BME Seminar Series: Mahesh Nagarajan

Characterization of Chondrocyte Patterns in Normal and Osteoarthritic Human Patella Cartilage on Computed Tomography Phase-Contrast Imaging through Texture Analysis

Abstract

Previous studies have shown that computed tomography phase-contrast imaging provides excellent visualization of zonal patterns in the cartilage matrix. Here, the differences between chondrocyte patterns observed in the radial zone of normal and osteoarthritic human patella cartilage were quantitatively evaluated through texture analysis techniques. Regions of interest (ROI) were extracted from the radial zone in high resolution CT-phase contrast images of 2 osteochondral cylinders, one healthy and one osteoarthritic.

Gray-level patterns observed in these ROIs were characterized through second order statistical features (gray level co-occurrence matrices) and more sophisticated morphological (Minkowski Functionals) and geometric (Scaling Index Method) techniques. The extracted features were then subjected to a machine learning task to determine how accurately they could predict the character of the cartilage sample the corresponding ROI was extracted from. Our results show that texture analysis-based characterization of chondrocyte patterns in the radial zone of the cartilage matrix can predict the character of ROIs extracted from CT phase-contrast images with high diagnostic accuracy.