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

The identification of pulmonary air bubbles might play a significant role for medical diagnosis of pulmonary pathologies. A method to segment these abnormal pulmonary regions on volumetric data, using a deformable 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.