Development of a New Image Registration-Based Method for Tracking Vertebral Motion in Kinetic Sagittal Mri

Conflicts of interest: No conflicts of interest Funding: Exeter NIHR Biomedical Research Centre (BRC) Background Nonspecific low back pain (NSLBP) is highly prevalent, with a significant societal and personal impact, yet the underlying mechanisms remains largely unknown. Evidence suggests a role for altered mechanical function: from local tissue changes to altered coordination patterns. It is important to move beyond static imaging in NSLBP to understand possible dynamic processes. This presentation focusses on the development of image registration-based tracking of vertebral bodies in dynamic MRI images. Methods Ten healthy participants underwent imaging in a 3T Siemens Magnetom MRI scanner using HASTE and TRUFISP sequences. They flexed and extended their lumbar spines passively or actively. 30–40 frames were obtained per session. Image-registration parameters tested were: cross-correlation vs. normalised mutual information similarity measures, and gradient descent vs. exhaustive search optimisation. Imaging preprocessing methods tested were: histogram matching, contrast stretching, difference of Gaussians (DOG), Gaussian blurring, and gradient filters: Sobel, Canny, Prewitt, Farid, Scharr and Roberts. Image registration results were visually assessed, and similarity space was inspected as a function of displacements and rotations. Results Inspection of the similarity space for the tested parameters indicated that MRI image artefact (aorta and CSF) was the principle reason for similarity minima not corresponding to the correct registration, hence failure. Additionally, multiple local minima resulted in the failure of gradient descent optimisation. DOG was most successful at overcoming the problem of artefact along with a modified exhaustive search method. Conclusion Vertebral body tracking is possible in high-speed MRI images despite MRI image artefact. Future directions to reduce need for manual tuning of parameters will be presented.