Fuzzy connectedness image co-segmentation for hybrid PET/MRI and PET/CT scans

In this paper, we presented a 3-D computer-aided co-segmentation tool for tumor/lesion detection and quantification from hybrid PET/MRI and PET/CT scans. The proposed method was designed with a novel modality-specific visibility weighting scheme built upon a fuzzy connectedness (FC) image segmentation algorithm. In order to improve the determination of lesion margin, it is necessary to combine the complementary information of tissues from both anatomical and functional domains. Therefore, a robust image segmentation method that simultaneously segments tumors/lesions in each domain is required. However, this task, named cosegmentation, is a challenging problem due to (1) unique challenges brought by each imaging modality, and (2) a lack of one-to-one region and boundary correspondences of lesions in different imaging modalities. Owing to these hurdles, the algorithm is desired to have a sufficient flexibility to utilize the strength of each modality. In this work, seed points were first selected from high uptake regions within PET images. Then, lesion boundaries were delineated using a hybrid approach based on novel affinity function design within the FC framework. Further, an advanced extension of FC algorithm called iterative relative FC (IRFC) was used with automatically identified background seeds. The segmentation results were compared to the reference truths provided by radiologists. Experimental results showed that the proposed method effectively utilized multi-modality information for co-segmentation, with a high accuracy (mean DSC of 85%) and can be a viable alternative to the state-of-the art joint segmentation method of random walk (RW) with higher efficiency.