"Automatic" multimodal medical image fusion

Abstract

This paper addresses the multimodal medical image fusion problem. To deliver the expected fusion accuracy, most of the state-of-the-art fusion algorithms have the typical requirement - a set of fiducial points between the two modalities of the images to "guide" the fusion. This paper aims at removing this requirement, resulting in an "automatic" multimodal medical image fusion methodology based on an innovative model using ElectroStatic Equilibrium theory called ESE that can be used in clinical diagnoses and evaluations with accepted fusion accuracy. By "automatic", it is meant that the fusion algorithm per se does not require fiducial points; it does require a certain form of human interactions in terms of providing users a list of parameter settings at the beginning of the fusion, that are case-based, anatomy-based, and image-modality-based, and it does even allow users to have an option to change the specific values of the parameter settings to accommodate specific clinical needs. This "automatic" approach allows radiologists to save their time/effort in identifying and marking the fiducial points in the images, allows physicians and radiologists to apply their domain expertise more intelligently in "playing with" different parameter settings in a higher level when running this algorithm in diagnoses, and also allows patients to avoid the inconvenience to be placed under the fiducial markings. Preliminary evaluations against one of the existing fusion methods have shown that ESE holds a great promise in future medical applications.