Investigation of heterogeneous tumor response in adaptive radiation therapy for patients with lung cancer

Document Type

Conference Proceeding

Publication Date


Publication Title

Int J Radiat Oncol Biol Phys


Purpose/Objective(s): In PET-guided adaptive radiotherapy (RT), metabolic activity changes at individual voxels cannot be derived until pre-treatment CT (CT1) images are appropriately registered to during-treatment CT (CT2) images. This study aims to investigate the feasibility of using deformable image registration (DIR) techniques to quantify radiation-induced metabolic changes on PET images. Purpose/Objective(s): Five patients with non-small cell lung cancer (NSCLC) treated with adaptive radiotherapy, were selected. PET/CT image-sets were acquired 2 weeks before RT and 18 fractions after the start of RT. DIR was carried out to register the CT1 to the CT2 using two image intensity-based deformable registration methods: B-spline and diffeomorphic Demons algorithms. The resultant displacements on the bounding box of the tumor were then used as boundary conditions a hybrid finite element method (Bspline/Demons+FEM). Bitmap masks were generated for gross tumor volumes (GTVs) in pre-treatment CT images. Each mask was divided into multiple layers from outer to interior, and changes in metabolic activity were calculated on each layer for different deformable registrations. The quality of these registrations was assessed based on the consistency of metabolic tumor index (MTI) calculated on the original and mapped GTVs. Results: MTI changes before and after mapping using B-Spline, Demons, hybrid-B-Spline, and hybrid-Demons registrations were 20.2%, 28.3%, 8.7%, and 2.2% on average, respectively. Average displacement error was 3.5±0.7mm for B-Spline registration and 2.1±0.4mm for B-Spline+FEM registration. Changes of SUVmean between the pre-RT PET and the during-RT PET were 49.3±13.4% and 40.2±13.3% for hybrid-B-Spline and hybrid-Demons deformation maps. For three of the patients with large tumors, SUV changes averaged in the middle layers of the tumors were about 2 times the changes in the central and peripheral layers, while relative changes in the activity were about 1.3 times in these layers, showing less variation. Conclusion: As opposed to image-intensity-based DIR algorithms, hybrid FEM modeling is independent of image intensity changes induced by tumor regression. Consequently the accuracy of the B-Spline and Demons registrations in the tumor region is improved. The hybrid modeling technique makes it possible to compare metabolic activities between two PET images for patients with tumor regression and reduces the uncertainty in comparing associated voxels in different regions of two images.





First Page


This document is currently not available here.