Automatic femur segmentation and condyle line detection in 3D MR scans for alignment of high resolution MR

This paper describes an automatic algorithm to extract the knee frame of reference from 3D MR isotropic scans. The method ultimately seeks to determine two lines that are tangent to the bottom of the condyles in an axial and a coronal plane. It consists of three major parts, initial detection of the knee joint using Hidden Markov Models, femur segmentation using Random Walker segmentation, and finally condyle detection. We demonstrate on 30 datasets that our algorithm is very robust and performs at the same level as a human reader.