Multi-scale Statistical Grey Value Modelling for Thrombus Segmentation from CTA

In this paper we present, evaluate, and discuss two multiscale schemes for modelling grey-level appearance in a deformable model for the segmentation of abdominal aortic aneurysm thrombus from CT angiography scans. The methods are initialised with the lumen boundary, and the image-based deformation force is based on a non-parametric statistical grey level model built from training data obtained at multiple scales. The image force direction and magnitude are based on a fit between the trained model and the data. Two multi-scale schemes are used for deformation. In one scheme, the boundary is progressively refined based on training data obtained at each scale, in a coarse-to-fine approach, and in the other, the training data obtained at all scales are used simultaneously to drive the deformation. The two schemes are evaluated and compared to the single scale scheme based on a leave-one-out study of nine patients.