KEY POINTS
- This single-institution implementation study included a paired validation cohort of 10 head and neck cancer cases and a clinical evaluation of 1,000 consecutive treatment plans: 500 manual plans created before implementation and 500 automated plans created during three years of routine use.
- In validation testing, automated plans maintained planning target volume coverage and similar global maximum dose (107.4% versus 107.3%) while reducing maximum doses to the brainstem by 5.1 Gy and spinal cord by 2.9 Gy, both P < .03.
- Across 50 blinded physician ratings, 94% of automated plans and 86% of manual plans were considered clinically acceptable. Automated plans were preferred by the majority for 7 of 10 cases, one was considered equivalent, and manual plans were preferred for two.
- In the clinical cohorts, automated planning maintained target coverage and similar global maximum dose while reducing brainstem and spinal cord maximum doses by 3.6 Gy and 2.1 Gy, respectively, both P < .001. Significant reductions were also observed for the salivary glands, oral cavity, cochleae, larynx, mandible, and lips.
- Plan preparation took under 5 minutes. Full optimization required approximately 50–60 minutes on standard workstations and 15–20 minutes with graphics processing unit acceleration.
CLINICAL TAKEAWAY
Automated planning can support reproducible head and neck treatment planning with preserved target coverage, improved organ-at-risk dosimetry, and limited manual intervention. However, the clinical comparison used unmatched pre- and post-implementation cohorts, permitted minor planner refinements, and collected no toxicity outcomes, so the findings support technical implementation rather than proving clinical benefit.