Similarity Measures in Medical Image Registration A Review Article

Abstract

Image registration is one of the most important problems in medical image analysis. It refers to the process of geometric alignment between two images based on correspondence. A crucial step in medical image registration process is to determine a similarity measure. There are various similarity measure techniques in this field that are applied in different registration applications. Selecting an effective similarity measure is a challenging problem, and this choice affects the accuracy of the registration results. According to past research, the similarity measures have extended from traditional to deep learning based methods. Our goal in this paper is to provide a literature review on various similarity measure techniques in medical image registration, classify them, and introduce main challenges. Thus Similarity measure techniques based on various registration approaches have been classified into two main classes and several subclasses namely distance based, correlation based, and information based in traditional methods; and statistical based, learning based, and similarity measure based loss function in learning based methods. Based on this classification, methods are introduced and each category is evaluated based on accuracy, speed, robustness, and complexity. Finally, recognizing and evaluating the different similarity criteria will help to select the appropriate similarity measure according to the intended application.