A Hybrid EMPCA-Scott Approach for Estimating Probability Distributions in Mutual Information

Mutual information (MI) has been widely used as a similarity measure in medical image registration (MIR) for more than a decade. Preliminarily, a number of bins, which is used in evaluating the probability distribution of grey levels, have influenced the performance of MI. Consequently, it can affect the accuracy of MIR and may lead to a wrong medical diagnosis. In an analysis of images from different modalities, it is more challenge to choose such an optimal number of bins. In order to accomplish such a challenge, we propose a new hybrid method based on the technique of an expectation maximization for principal component analysis (EMPCA)[1] and the concept of Scott's rule [2]. We have analyzed and evaluated our approach based on MRI images: T1, T2 and PD. The experimental results have shown that this approach can find a more appropriate bin number which can improve the performance of MI in terms of its accuracy compared to others.