R. Bundschuh, G. Delso, A. Martinez-Möller, S. Nekolla, S. I. Ziegler, M. Schwaiger Segmentation of lesions with inhomogeneous activity distribution in PET Journal of Nuclear Medicine 2008, 49 (suppl. 1), p381. Proceedings of Annual Meeting of Society of Nuclear Medicine (SNM 2008), New Orleans, LA, USA, June 2008 (bib)
	Objectives Tumor volume segmentation based on PET is of great interest in radiotherapy treatment planning. Currently most algorithms are based on a threshold which seems to be appropriate in phantom studies with homogeneously filled spheres. In general, tumors can not be considered homogeneous. Therefore we analyzed segmentation methods for inhomogeneous lesions. Methods: 3 different methods were compared: threshold of maximum (M1) or mean (M2) activity uptake in the lesion, and a gradient method based on the Marr-Hildreth operator (M3). The methods were applied to data from 4 phantom measurements using radioactive wax simulating inhomogeneous lesions (40 mm diameter) of different activity distribution. In addition, inhomogeneous lesions of different size (10 – 50 mm diameter) in the lung of a patient were simulated by Monte Carlo methods. Practicability of the algorithms was tested with 10 data sets of patients with solitary lung lesions. Results: In the phantom study lesion volumes determined by of M1 (M2; M3) were 54 (48; 3)% too small in case of higher activity concentrations in the center of the lesion. With lower uptake in the center of the lesion, M3 showed with 20% error only a slightly improved result compared to M1 (26%) and M2 (30%). For homogeneous lesions the results of all three methods were within 3% of the real volume. Similar results were found with the Monte Carlo data: All methods worked well in the homogeneous case, M1 and M2 resulted in underestimation of the volume of up to 77% in inhomogeneous lesions. For M3 this underestimation was much lower with a mean of 13% (range 5-24%). Especially in lesions >20 mm, M3 showed better results. Conclusions: In case of inhomogeneous uptake, current threshold based segmentation methods fail. Gradient based methods are superior in this case.
	This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each authors copyright. In most cases, these works may not be reposted without the explicit permission of the copyright holder.