3D reconstruction of intervertebral discs from T1-weighted magnetic resonance images

Low back pain is a current and increasing problem closely related to intervertebral disc degeneration, which is responsible for over 90% of spinal surgical procedures. In clinical routine, clinicians base their diagnosis of disc degeneration on 2D analysis of Magnetic Resonance (MR) images. In this work, an automatic 3D segmentation method, based on active shape models, is presented for both degenerated and normal intervertebral discs. A database of 25 intervertebral discs was used to semi-automatically build a shape statistical model and intensity models. Then, a 3D reconstruction was achieved by using those models to deform an initial shape. The method was evaluated using the 25 intervertebral discs and a leave-one-out cross validation, resulting in a mean shape accuracy of 1.6mm and a mean dice similarity index of 83.6%. This automatic and accurate 3D reconstruction method opens the way for an improved diagnosis of disc degeneration.