Dealing with multiple types of expert knowledge in medical image segmentation: a rough sets style approach

The fundamental concept of rough sets, upper and lower approximations, provide powerful ways of representing uncertain boundary of regions in images. However, there exist a few studies that discuss effectiveness of this concept in the field of medical image processing, where domain knowledge of experts plays a key role in determining boundaries between anatomically meaningful regions of interests (ROIs). This paper discusses how the expert knowledge can be manipulated in medical image segmentation, especially, how can one treat multiple types of anatomical knowledge about a ROI, such as morphology and location, using upper and lower approximations.