Knee Cartilage Segmentation Algorithms: a Critical Literature Review

Abstract

The objective of this study was to give a review of the knee cartilage segmentation methods presented in the literature, from a critical perspective. Segmentation allows quantitative and qualitative analysis of the cartilage from the morphological point of view. This is important in clinical applications, because it allows early diagnosis of pathological changes in the cartilage, such as osteoarthritis. Cartilage extraction can be done in various approaches, which can be grouped into edge tracking, intensity-based (thresholding, texture analysis, watershed transform), supervised learning (kNN classifiers, support vector machines, ensemble learning), and energy minimization (active contours, statistical shape models and surfaces, graph-cuts). Deciding which method is more applicable for particular situation greatly depends on what trade-off the user is willing to accept with respect to the robustness to the quality of the input data, computational time, and accuracy. It is often required to specify the region-of-interest or to indicate the most typical candidates for the cartilage and non-cartilage tissues. This step may require user interaction or can be fully automatic. Most of the automatic segmentation methods contain an additional pre-processing step, where the bones or bone-cartilage interfaces are extracted, since bone is an important landmark when searching for the cartilage.