Abstract: The aim of the present work is the analysis and mining of the informative content related to pathological liver tissues, when acquired by triphasic CT, with the proposal of a data-fusion approach, which is able to visualize and show up such a content in the best way as a support to the medical diagnosis. Since the huge amount of CT volumes to be analyzed in a limited time is the major cause of sensitivity loss during the diagnosis process, a better chance of detection and localization of the pathology can be derived from the method here proposed. This method can be a valid support to the current medical practice, even in the cases where pathology is at the very early stage and has a large probability to be missed by a visual inspection. As expected when analyzing the three phase volumes, one can note that the injection of a contrast agent causes significant changes in the radiological finding for both pathological and healthy parts of the liver. Thanks to a specific statistical analysis performed on a training dataset of real cases, the described study was focused on the characterization of hepatocellular carcinoma (HCC) tumor tissues and liver tissues. In order to detect and discriminate tumor from liver parenchyma, we here propose using both steady-state and dynamic features. Some common patterns have been observed suggesting rules, which have been confirmed by radiology specialists. Based on the rules and the best discriminant power of some of the characterizing features, a new color data fusion approach is then proposed and discussed which improves the mass visibility while increasing contrast with respect to surrounding parenchyma.

Keywords: Triphasic CT, tissue characterization, feature analysis, data fusion, color distance.