Bayesian segmentation of chest tumors in PET Scans using a Poisson-Gamma mixture model
Résumé
This paper presents a Bayesian algorithm for PET image segmentation. The proposed method, which is derived from PET physics, models tissue activity using a mixture of Poisson-Gamma distributions. Moreover, a Markov field is proposed to model the spatial correlation between mixture components. Then, segmentation is performed using an Markov chain Monte Carlo algorithm that jointly estimates the mixture parameters and classifies voxels. The performance of the proposed algorithm is illustrated on synthetic and real data. Experimental results on real chest PET images suggest that the proposed method can correctly segment both small and large tumors.