KEY POINTS
- This single-institution retrospective study included 55 patients with localized prostate cancer: 30 cases for model construction and closed-loop validation, and 25 independent cases for open-loop and cross-prescription validation.
- A RapidPlan model trained for conventional fractionation at 78 Gy in 39 fractions was reused without retraining for moderate hypofractionation at 60 Gy in 20 fractions under a different, CHHiP-based dose-constraint framework.
- In cross-prescription validation, model-generated and manual plans had comparable overall plan quality scores: 89.5 ± 34.5 versus 91.2 ± 41.8 (p=0.287).
- Manual plans achieved slightly better target homogeneity: planning target volume D98% was 59.3 versus 59.0 Gy, while D2% was 62.7 versus 63.2 Gy with model-generated plans (both p<0.001). These differences remained within institutional acceptance criteria.
- Model-generated plans reduced rectal V57 Gy (4.4% versus 5.2%, p=0.007) and several bladder dose-volume endpoints, while bladder V60 Gy and most rectal endpoints were not significantly different.
CLINICAL TAKEAWAY
A prostate knowledge-based planning model may be reusable across conventional and moderately hypofractionated prescriptions when anatomy, beam geometry, and optimization priorities remain broadly similar. The approach could reduce repeated model-building work, but the evidence is limited to one institution, one planning platform, and dosimetric endpoints; local cross-prescription validation remains essential.