Automatic Segmentation Of Meniscus Using Locally-Weighted Voting Based On Multi-Atlas And Edge Classification In Knee Mr Images

We propose an automatic segmentation of meniscus from knee MR images using multi-atlas segmentation and patch-based edge classification. In order to prevent mis-registration of the meniscus with the large structures, meniscus is localized after segmenting the bone and articular cartilage in advance. To segment the meniscus with large shape variations and to remove leakage to the collateral ligaments, meniscus is segmented using shape-and intensity-based locally-weighted voting (LWV) and patch-based edge classification. Experimental results show that the Dice similarity coefficient of the proposed method provides an average of 80% or more of the two manually outlined results and is improved over the multi-atlas-based LWV.