MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

MRI-Based Prostate Proton Radiotherapy Using Deep-Learning-Based Synthetic CT

G Shafai-Erfani*, Y Liu , Y Lei , Y Wang , T Wang , S Tian , A Jani , M McDonald , W Curran , T Liu , J Zhou , X Yang , Department of Radiation Oncology and Winship Cancer Institute, Emory University, Atlanta, GA 30322

Presentations

(Wednesday, 7/17/2019) 1:45 PM - 3:45 PM

Room: Stars at Night Ballroom 2-3

Purpose: To develop and verify a novel deep-learning-based scheme to construct synthetic CTs (SCT)s with adequate precision needed for proton treatment planning for prostate cancer, based on routine MRI as the sole imaging modality.

Methods: We propose a 3D cycle-generative adversarial network (cycle-GAN) framework with dense blocks to be trained for adept nonlinear mapping between co-registered MR and CT image pairs. In training stage, 3D MR image patches were used by a generator to construct SCTs, which in turn were used by an inverse generator, and were assessed by a discriminator. In prediction stage, SCTs were generated for the study cohort of 17 patients. To comprehensively evaluate the efficacy of our method, we explored its performance in: 1) Image quality, using mean absolute error (MAE), peak signal-to-noise ratio (PSNR), and normalized cross-correlation (NCC) metrics; 2) Dosimetric analysis using several DVH endpoints, voxel-based dose difference, and gamma analysis at 1mm/1%, 2mm/2%, 3mm/3% criteria with 10% dose threshold; 3) Proton distal range difference; 4) Individual pencil beam Bragg-peak shift.

Results: Average MAE, PSNR, and NCC among the study cohort were found to be 51.32±16.91 HU, 24.2±2.46 dB, and 0.93±0.03, respectively. Relative differences for DVH endpoints within PTV were generally less than 1%. Mean value of voxel-based dose differences and absolute dose differences were -0.07±0.07% of the prescribed dose. Gamma-index analysis at 1mm/1%, 2mm/2%, 3mm/3% reveals on average 92.39±5.97%, 97.95±2.95% and 98.97±1.62% pass rate, respectively. Mean proton beam distal range shifts and mean Bragg-peak shift were found to be 0.23±0.25 cm, and 0.18±0.07 cm, respectively.

Conclusion: Our method demonstrates competency in constructing SCTs with satisfactory results within the areas of image quality, dosimetric precision, proton distal range and Bragg-peak shift, which suggests the feasibility of further development of an MRI-only workflow for prostate proton radiotherapy.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email