Automatic Segmentation of Articular Cartilage from MRI : A Multi-Contrast and Multi-Dimensional Approach

Introduction The regional variation in the morphology (thickness or volume) of articular cartilage is frequently used for evaluating the initiation and progression of osteoarthritis [1]. Quantifying regional cartilage thickness or volume requires MR image segmentation (classification) and three-dimensional reconstruction [2]. A number of computational methods have been used to automate the segmentation of articular cartilage from a gray scale MR images taken with a single sequence. Yet, fully automatic segmentation seems to be a difficult goal to achieve. There exist many different MR sequences that utilize tissue properties such as T1 and T2 relaxation times to increase the contrast between cartilage and its surrounding soft tissues in joints. Multiple sets of MR images taken with different sequences provide different contrast mechanisms between tissues and will help separate different tissues [3]. The purpose of this study was to evaluate segmentation of knee articular cartilage automatically from multiple sets of MR images using a support vector machine (SVM) method [4], a kernel-based machine learning algorithm.