Segmentation of sputum color image for lung cancer diagnosis based on mean shift algorithm

This paper presents the mean shift segmentation algorithm for segmenting the extracted sputum cells into nuclei and cytoplasm regions. The segmentation results will be used as a base for a Computer Aided Diagnosis (CAD) system for early detection and diagnosis of lung cancer. The mean shift is a mode seeking process on a surface design with a kernel. Also it will be used as a strategy to perform multistart global optimization. The histogram analysis is used to find the best distribution of the nuclei and cytoplasm sputum cell pixels and to find the best color space that can be used to perform the mean shift segmentation. The Mena shift method offers better performance compared to other segmentation algorithm including Hopefield Neural Network (HNN). The new method is validated on a set of manually defined ground truths sputum images.