Metacarpal Bones Localization in X-ray Imagery Using Particle Filter Segmentation

Abstract

Statistical methods such as sequential Monte Carlo Methods were proposed for detection, segmentation and tracking of objects in digital images. A similar approach, called Shape Particle Filters was introduced for the segmentation of vertebra, lungs and hearts [1]. In this contribution, a global shape and a local appearance model are derived from specific object annotated X-ray images of the metacarpal bones. In the test data a unique labeling of the bone boundary and the background points and a manual annotation is given. Using a set of local features (Haar-like) in the neighborhood of each pixel a probabilistic pixel classifier is built using the random forest algorithm. To fit the shape model to a new image, a label probability map is extracted and then the optimal shape is obtained by maximizing the probability of each landmark with the Differential Evolution algorithm.