GPU acceleration for liver enhancement and segmentation

Abstract

This doctoral thesis deepens the GPU acceleration for liver enhancement and segmentation. With this motivation, detailed research is carried out here in a compendium of articles. The work developed is structured in three scientific contributions, the first one is based upon enhancement and tumor segmentation, the second one explores the vessel segmentation and the last is published on liver segmentation. These works are implemented on GPU with significant speedups with great scientific impact and relevance in this doctoral thesis The first work proposes cross-modality based contrast enhancement for tumor segmentation on GPU. To do this, it takes target and guidance images as an input and enhance the low quality target image by applying two dimensional histogram approach. Further it has been observed that the enhanced image provides more accurate tumor segmentation using GPU based dynamic seeded region growing. The second contribution is about fast parallel gradient based seeded region growing where static approach has been proposed and implemented on GPU for accurate vessel segmentation. The third contribution describes GPU acceleration of Chan-Vese model and cross-modality based contrast enhancement for liver segmentation.