(Tuesday, 7/16/2019) 10:00 AM - 10:30 AM
Room: Exhibit Hall | Forum 2
Purpose: To develop and validate a deep-learning-based method to derive electron density information from routine abdominal MR images for potential MRI-based liver photon and proton treatment planning.
Methods: A novel 3D dense-cycle generative adversarial network (dense-cycle-GAN) was proposed to capture the structural and textural relationship between CTs and MRs for synthetic CT (sCT) generation. A novel compound loss function was employed to differentiate the structure boundaries with significant HU variations. The proposed dense-cycle-GAN consisted of 4 generators with dense-blocks and 2 discriminators. A cohort of 21 patients with co-registered abdominal CT-MR pairs was used to evaluate the algorithm by leave-one-out cross-validation. CT-based plans for photon and proton radiotherapy were created separately on Eclipse and RayStation and then evaluated on sCTs for dosimetric comparison.
Results: Imaging endpoints including the mean absolute error (MAE), peak signal-to-noise ratio (PSNR), and normalized cross-correlation (NCC) were 72.87±18.16 Hounsfield unit (HU), 22.65±3.63 dB and 0.92±0.04, respectively. The average pass rate of gamma analysis was 100% with 2%/2 mm acceptance criteria for the photon plans and 97.22±3.90% for the proton plans. No significant differences were observed in PTV and OAR DVH metrics including D10, D50, D95, Dmin, Dmean, and Dmax (p > 0.05) for both radiation modalities. For the proton plans, the median absolute range difference was 0.17cm, and 95% of the beam range differences fell into the MGH acceptance criteria. The median individual pencil beam Bragg peak shift was 0.12 cm.
Conclusion: This work generated abdominal sCT from routine MR images based on our proposed dense-cycle-GAN algorithm. Proton dose calculations are more sensitive to HU accuracy than photon plans. The image similarity and dosimetric agreement between the sCT and the original CT warrant further development of an MRI-only workflow for both photon and proton liver radiotherapy.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
