Multi-Atlas based Hip Bone Segmentation from Routine Clinical Hip MRIs of Osteoarthritis Patients

Abstract

Magnetic Resonance Imaging (MRI) scans of the hip joint are used to diagnose hip osteoarthritis (OA) in routine clinical procedures. MRIs provide better visualization of biochemical degeneration patterns of bones and cartilage with the progression of the disease. In advanced stages, quantitative assessments of the hip bones are done utilizing computer-assisted techniques such as 3D models of the bones to plan surgical treatments. Bone segmentation is a vital step in constructing accurate 3D bone models from imaging modalities. This study aims to segment proximal femur and innominate bone from routine clinical hip MRIs taken from elderly and OA patients. Images from both cohorts show degenerated bones with Bone Marrow Lesions (BMLs) and Subchondral Bone Cysts (SBCs) with a high prevalence in the OA patients cohort. This study proposes to utilize a multi-atlas based segmentation framework with an intermediate template to segment the bone areas automatically from hip MR images. The proposed method achieved accurate automated segmentations with the mean Dice Similarity Coefficient (DSC) values of 0.938 and 0.897 for the proximal femur and innominate bone, respectively, on OA patient's MR images. Slightly higher accuracy was recorded for the MRIs from asymptomatic individuals (DSC values: proximal femur 0.959, innominate bone 0.898).