Room: Stars at Night Ballroom 1
Purpose: Development and multi-institutional validation of an autoplanning software based on a genetic optimizer.
Methods: A genetic optimizer was implemented in the RayStation treatment planning system (Version 8A) using Python code inside a platform called Genetic Planning Solution (GPS2.1).For each of the ten participating Italian centers we compared for ten of their prostate patients the manually generated, clinically delivered plan (Manual) with the AutoGPS plan. Among centers there was a wide variation in delivered dose, fractionation schedule, and planning protocol. Nonetheless, AutoGPS plans were generated with a standard configuration of the GPS, i.e. no center-specific tuning was applied. Plan evaluations considered parameters as Dmean and the DVH based metrics for Rectum and Bladder, as suggested by QUANTEC. In addition, plan analysis software (PlanIQ, Sun Nuclear Corp., Melbourne, FL) was used, it provided an objective approach for plan quality evaluation using a unique Percentage Planning Quality Metric (PQM%), taking into account both PTV coverage and OARs sparing.
Results: PTV coverage differences were negligible(97.0% and 96.8% for Manual and AutoGPS plans, respectively), while OAR doses were significantly lower in AutoGPS plans (p<0.05) . Bladder D25% and Dmean were 10.1% and 11.6% lower in the AutoGPS plans, and Rectum D15% and Dmean were reduced by 12.1% and 15.6%. Mean doses to the Left and Right Femural Heads were 5.5% and 6.8% lower in AutoGPS plans. In the AutoGPS plans, the mean PQM was improved by 9.8% (p<.0001).
Conclusion: Without any center-specific tuning of the developed genetic optimizer, the quality of the automatically generated prostate plans superseded that of the manually generated plans.
Funding Support, Disclosures, and Conflict of Interest: Alparone A. and Zara S. are employers of Tecnologie Avanzate s.r.l.that is the Company involved in Italy in the free distribution of the software GPS running on the RayStation platform