Anterior cruciate ligament segmentation in knee MR images with superpixel graph cuts

Abstract

Anterior cruciate ligament (ACL) is a soft tissue in knee joint which plays an important role in maintaining knee joint stability and the segmentation of this fibrous tissue is crucial for the diagnosis of ACL tears and surgical guidance in ACL reconstruction surgery. However, an automatic segmentation of ACL from knee magnetic resonance image (MRI) is a challenging task due to 1) similar intensity distribution ACL shares with adjacent soft tissues e.g. PCL and cartilage, which causes leakage of the segmented label into these adjacent soft tissues, and 2) inhomogeneous intensity region inside the ACL in knee MRI. In this thesis, an automatic ACL segmentation from 3D knee MR images using graph cuts is proposed. To overcome the difficulties, we propose two additional constraints for the graph cuts to reflect the 2D and 3D shape information of ACL on its segmentation, and incorporate the superpixels for graph cuts to make the segmentation method robust to intensity inhomogeneity of ACL. As a result, thanks to the proposed shape constraints, the proposed method cuts the leakage of segmented ACL labels and shows robustness to intensity inhomogeneity. Compared to the preceding work on ACL segmentation [1] and the original graph cuts [2,3], the proposed method shows overall improved performances.