Active contour segmentation of MR images for quantitative cartilage measurements: computational method, interobserver reproducibility and comparison to manual delineation

Introduction Many joint deseases are related to damages of the articular cartilage layer. These damages may be quantified by measurements of the cartilage thickness distribution throughout the joint surface, supplying the physician with a paramter allowing to stage joint dysfunctions or to monitor therapeutic interventions, such as chondroprotective treatment or cartilage transplantation. With magnetic resonance imaging (MRI), a non-invasive technique has become available to provide accurate quantitative data on cartilage volume [1,2] and thickness [2,3], using highresolution, fat-suppressed 3D gradient-echo sequences and 3D image processing [3]. However, a critical and very time consuming step remains the segmentation of the cartilage from the surrounding tissue. The manual delineation of the cartilage boundaries is not only tedious, but the segmentation result depends strongly on the observer, limiting the accuracy and the reproducibility of the method. Therefore we developed an active contour (snake) [4] approach using a B-spline curve representation [5] that fulfills the requirement of real time performence, the user being able to interactively initialize and controll the algorithm.