Learning to Locate Cortical Bone in MRI

• Published in: Machine Learning in Medical Imaging pp. 168 - 175
• Main Authors: Hermosillo, Gerardo, Raykar, Vikas C., Zhou, Xiang
• Format: Book Chapter
• Language: English
• Published: Berlin, Heidelberg Springer Berlin Heidelberg
• Series: Lecture Notes in Computer Science
• Subjects: non-rigid registration; sparse logistic regression; template features; whole-body MR imaging
• ISBN: 9783642354274; 3642354270
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-642-35428-1_21

Automatic analysis of MR images requires the correct identification of the various tissues within them. Cortical bone is the most challenging tissue to identify in MR. We present an algorithm to automatically predict the cortical bone locations from whole-body MR Dixon images. Our algorithm combines local information from MR with global information borrowed from exemplar patients with co-registered MR and CT images. The local information is calculated using a classifier trained to discriminate bone from soft tissue using new multi-image template features. The global information is incorporated by retrieving annotated bone maps of the exemplars using a new, non-rigid registration algorithm. We combine the local and global information by an iterative filtering precedure.