A Level-Set Based Volumetric CT Segmentation Technique: A Case Study with Pulmonary Air Bubbles

Abstract

The identification of pulmonary air bubbles plays a significant role for medical diagnosis of pulmonary pathologies. A method to segment these abnormal pulmonary regions on volumetric data, using a model deforming towards the objects of interest, is presented. We propose a variant to the well known level-set method that keeps the level-set function moving along desired directions, with an improved stopping function that proved to be successful, even for large time steps. A region seeking approach is used instead of the traditional edge seeking. Our method is stable, robust, and automatically handles changes in surface topology during the deformation. Experimental results, for 2D and 3D high resolution computed tomography images, demonstrate its performance. © Springer-Verlag 2004.