Medical image registration is an important but often challenging aspect for clinical image analysis. It has applications in treatment planning requiring image fusion, or inter-subject atlas based analyses, as well as longitudinal analyses. Spine registration presents extra challenges because of the variability in the field of view (FoV) of the spinal column between different image series and many vertebrae having a similar appearance leading to many local registration minima. To help improve spine registration robustness, we generate a labelled dataset of cervical spine magnetic resonance imaging (MRI) and successfully apply a Mask R-CNN model to localize and label vertebra. An automated method to generate labelled bounding boxes and masks can then be used to seed initial alignment or crop to appropriate FoV for subsequent affine and deformable spine MRI registration.
PDF