Segmentation of Medical Image Data using Level Set Methods

The field of medical image analysis is becoming an increasingly important part of the medical profession. Advancements in the field of medical imaging techniques results in images and volumes with an increasing level of detail. Effective methods are needed to extract information from this ever increasing ammount of data, making the field of image analysis more important than ever. Segmentation is an important part of the medical image analysis process. It is used to extract visualize and process relevant anatomical structures within the body. In this project we explore a spesific segmentation approach known as the level set method to extract medical data. We wanted to explore its ability to extract data from volumes of different modailites, such as CT and MRI. The level set method was implemented using the sparse field approach which is a version optimized for serial execution on the CPU. In addition we explored the possibility of parallelizing it using CUDA on the GPU. The results shows that the implemented sparse field method produces good results and is exellent at preventing leakage where other similar methods would struggle. However, level set methods have some problems segmenting images with low variance, causing leakages, which is also present in the implemented sparse field algorithm. The program was parallelized in the GPU using the CUDA technology. The sparse field method is however optimized for serial implementation, which resulted in little performance increase