Kidney Segmentation in Scintigraphic Sequences Data using Multi-Agent Approach
Aim: The aim of this paper was to define a robust method, allowing the effective detection of structures whose contours change during the time. Methods: We integrated an agent model based on a spatiotemporal descriptor for the points of interest detection and Fast Marching Method used for kidney segmentation and tracking in scintigraphic sequences. The proposed agent model contains both type of agents: supervising and explorer agents. As soon as the spatio-temporal descriptors HOG3D detect the points of interest, the supervising Agent create explorer agent on each point of interest in the image. All explorer agents evolve according to the Fast Marching Method. In case of conflict between two agents, the supervising agent should intervene immediately to manage this conflict. Results: Our system was applied experimentally on synthetic sequences then on real scintigraphic sequences for the segmentation of the two kidneys. We have found an acceptable performance in the segmentation phase, approved and validated by experts in nuclear medicine. Conclusions: Our method achieves high accuracy in kidney segmentation, considerably reducing the time and labor required for contour delineation. In addition, the method can be expanded to 3D segmentation directly without modification.