Multi-modal Multi-Atlas Segmentation using Discrete Optimisation and Self-Similarities

Abstract

This work presents the application of a discrete medical im-age registration framework to multi-organ segmentation in different modalities. The algorithm works completely auto-matically and does not have to be tuned specifically for dif-ferent datasets. A robust similarity measure, using the local self-similarity context (SSC), is employed and shown to out-perform other commonly used metrics. Both affine and de-formable registration are driven by a dense displacement sam-pling (deeds) strategy. The smoothness of displacements is enforced by inference on a Markov random field (MRF), using a tree approximation for computational efficiency. Consen-sus segmentations for unseen test images of the VISCERAL Anatomy 3 data are found by majority voting. 1