Comparing methods to improve cone-beam computed tomography for dose calculations in adaptive proton therapy

Proton therapy requires dose monitoring, often performed based on repeated computed tomography (reCT) scans. However, reCT scans may not accurately reflect the internal anatomy and patient positioning during treatment. In-room cone-beam CT (CBCT) offers a potential alternative, but its low image quality limits proton dose calculation accuracy. This study therefore evaluated different methods for quality-improvement of CBCTs (synthetic CTs; sCTs) for use in adaptive proton therapy of head-and-neck cancer patients.

Thirty-five CBCTs from twenty-four head-and-neck cancer patients were used to assess four sCT generation methods: an intensity-correction method, two deformable image registration methods, and a deep learning-based method. The sCTs were evaluated against same-day reCTs for CT number accuracy, proton range accuracy through single-spot plans, and dose recalculation accuracy of clinical plans via dose-volume-histogram (DVH) parameters.

All four methods generated sCTs with improved image quality while preserving the anatomy relative to the CBCT. The differences in absolute median proton range between sCT methods were small and generally less than the difference between sCT and reCT, which had median differences of 1.0-1.1 mm. Similarly, differences in DVH parameters were generally small between the sCT methods. While outliers were identified for all four methods, these outliers were often consistent for all sCT methods and could be attributed to anatomical and/or positional discrepancies between the CBCT and reCT.

All four sCT methods enabled accurate proton dose calculation and preserved the anatomy, making them of value for adaptive proton therapy.