Purpose: The purpose of this study is to assess the efficacy of Artificial Intelligence (AI)-generated treatment plans using Varian Ethos system.
Methods: Six large field pelvic/abdomen sites, treated with Varian Ethos v.2.0, were chosen for the present study. Ethos created 2 VMAT and 3 IMRT AI-driven plans for these cases, based on priority goals in the RT intent and the user chose the most optimal plan. For all cases, IMRT plans were dosimetrically superior to VMAT plans and were selected for treatment. Typically, these cases would have been treated using VMAT in Eclipse. Ethos VMAT plans were exported to Eclipse v.16.1 and optimized by an experienced dosimetrist, using the same treatment geometry. All plans were calculated for dose using AcurosXB and were normalized, such that, 95% of the PTV received 100% of the prescription dose.
Results: Global maximum doses for Ethos vs. Eclipse plans were 114.6% vs. 114.1%, respectively. PTV indices (D98/D50/D2) for Ethos vs. Eclipse VMAT plans were 98.3% vs. 98.4%, 104.2% vs. 105.0%, and 108.3% vs. 109.4%, respectively. Conformity and Homogeneity indices for Ethos and Eclipse plans were 1.06 vs. 1.08, and 0.10 vs. 0.10, respectively. On average, total plan monitor units were 985.7 and 946.2 for Ethos and Eclipse plans, respectively. Based on RTOG criterion, specific to the organ in question, both sets of plans were deemed to be clinically acceptable.
Conclusion: Varian AI-driven Ethos system creates highly-modulated IMRT plans that are comparable or superior to conventional IMRT plans generated by dosimetrists. VMAT plans from Ethos are not as good as their IMRT plans but comparable to typical VMAT plans created in Eclipse. The inferiority of Ethos VMAT plans seems to be systemic. It appears AI VMAT plans are not running sufficient iterations due to time constraints and, therefore, plans are being trapped in local minima.