作者: Abolfazl Mehranian , Habib Zaidi
DOI: 10.1109/NSSMIC.2014.7430926
关键词: Computer vision 、 Correction for attenuation 、 Attenuation 、 Mixture model 、 Artificial intelligence 、 Iterative reconstruction 、 Algorithm 、 Markov random field 、 Gaussian 、 Computer science 、 Prior probability 、 Image segmentation
摘要: The maximum likelihood estimation of attenuation and activity (MLAA) has been proposed to jointly estimate from emission data only. In this paper, we an improved MLAA algorithm by imposing MR spatial CT statistical constraints on the using a constrained Gaussian mixture model (GMM) Markov random field (MRF) smoothness prior. We compare MLAA-GMM with algorithms Rezaei et al Salomon as well 4-class MRAC method. Dixon images were segmented into outside air, fat soft tissue classes low-intensity class corresponding air cavities, bone susceptibility artifacts. To eliminate miss-classification bones surrounding tissue, unknown was expanded co-registered probability map. A 4 Gaussians (air, fat/soft bone) used for class, while unimodal others. evaluated simulation clinical datasets. bias in estimated against CT-based correction. Our results show that MLAA-Rezaei suffers scale noise problems. performance MLAA-Salomon is also affected depends highly quality segmentation, especially at air/bone interfaces vertebra. It demonstrated effectively exploits prior information, thereby noise-, crosstalk- scale-free maps. PET analyses showed outperformed corrected Therefore, method can pave way toward accurate emission-based TOF PET/MR imaging.