One RapidPlan model generated ocular stereotactic plans across three fractionation schedules

KEY POINTS

• The RapidPlan model was trained using 86 synthetic ocular cases and tested on 24 independent datasets, including 20 hybrid cases with magnetic-resonance-imaging-derived tumour contours and four synthetic cases with planning target volume overlap of the optic nerve.
• The same prescription-agnostic model generated plans for 25 Gy in one fraction, 42 Gy in three fractions and 50 Gy in five fractions using HyperArc with four partial arcs, 6 MV flattening-filter-free beams and Acuros XB.
• Each plan used one automated optimization without manual adjustment. Median Radiation Therapy Oncology Group conformity indices were 1.05, 1.06 and 1.05, while the Paddick conformity index was 0.86 for all three schedules.
• Median gradient indices were 3.41, 3.34 and 3.37, and median planning target volume D99% was 95.91%, 95.37% and 95.30% of prescription for one-, three- and five-fraction plans. Statistically significant differences were generally below 1% and considered clinically negligible.
• Median ipsilateral optic-nerve maximum doses were 4.77, 7.94 and 9.67 Gy across the three regimens. Optic nerve, chiasm, brain, brainstem and lacrimal-gland constraints were usually achievable, while lens and skin limits were more frequently exceeded.
• Median optimization and final dose-calculation time was 13.98 minutes. Median beam-on time decreased from 6.97 minutes for one fraction to 3.74 minutes for three fractions and 2.67 minutes for five.
• Portal-dosimetry pass rates exceeded 99% using 2%/2 mm criteria for all schedules. High-definition multileaf collimation improved conformity and reduced maximum dose, while 10 MV beams shortened delivery but generally increased dose gradient and selected ocular-organ doses.

CLINICAL TAKEAWAY

SOURCE

Medical Dosimetry