Segmentation of the Midbrain in Transcranial Sonographies using a Two–Component Deformable Model

We present an approach for automatic segmentation of the midbrain and substantia nigra in 2D transcranial sonographies (TCS). It combines principles from both deformable shapes and Active Contour Models (ACM). A set of ACM is coupled with a two–component Finite Element Model of the midbrain anatomy, and this shape representation is used in a deformable shape search for the midbrain. A successful search allows automatic initialisation of the ACM for the exact segmentation of the substantia nigra. For improving convergence independent from the initial estimate, deformations are driven by a classifier that distinguishes regional image features, and in this way steers the deformation to inter–tissue boundaries. Experimental results are promising and show that our approach can segment the midbrain from TCS automatically and accurately, compared with other image segmentation methods and manual segmentation.