Using Shape-Aware Models for Lumbar Spine Intervertebral Disc Segmentation

High incidence cases associated with back pain include intervertebral disc degeneration (IDD), or disc herniation, in the spinal lumbar region, as well as sciatica, pain in the legs due to IDD. This research aims to provide a more accurate and robust segmentation scheme for identification of spine pathologies, to assist with spine surgery planning and simulation. We are developing a minimally supervised 3D segmentation approach of lumbar spine herniated discs for MRI scans that exploits weak shape priors encoded in simplex mesh active surface models. In the event that the internal simplex shape memory influence hinders detection of pathology, user-assistance is allowed to turn off the shape feature and guide model deformation. We propose use of weak shape priors as a precursor to, and incorporation of, a shape-statistics feature for landmark-based semi-automatic segmentation of healthy intervertebral discs, and ultimately, for segmentation of vertebrae. Our framework enables the application of shape priors in the healthy part of the anatomy, and the disabling of these priors where inapplicable. Results were validated against expert-guided segmentation and demonstrate promising results with absolute mean segmentation error of less than 1 mm.