Continuous convex relaxation methodology applied to retroperitoneal tumors

Abstract

In this paper, two algorithms for the segmentation of tumors in soft tissues are presented and compared. These algorithms are applied to the segmentatiion of retroperitoneal tumors. Method: The algorithms are based on a continuous convex relaxation methodology with the introduction of an accumulated gradient distance (AGD). Algorithm 1 is based on two-label convex relaxation and Algorithm 2 applies multilabel convex relaxation. Results: Algorithms 1 and 2 are tested on a database of 6 CT volumes and their results are compared with the manual segmentation. The multilabel version performs better, achieving a 91% of sensitivity, 100% of specificity, 88% of PPV and 89% of Dice index. Conclusions: To the best of our knowledge, this is the first time that the segmentation of retroperitoneal tumors has been addressed. Two segmentation algorithms have been compared and the multilabel version obtains very good results