Traditionally, shape analysis is used for representation and statistical analysis of single objects, and often the goal is to discriminate between two populations of objects. Medial representations are geometric models that describe anatomical structures by defining the topology and shape of a structure's skeleton and then deriving the geometry of the structure's boundary from the skeleton. In this talk, we will discuss how "m-reps" were used in a certain clinical pediatric autism study involving sub-cortical brain structures. In order to differentiate between children with autism and typically developing children, we present a method called "distance weighted discrimination" (DWD) for analyzing multiple objects. It is a method similar to support vector machines and is a process for finding the best hyperplane that separates two populations. We will discuss different features, like shape, volume, and pose, to find out which is more significant for discriminating populations.
“ … methods in image analysis using medial representations Qinying Chen 3/20/2020 Figure 1: The task of category-specific … ”
Abstract
Traditionally, shape analysis is mostly used in representation and statistical analysis of single objects, and the goal is to discriminate between two populations of objects. I will focus on the method present in paper <em>Multi-object Analysis of Volume, Pose, and Shape using Statistical Discrimination</em>. This paper presents a new methodology of discriminant analysis for multiple objects. The discriminant method is called distance weighted discriminant (DWD). It is a method similar to SVM, but it is useful when presenting new and untrained samples. The advantage is its generalization ability in high dimension and low sample sizes settings. Essentially, distance weighted discrimination is a process of finding the best hyperplane that separate two populations. It is a method similar to SVM that uses an optimization method to find the maximum of the distance between the hyperplane and data points. This paper uses data from clinical pediatric autism study that includes a total of 70 samples. Then, using m-rep to discuss different features like shape, volume and pose and to find out which is more significant in discriminating populations.