Segmenting the lungs in medical images is a challenging and important task for many applications. In particular, automatic segmentation of lung cavities from multiple magnetic resonance (MR) images is very useful for oncological applications such as radiotherapy treatment planning. Largely changing lung shapes, low contrast and poorly defined boundaries make the lung cavities hard to be distinguished, even in the absence of prominent neighboring structures. In this paper, we utilized a modified geometric-based snake model which could greatly improve the model's segmentation efficiency in capturing complex geometries and dealing with difficult initialization and weak edges. This model integrates the gradient flow forces with region constraints provided by fuzzy c-means clustering. The proposed model has been tested on a database of 30 MR images with 80 slices in each image. The obtained results are compared to manual segmentations of the lung provided by an expert radiologist and with those of previous works, showing encouraging results and high robustness of our approach.
展开▼
