Fast Segmentation of Focal Liver Lesions in Contrast-Enhanced Ultrasound Data
Spyridon Bakas, Chatzimichail, Katerina, Labbe, Bastien, J.A.Hunter, Gordon, Sidhu, Paul S., Dimitrios Makris · Research Repository (Kingston University London) · 2014
Assessment of focal liver lesions (FLLs) in Contrast-Enhanced Ultrasound data requires initialisation tasks that are currently performed manually by experienced radiologists. These tasks lead to subjective results, are time-consuming and prone to misinterpretation and human error. This paper describes an attempt to improve this clinical practice by proposing a complete pipeline for the automation of initialisation tasks, such as the identification of a frame where the FLL is well-distinguished, the segmentation of the FLL and the conical area including the ultrasonographic image. The currently proposed novel contribution to automate the FLL segmentation is a fast two-step method, initialised only by a single seed-point, which firstly approximates the FLL by an ellipse and then further refines its shape by iteratively classifying boundary pixels.