Graph cut-based segmentation for intervertebral disc in human MRI.

We introduce an automated algorithm for the 2D segmentation of both healthy and degenerated lumbar intervertebral discs (IVD) from T2-weighted Turbo Spin Echo(TSE) sagittal spine Magnetic Resonance Images (MRIs). Our approach employs a fast algorithm addressing the s-t max-flow/min-cut problem, incorporating anatomical knowledge of soft tissues in the human body. In the initial phase, preprocessing is applied to the input image to eliminate intensity inhomogeneity and noise. A graph is then constructed from the image pixels, and seed points are automatically initialised using a growing bounding box. In the second phase, the method applies the s-t max-flow/min-cut algorithm to separate an IVD from the background. This method effectively detects degenerated and healthy IVDs by applying the s-t max-flow/min-cut algorithm within a directed graph. The polynomial time complexity of this approach enables the exploration of a globally optimal solution, eliminating the need for user interaction in seed point selection. Validation of the algorithm on a dataset of 15 patients demonstrates its efficient segmentation performance, achieving a Dice Similarity Coefficient (DSC) of 89%.