Investigation of methods for intensity based kidney registration with MRI

Abstract

Background: The kidneys are an important part of the body and an absence of the functions a kidney provide is be lifethretening. To be able to provide correct treatment, data needs to be collected. A safe and efficient way of collecting this data is the use of MRI which can be used to asses both function, structure and pathopysiological changes. But to be able to access this data, images need to be collected over different time instances. This results in a issue: the movement of the kidneys due to breathing and heartbeats introduced during this time period of scanning complicates the analysis of the data. To compensate for these motions, image registration can be used. This is the process of aligning the images so that they have a pixel to pixel correspondance. Aim: The purpose of this study was to assist in the search for an automated registration method that can compensate for motions in kidney MRI images. Method: The method used in this project is literature study and performance study. The literature study was used to get a broad understanding of the topic and to gather information about promising techniques. These techniques were then implemented in the performance study. The main parts investigated was: i) the registration process, ii) choice of transform, iii) choice of similarity measure. Results and conclusion: Kidney registration is a complex task including many possible combinations of settings. Mutual information (MI) outperformed Cross-correlation (CC) as a similarity measure. When it comes to different combinations of registration processes and transformation models, the result does not show a clear optimal combination. Thus, this study urges the complexity of selecting registration settings and there are many areas where further research is needed.