Adipose tissue segmentation in whole-body MRI

Abstract: Adipose tissue volume and distribution is related to metabolic diseases such as diabetes and atherosclerosis. This relationship is in focus for much research, much due to a worldwide increase in obesity. It is in many cases of interest to calculate the amount of adipose tissue in different compartments within the body. Commonly used methods are however prone to introduce errors due to partial volume effects. Previous studies have successfully segmented three adipose tissue compartments from abdominal two-point Dixon fat-water MRI volumes using Morphon registration and atlas segmentation. This thesis extends upon the previous work by enabling segmentation of whole-body MRI volumes and by improving the registration with the use of both fat and water data. Possible methods for bone marrow segmentation are also tested and evaluated. The methods presented seem to be sufficient for creating whole-body volumes from a set of smaller volumes. The adipose tissue segmentation was adequate for subjects with relatively small volumes of adipose tissue, whereas segmentation of subjects with large amounts of adipose tissue require further improvement. Of the evaluated methods for bone marrow segmentation one seemed to perform adequately on all the tested datasets. Due to the few datasets available for testing it was not possible to draw any general conclusions as to how well the presented methods perform.