Joint Multimodal Segmentation of Clinical CT and MR from Hip Arthroplasty Patients

Magnetic resonance imaging (MRI) is routinely employed to assess muscular response and presence of inflammatory reactions in patients treated with metal-on-metal hip arthroplasty, driving the decision for revision surgery. However, MRI is lacking contrast for bony structures and as a result orthopaedic surgical planning is mostly performed on computed tomography images. In this paper, we combine the complementary information of both modalities into a novel framework for the joint segmentation of healthy and pathological musculoskeletal structures as well as implants on all images. Our processing pipeline is fully automated and was designed to handle the highly anisotropic resolution of clinical MR images by means of super resolution reconstruction. The accuracy of the intra-subject multimodal registration was improved by employing a non-linear registration algorithm with hard constraints on the deformation of bony structures, while a multi-atlas segmentation propagation approach provided robustness to the large shape variability in the population. The suggested framework was evaluated in a leave-one-out cross-validation study on 20 hip sides. The proposed pipeline has potential for the extraction of clinically relevant imaging biomarkers for implant failure detection.