A COMBINED FEATURE ENSEMBLE BASED MUTUAL INFORMATION SCHEME FOR ROBUST INTER-MODAL, INTER-PROTOCOL IMAGE REGISTRATION

Abstract

In this paper we present a new robust method for medical image registration called combined feature ensemble mutual information (COFEMI). While mutual information (MI) has become arguably the most popular similarity metric for image registration, intensity based MI schemes have been found wanting in inter-modal and interprotocol image registration, especially when (1) significant image differences across modalities (e.g. pathological and radiological studies) exist, and (2) when imaging artifacts have significantly altered the characteristics of one or both of the images to be registered. Intensity-based MI registration methods operate by maximization of MI between two images A and B. The COFEMI scheme extracts over 450 feature representations of image B that provide additional information about A not conveyed by image B alone and are more robust to the artifacts affecting original intensity image B. COFEMI registration operates by maximization of combined mutual information (CMI) of the image A with the feature ensemble associated with B. The combination of information from several feature images provides a more robust similarity metric compared to the use of a single feature image or the original intensity image alone. We also present a computer-assisted scheme for determining an optimal set of maximally informative features for use with our CMI formulation. We quantitatively and qualitatively demonstrate the improvement in registration accuracy by using our COFEMI scheme over the traditional intensity based-Mi scheme in registering (1) prostate whole mount histological sections with corresponding magnetic resonance imaging (MRI) slices; and (2) phantom brain T1 and T2 MRI studies, which were adversely affected by imaging artifacts