A localised elastic net technique for lung boundary extraction from magnetic resonance images

This paper is concerned with the key process of segmentation of the input image and/or extraction of desired features. In computer vision systems, with particular application to segmenting the human lung boundaries from magnetic resonance images of human torsos. In this application, once the basic segmented boundaries have been constructed (in this case by a multi-layer perceptron), further image processing is required to remove noise and complete the segmented boundaries with the eventual aim of removing the need for human expertise from the lung extraction process. In this paper, we present some significant new results on a deformable model (based on Durbin and Willshaw´s elastic net) to replace the existing post-processing techniques. A method for comparing the similarity between the expert and computer extracted boundaries is introduced, and it is shown that the new algorithm does indeed offer significant progress on the general problem area. Finally, some areas for short to medium term further development are identified