KEY POINTS
- This retrospective model-development and validation study used imaging from 148 head-and-neck cancer patients: 90 for training, 38 for validation, and 20 for independent testing.
- A UNet3+ generative adversarial network generated synthetic computed tomography from cone-beam computed tomography, including anatomy outside the cone-beam field of view.
- In the independent test cohort, mean absolute error was 49.9 ± 11.0 Hounsfield units globally, 40.2 ± 8.2 inside the cone-beam field of view, and 61.4 ± 19.0 outside it.
- Dose recalculation in a commercial treatment planning system showed mean planning target volume metric differences below 1.6%, with maximum observed difference 5.1%.
- Global gamma pass rates were 99.0% at 3%/2 millimetres and 98.7% at 2%/2 millimetres, using a 10% dose threshold. Blinded clinical realism assessment had a median score of 4 of 5.
CLINICAL TAKEAWAY
Synthetic computed tomography from cone-beam computed tomography may enable daily dose recalculation for head-and-neck adaptive radiotherapy on conventional linear accelerators, including regions beyond the cone-beam field of view. The dosimetric performance is technically encouraging, but the evidence remains limited by the small retrospective test cohort, single-vendor imaging data, and reliance on deformably registered computed tomography as the reference standard.