Automated segmentation of the menisci from MR images

Pathologic processes active in early-stage knee joint osteoarthritis may also affect the integrity of the crescent-shaped fibrocartilagenous structures called menisci. Magnetic resonance imaging can allow the detection of these structural changes, however, large-scale clinical application remains limited by tedious and labor-intensive techniques for volumetric measurement. Towards automating these quantitative measurements, we have currently developed a scheme that allows the automatic segmentation of the menisci from MR images of healthy knees. This scheme utilizes prior automatic bone and cartilage segmentations to provide spatial localization, before shape model fitting and tissue classification are used to segment the menisci. The accuracy and robustness of the approach was experimentally validated using a set of 14 fat suppressed Spoiled Gradient Recall MR images. An average Dice Similarity Coefficient of 0.75 and 0.77 was obtained for the medial and lateral meniscus, illustrating the accuracy of the approach, while the coefficient of variation for volume was 2.29 and 1.50, respectively.