Automated lung CT image segmentation using kernel mean shift analysis

Abstract

With improvement technology in medical science, using methods based on machine vision technics become more considerable. Automatic methods in clinical practice provide fast and accurate analysis of scanned images indisease diagnosing. Within these methods, medical image segmentation plays more important role in separation of defective cellular from healthy organs. By performing an accurate segmentation, medicines can detect indistinguishable parts of scanned images, classify them and search over a database to find similar cases. In this paper; we proposed an efficient and adaptive method for segmentation of lung CT images. The proposed algorithm uses adaptive mean shift method that estimate the bandwidth parameter by using fixed bandwidth estimation. Because of close dependency of kernel density estimation method to the bandwidth parameter, Particle Swarm Optimization algorithm is used to optimize this parameter. This method is achieved better segmentation that can carry out small lung nodules and detecting regions within an CT image. Experimental results on a large dataset of diverse lung CT images prove that the proposed algorithm accurately and efficiently detects the borders and regions of lung images.