Hybrid PET/MRI co-segmentation based on joint fuzzy connectedness and graph cut. (October 2017)

• Record Type: Journal Article
• Authors: Sbei, Arafet; ElBedoui, Khaoula; Barhoumi, Walid; Maksud, Philippe; Maktouf, Chokri
• Is Part Of: Computer methods and programs in biomedicine. Volume 149(2017)
• Page Start: 29
• Page End: 41
• Publication Date: 2017-10
• Subjects: GCsummax -- Iterative relative fuzzy connectedness -- Graph cut -- Co-segmentation -- PET/MRI; Medicine -- Computer programs -- Periodicals; Biology -- Computer programs -- Periodicals; Computers -- Periodicals; Medicine -- Periodicals; Médecine -- Logiciels -- Périodiques; Biologie -- Logiciels -- Périodiques; Biology -- Computer programs; Medicine -- Computer programs; Periodicals; Electronic journals; 610.28
• DOI: 10.1016/j.cmpb.2017.07.006
• Languages: English
• ISSNs: 0169-2607

Abstract

Highlights: Tumor segmentation from hybrid PET/MRI scans. Iterative Relative Fuzzy Connectedness (IRFC) optimizes seeds initialization. Min-cut/max-flow technique allows the boundary smoothing. A visibility weighting scheme was adapted to achieve the task of co-segmentation. The proposed method was efficiently validated on several clinical studies. Abstract: Background and Objective: Tumor segmentation from hybrid PET/MRI scans may be highly beneficial in radiotherapy treatment planning. Indeed, it gives for both modalities the suitable information that could make the delineation of tumors more accurate than using each one apart. We aim in this work to propose a co-segmentation method that deals with several challenges, notably the lack of one-to-one correspondence between tumors of the two modalities and the boundaries' smoothing. Methods: The proposed method is designed to surpass these limits, we propose a segmentation method based on the GC s u m m a x technique. The method takes the advantage of Iterative Relative Fuzzy Connectedness (IRFC) on seeds initialization, and the standard min-cut/max-flow technique for the boundary smoothing. Seed initialization was accurately performed thanks to high uptake regions on PET. Besides, a visibility weighting scheme was adapted to achieve the task of co-segmentation using the IRFC algorithm. Then, given the co-segmented regions, we introduce a morphological-based technique that provides object seeds to standard Graph Cut (GC) allowing it to avoid the shrinking problem. Finally, for each modality, the segmentation task is formulated as an energy minimization problem which is resolved by a min-cut/max-flow technique. Results: The overlap ratio (denoted DSC) between our segmentation results and the ground-truth for PET images is 92.63 ± 1.03, while the DSC for MRI images is 90.61 ± 3.70. Conclusions: The proposed method was tested on different types of diseases and it outperformed the state-of-the-art methods. We show its superiority in terms of assymetric relation between PET and MRI and tumors heterogeneity.