Segmentation of noisy PET images using Bayesian matting

Abstract

Accurate and robust image segmentation is identified as one of the most challenging issues in Positron Emission Tomography (PET). The low spatial resolution, signal dependant noise levels and complex nature of anatomical structures have negative impact on qualitative and quantitative image segmentation analysis. Several unsupervised methods such as, Fuzzy C-Means (FCM) clustering, active contour modeling are usually used in segmenting medical images. However, these methods are sensitive to both noise and intensity inhomogeniety, as they ignore the spatial information. In this paper, we propose a methodology which segments noisy PET images incorporating an efficient denoising technique using transform domain filters to remove the noise followed by an active contour method to segment the Region Of Interest (ROI). Finally, the segmented output is fine tuned using Bayesian matting approach. Experimental results show that the proposed approach improves the overall segmentation accuracy.