Local Shape Modelling for Brain Morphometry using Curvature Scale Space

Abstract. Methods to capture the morphological variability of the human brain are needed to improve the diagnosis and treatment of neuro-degenerative diseases. This work presents an approach for local shape modelling that models self-similar curves to derive shape information. The method is based on two main techniques: Curvature Scale Space (CCS) as a way to represent contours and obtain a set of meaningful shapes to be analysed, and Statistical Shape Models to characterize the shape variation. The evolution of the curve in the CSS is used to partition the contour into locally similar parts which are then aligned and their variability modelled by PCA. We present the method and demonstrate results on a set of white-matter brain contours.