF.