MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Stopping Power Map Estimation From Cone-Beam CT Using Deep Learning for CBCT-Guided Adaptive Radiation Therapy

J Harms*, Y Lei , T Wang , B Ghavidel , W Stokes , T Liu , W Curran , J Zhou , M McDonald , 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: In intensity-modulated proton therapy (IMPT) protons are used to deliver highly conformal dose distributions, targeting tumors and sparing organs-at-risk. However, due to uncertainties in both patient setup and relative stopping power (RSP), margins are added to the treatment volume during treatment planning, leading to higher doses to normal tissues. Cone-beam CT (CBCT) images are taken daily before treatment, however the poor image quality of CBCT limits the use of these images to position verification. In this work, we use a deep-learning based method to predict stopping power maps from daily CBCT images, allowing for online dose calculation in a step towards adaptive radiation therapy.

Methods: Patients were simulated using a Siemens TwinBeam dual-energy CT scanner, and the mixed-energy CT images were used. The Gammex RMI 467 electron density phantom is used to build a calibration curve that relates measured Hounsfield units (HU) to experimentally-measured RSP for each insert. To train the deep learning algorithm, CT simulation images from 20 head-and-neck cancer patients were converted to stopping power maps. RSP maps are then registered to daily CBCT images to create training pairs, and the algorithm learns a mapping from CBCT to RSP. Leave-one-out validation was used for evaluation, and mean absolute error (MAE), mean error (ME) and normalized cross-correlation (NCC) were used to quantify the differences between the CT-based and CBCT-based RSP maps.

Results: The CT HU-RSP calibration curve has an ME (and MAE) of 0.03±0.03 across 17 materials. The CBCT-based RSP generation method was evaluated on 20 head-and-neck cancer patients. The mean MAE between CT-based and CBCT-based RSP was 0.04±0.02, the mean ME was -0.01±0.03 and the mean NCC was 0.97±0.01 for all patients.

Conclusion: The proposed method provides sufficiently accurate RSP map generation from CBCT images, possibly allowing for CBCT-guided adaptive treatment planning for IMPT.

Funding Support, Disclosures, and Conflict of Interest: This research is supported in part by the National Cancer Institute of the National Institutes of Health under Award Number R01CA215718 (XY).

Keywords

Not Applicable / None Entered.

Taxonomy

IM- Cone Beam CT: Machine learning, computer vision

Contact Email