Fast automated segmentation and visualisation methods for MR images of the knee joint in arthritis

Abstract

To highlight the diagnostic information in MR slice images of early knee joint damage, due to osteoarthritis (OA) or trauma, we have developed techniques for automated segmentation and rendering of the major joint structures. Using a 1.5 T GE body scanner we obtained sagittal T1-weighted spin echo images from 4 volunteers and 2 patients with OA. Segmentation was performed using MIDAS (Medical Information Display and Analysis System: Imperial College). From seed points on a midline "training" slice, the program searched all slices, locating the boundaries of the femur, tibia, patella, menisci and cruciate ligaments. The overall process was many times faster than manual segmentation. The joint components were then rendered as a composite 3D object allowing "virtual dissection" by the user. Osteophytes and soft-tissue abnormalities were revealed, viewable from any angle. The authors are now using newer MR sequences to detect the articular cartilage so that this can be added to the 3D model to indicate damage from OA. They believe that refinement of these techniques will enable visualisation of early changes to the joint structures, proving clinically useful in diagnosis and management of OA.