2013 - Assessing the Efficacy of an AI-Assisted TPS for Patients with Oligometastatic Cancer
Presenter(s)
N. Sarigul1, R. Chicfeh1, B. E. Ispir2, G. A. Kavak3, E. Pearson1, and B. Aydogan4; 1Department of Radiation and Cellular Oncology, The University of Chicago Medicine, Chicago, IL, 2Acibadem Healthcare Group, Department of Radiation Oncology, Istanbul, Turkey, 3Gaziantep University, Department of Radiation Oncology, Gaziantep, Turkey, 4Department of Radiation and Cellular Oncology, The University of Chicago, Chicago, IL
Purpose/Objective(s): To assess the performance of a commercial AI-assisted Treatment Planning System (AI-aTPS) in planning oligometastatic cancer treatments for patients previously treated with multi-isocenter plans targeting up to 5 tumors
Materials/Methods: Study cohort includes 200 patients, previously treated under IRB approval, using C-arm linacs with doses ranging from 30 to 45 Gy over 3 fractions. Patients were categorized by the number of isocenters (1-2) and targets (up to 5). Tumor locations included the lung, liver, pancreas, iliac region, ribs, and abdomen. AI-aTPS planning utilized automated contouring, organ specific planning criteria, and site-specific planning templates. The primary planning objectives were V95%=95% and V70%=99.5% for PTV while ensuring organ at risk (OAR) doses remained comparable to or less than those in the clinically approved plan. Plans were generated using AI-aTPS with a 12-field intensity-modulated radiation therapy (IMRT) technique. All plans were created by an inexperienced researcher and subsequently evaluated by experienced physicists. Planning times were recorded, and dose coverage was compared to the clinical plans used for patient treatment.
Results: AI-aTPS reduced planning time was 34 ± 5.6 minutes, providing substantial improvement over current practice which may take up to several days for complex cases. AI-aTPS plans met PTV dose coverage objectives in 56% of cases, compared to 50% with clinical plans. For V70%, AI-aTPS and clinical plans produced effectively identical results (100% coverage) in 48% of cases, while AI-aTPS achieved superior dose coverage in 35% of cases, and clinical plans outperformed AI-aTPS in 16%. For V95%, AI-aTPS provided superior coverage in 52% of cases, with clinical plans outperforming in 47% and the remaining 1% of cases showed identical results between the two systems. For approximately 62% of patients, an inexperienced planner was able to quickly generate clinically acceptable plans. However, the remaining cases required additional refinements from an experienced planner. Expertise was particularly needed for tumors in challenging locations, such as liver segments SII and SIII (adjacent to the heart and stomach), pancreatic tumors within 0.5 cm axially of the duodenum and/or stomach, and cases where meeting the D0.03cc constraint for certain OARs proved difficult.
Conclusion: AI-aTPS significantly reduced planning times and reliance on planner experience while improving PTV dose coverage through auto-contouring, planning templates, and tumor-location classification. These findings indicate that even highly complex cases with multiple metastatic tumors can be planned in under an hour.