The accurate and early diagnosis of osteoporosis and the assessment in response to therapy are critical for patient management but still remain a challenge for clinicians. There have been recent advancements in diagnostic imaging techniques to improve the assessment of bone quality. There are several different imaging techniques which can be used for the assessment of bone quality both in vivo and in vitro including multi-detector Quantitative Computed Tomography (QCT), High Resolution peripheral Quantitative Computed Tomography (HR-pQCT), Micro Computed Tomography (μCT), Synchrotron Radiation Micro Computed Tomography(SRμCT), and Magnetic Resonance Imaging (MRI) which each have advantages and limitations. The purpose of this thesis is to develop robust image registration techniques for CT and MR-based musculoskeletal images and determine if there is an improvement in the accuracy of longitudinal studies or an enhancement in the understanding of bone quality by combining images from different imaging techniques.
An automatic inter-modal rigid registration method based on normalized mutual information was implemented to allow for the direct spatial comparison of tissue mineralization distributions of ex vivo bone tissue specimens in μCT and SRμCT images. The registration method successfully aligned images acquired using μCT and SRμCT in five specimens of the femoral head, four specimens of the vertebral body, and five specimens of the proximal tibia. This allowed the first direct comparison of tissue mineral density (TMD) between the two modalities.
A normalized mutual information registration method was applied to a set of 49 radius images and 51 tibia images of postmenopausal osteopenic women acquired on MRI and HR-pQCT. The registration method successfully registered all images and the robustness of the method was established. The amount of cortical porosity identified in the HR-pQCT images that contained bone marrow as visualized on the MR images was then quantified.
Image registration methodologies to align MR images in longitudinal studies were also developed. An automatic registration method based on a mutual information measure was implemented for the alignment of high-resolution MR images of trabecular bone in vivo. The robustness and reproducibly of the registration method was established on MR images of the proximal femur of six normal healthy volunteers. The improvement in measurement accuracy in a longitudinal study was demonstrated on MR images of the proximal femur of twenty-four postmenopausal osteopenic women who were scanned at 0 and 12 months.
The automatic registration method was then extended to prospective registration that allowed follow-up images to be acquired in the same orientation as baseline images. The feasibility of prospective registration for MR images of trabecular bone was demonstrated on the distal tibia of five volunteers and the knee of one volunteer. The prospective registration ensured that the same region was analyzed in both the baseline and follow-up images, saved post processing time, preserved the reproducibility of the trabecular bone parameters, and required no interpolation.
The results of this project suggest that the adoption of image registration into the analysis of musculoskeletal images of bone improves the accuracy, reproducibility, and precision of longitudinal and comparative studies.