Graph-cut energy minimization for object extraction in MRCP medical images.

Bile duct identification and extraction in magnetic resonance cholangiopancreatography (MRCP) images, is a necessary step in the development of computer-aided diagnosis systems using such images. MRCP is becoming the de facto modality in the diagnosis of biliary diseases and even in the pre-surgical workup for liver transplants. The energy minimization graph-cut method is a proven technique in the extraction of objects in natural images, and even used in 3D reconstruction. This paper proposes several versions of the graph-cut approach for the extraction of the biliary structures in MRCP images. The schemes include a fully interactive lazy snapping method, a manual point selection method for minimal user interaction and an automated phase unwrapping via max flows (PUMA) implementation. The performance of the algorithms vary, but the results support that the scheme is a promising semi-automated object extraction scheme for the significant biliary structures in medical MRCP images.