KEY POINTS
- The GPU-based proton and ion dose engine, gPRIDE, was designed to calculate dose within realistic spatially varying magnetic fields. It incorporated fringe-field focusing correction, region-of-interest dose calculation, simplified ray tracing, and parallelization at pencil-beam and voxel levels.
- Validation used Monte Carlo simulations as the reference in a water phantom and four patient anatomies—prostate, liver, lung, and brain—under 0–1.5 T fields with beams either parallel or perpendicular to the main magnetic field.
- Water-phantom gamma passing rates exceeded 99.8% at 2%/2 mm across all evaluated field strengths. Disabling the focusing correction reduced perpendicular-orientation agreement at 1.5 T from 99.87% to 97.61%, showing that fringe-field focusing could not be ignored.
- At 1.5 T, prostate, liver, and brain cases achieved 99.06–99.99% gamma passing rates at 2%/2 mm in both beam orientations. Calculation times ranged from 7.45 to 13.68 seconds for these plans.
- Lung performance was lower because of tissue heterogeneity: passing rates were 92.27% inline and 94.03% perpendicular at 2%/2 mm. They improved to 95.60% and 96.83%, respectively, at 3%/3 mm.
- gPRIDE required 4.44–13.68 seconds per patient plan on an NVIDIA RTX 3060 and less than 1.4 GB of GPU memory. The corresponding Monte Carlo jobs required approximately 3,675–7,952 seconds each, despite being divided across 40 parallel tasks.
- Clinical plans had originally been created without a magnetic field and required manual gantry, source, and isocentre adjustments to compensate for beam deflection. Validation did not include physical measurements or treatment delivery, and accuracy remained limited by the pencil-beam model in lung and potentially for off-centre beams or large treatment fields.
CLINICAL TAKEAWAY
The calculation speed is compatible with online adaptive planning, and agreement was excellent outside highly heterogeneous lung anatomy. MRI-guided proton therapy remains experimental, however, and gPRIDE still requires measurement-based commissioning, integrated magnetic-field-aware optimization, and improved heterogeneous-tissue modelling before clinical deployment.