Multi-Criteria Optimization for Real-Time Planning of Lung Cancer Radiotherapy
W.T. Watkins1*, W.Y. Song2, J.R. Merrick1, E Weiss1, G.D. Hugo1, J.V. Siebers1, (1) Virginia Commonwealth University, Richmond, VA (2) University of California, San Diego, La Jolla, CASU-E-T-648 Sunday 3:00PM - 6:00PM Room: Exhibit Hall
Purpose: Multi-criteria optimization (MCO) is implemented for planning lung cancer radiotherapy treatments to clarify patient-specific tradeoffs and allow real-time plan decision making.
Methods: For four locally advanced lung cancer patients (P1-P4), a basis set of MCO plans are constructed and compared to plans determined from fixed-objective (FO) optimization for organs at risk (OARs). All optimized plans include constraints on target-D95>70 Gy and spinal cord Dmax<45 Gy. Five MCO basis plans are designed per patient through weight variation of four non-zero dose-volume objectives (DVOs) for ipsilateral lung (iLung), contralateral lung (cLung), heart, and esophagus. The five basis plans are optimized according to: (1) simultaneous minimization of four OAR-DVOs and (2-5) weight variation for one OAR-DVO.
Results: Patient-specific tradeoffs between OAR objectives are revealed with MCO which are not evident in FO-optimization. For P1, MCO basis plans vary iLung-V20 from 46% to 65% and show that V20<46% is not achievable; the FO iLung-V20 is 54%. For P2,the FO-plan trades off a 1% reduction in iLung-V20 for a 29% increase in esophagus-V20. An interpolated MCO plan, in this case, takes advantage of this tradeoff to reduce esophagus V20 by from 41% to 8%. P3 shows increasing heart-V20 by 35% (from 20% to 55%) results in a 7% (34%-27%) reduction in iLung-V20. With P4, MCO reveals a tradeoff between the two lungs; varying iLung-V20 from 23% to 30% corresponds to cLung-V20 varying from 23% to 17%. The FO plan treats cLung up to the V20 objective (to 29%) without penalty. MCO shows the ability to reduce OAR dose-volumes, but often led to increased PTV hotspots.
Conclusion: Analysis of MCO plans clarifies conflicting objectives and exposes inherent limitations due to patient geometry. Real-time planning is possible with a small set of MCO plans, and achieves plans which are superior to FO-optimization.
Funding Support, Disclosures, and Conflict of Interest: Supported by NIH P01-CA-116602 and Philips Medical Systems