Accurate segmentation of knee cartilage in magnetic resonance (MR) images to automatically measure thickness values is a basic need in diagnosis and treatment of osteoarthritis. A new fast algorithm (comprising an Edge-enhanced Distance Transform aiding a narrowband Directional Gradient Vector Flow geometric Snake) was evaluated on >20 synthetic and real MR images, including cartilage. We refer to this as the Chessboard Directional Compensated GVF snake, or CDCG. The method integrates general GVF forces with upwind direction constraints and implements a fast directional distance transform outside narrow bands masked over the bone boundaries. The numerical solution is built on a level set approach. The algorithm is robust to weak image features and even invisibly contacting boundaries. Compared with the GGVF Snake, our method is more tolerant of disconnected or fuzzy edges and noise; more precise; and nearly 3 times faster. The approach supports multiple Snakes and allows effective detection of cartilage thinning.