Segmentation of flexor tendons within carpal tunnel from magnetic resonance image

Carpal tunnel syndrome (CTS) is one of the most common peripheral neuropathies. For evaluation of the CTS, segmentation of carpal tunnel and its contents from magnetic resonance (MR) images is rather important. In this paper we propose a new segmentation framework for the flexor tendons within the carpal tunnel on MR images. The proposed method consists of two stages including construction of reference tendon contour (RTC) and registration-based segmentation. Initially, we segment the flexor tendons from the most proximally cross-sectional image using watershed algorithm. The resulting segmented tendons then serve as the RTCs and are applied to the registration-based segmentation stage. Each of the RTCs in this stage is registered to its distally adjacent image by a rigid transformation and then its shape is refined by snake deformation. The registration-based segmentation is sequentially performed on each pair of adjacent cross-sectional images along the proximal to distal direction until the most distal image is segmented. Consequently, the flexor tendons in the entire MR volume can be automatically segmented. In the experiments three MR volumes were used to validate the accuracy of the proposed method. Compared to the manual results, our proposed method showed good agreement on the tendon segmentations with dice similarity coefficient above 0.8.