Daily proton dose re-calculation on deep-learning corrected cone-beam computed tomography scans.

Area of Science:

• Medical Physics
• Radiotherapy
• Artificial Intelligence

Background:

• Accurate dose calculation is crucial for adaptive proton therapy.
• Synthetic CT (sCT) generation from cone-beam CT (CBCT) needs stable performance across treatment fractions.

Purpose of the Study:

• To evaluate a 3D deep learning (DL) network for sCT generation from CBCT in prostate cancer patients.
• To assess the feasibility of DL-based sCT for adaptive proton therapy over the full treatment course.

Main Methods:

• A 3D DL network was trained and tested using data from 25 prostate cancer patients.
• Generated sCT images were compared to planning CT and repeat CT (reCT) for CT number accuracy, proton range, and dose distribution.
• Evaluated CT number accuracy in ROIs (bladder, prostate, femoral heads), proton range accuracy, and dose trends in target coverage.

Main Results:

• sCT images exhibited comparable image quality and anatomy preservation to planning CT.
• Mean CT numbers in ROIs were comparable between sCT and CT.
• The largest median proton range difference was 1.9 mm, with excellent target coverage (V95%≥99.6%) in dose calculations.

Conclusions:

• The DL network successfully generated high-quality sCT images suitable for adaptive proton therapy.
• sCT images demonstrated comparable CT numbers, proton range, and dose characteristics to fan-beam CT.
• The DL network's robustness to intra-patient variations supports its use in adaptive radiotherapy.