Cross-modality-guided contrast enhancement on liver segmentation

Obtaining rapid and accurate segmentation of organs remains an important and challenging task. Liver segmentation algorithms are slow and inaccurate due to noise and low-quality images from abdominal computed tomography (CT) scans. Chan- Vese is an image segmentation method with great robustness against noise, however it is quite slow due to the computation of complex partial differential equations, especially for large medical data sets. The very low contrast of the liver obtained by CT images reduces the overall quality of liver segmentation. In this work we propose two improvements to obtain optimal liver segmentation: contrast enhancement using a guided crossmodality technique as a preprocessing prior to liver segmentation; as well as the GPU paralysis of the Chan-Vese method. The results present a notable improvement in the sensitivity and precision of liver segmentation, as well as a much faster calculation, desirable to obtain liver segmentation in real-time during the surgery.