A Hybrid Approach to the Segmentation of Lungs from Digital Chest X-Rays

摘要

The accurate segmentation of lungs is a prerequisite in many Computer Aided Diagnostic systems. If the algorithm used in such systems is guided by the domain information, apart from the image information, then the results will be closer to human interpretation. There are multiple approaches suggested in the literature to solve the problem, but an accurate and faster approach is yet to be developed. In this paper a method to segment out the lungs using a mix of rule based approaches and shape model is presented. The whole task of lung boundary detection is divided in to smaller sub-tasks, each further approached with methods guided by the knowledge of anatomy. The algorithm has been tested on images taken from the available public database, and the results derived are promising. A similar approach can be applied to other segmentation tasks in medical domain or elsewhere, where the object of interest lacks a standard shape and location feature.