Pyramid-transform based lung-tissue MRI image segmentation

This paper presents a framework based on a pyramid-transform for the segmentation of lung tissue in MRI images. We show that a multi-scale framework is beneficial to the segmentation of tissues; specifically, we show that removal of a large-scale anatomical structure within the lung can be accomplished without the need to provide interactive manipulation of objects within the image. Moreover, we show that higher performance in boundary detection can be achieved using an edge detector operator in combination with a steerable filter bank compared to using the edge detector directly on the image. Currently, early stages of the segmentation are based on the assumption that the lungs are embedded in a high intensity background, as is the case in MRI upper abdomen images. Generalization of the proposed methods to other types of images will require further development of the algorithms but, based on the results of large-scale structure removal presented here, we postulate that further use of a pyramid structure can further benefit the process of auto-segmentation