Room: AAPM ePoster Library
Purpose: Daily volumetric imaging is a key component of adaptive proton therapy (APT). While such data can be acquired using cone-beam CT (CBCT), scatter artifacts make uncorrected CBCT images unsuitable for APT dose calculation. Scatter artifacts can be accurately corrected with Monte Carlo (MC) simulations, however MC is too computationally demanding for real-time use. In this work, we evaluate the performance of a MC trained U-shape deep convolutive neural network (U-net) to correct scatter artifacts in CBCT images and enable fast and accurate APT dose calculation.
Methods: CBCT projections are generated for 48 head and neck patients using a GPU accelerated MC code (MCGPU), providing a total of 17,280 pairs of uncorrected and scatter-free projections. A U-net is trained to estimate scatter distributions from uncorrected CBCT projections. Patients are distributed in training (29), testing (9) and validation (10) sets. The accuracy of the scatter correction is experimentally evaluated using CT and CBCT images of an anthropomorphic head phantom. The potential of the method for head and neck APT is assessed by comparing proton therapy dose distributions calculated on scatter-free, uncorrected and scatter-corrected CBCT images.
Results: The mean HU difference between scatter-free and scatter-corrected images is 0.8 ±26 HU, compared to 49.4 ±119 HU for the uncorrected images. In the head phantom, the root-mean square difference of proton ranges calculated in the reference CT and corrected CBCT is 1.41 mm. The average 2%/2mm Gamma pass rate for proton therapy plans optimized in the scatter free images and re-calculated in the scatter-corrected ones is 98.89%. All CBCT projection volume could be corrected in less than 4 seconds.
Conclusion: The potential of a MC trained U-net to correct CBCT images for head and neck APT is demonstrated. The method achieves an accuracy similar to a full Monte Carlo scatter-correction, while being considerably faster.
Funding Support, Disclosures, and Conflict of Interest: Support from NSERC (PDF-532784 - 2019) and NCI (R01 229178)