Segment-growing hierarchical model for bile duct detection in MRCP

Magnetic resonance cholangiopancreatography (MRCP) is the popular diagnostic imaging sequence for the diagnosis and surgery workup for the pancreatobiliary system and liver. The technique is relatively noisy and suffers from imaging characteristics such as the partial volume effect and varying acquisition orientation, making automatic analysis of the images difficult. This paper explores some of the popular image processing techniques with the goal of selecting suitable features in MRCP images, as a basis for preliminary computer-aided diagnosis systems in biliary structure image reconstruction and disease detection. Visual results and observations are given and analyzed. The findings support that many popular techniques such as texture analysis fail to highlight the structures of interest in MRCP images, whereas multi-scale, multi-resolution and dynamic thresholding achieve better success. The proposed multi-scale combination technique known as the Segment-Growing Hierarchical Model produced good visual results for detection of the bile ducts.