Dose-Guided Automatic IMRT Planning: A Feasibility Study
Y Sheng*, T Li, L Yuan, F Yin, Q Wu, Duke University Medical Center, Durham, NCTH-C-137-11 Thursday 10:30AM - 12:30PM Room: 137
Purpose: To develop and evaluate an automatic planning technique for prostate cancer utilizing prior expert plan's dose as guidance.
Methods: A plan atlas was created using 10 expert prostate IMRT plans covering a variety of typical anatomical configurations. The target includes both prostate and seminal vesicle.
Additional 9 patients (query cases) were used to test the automatic planning, which started by matching the query case to the atlas based on the overall PTV-OAR anatomical configuration. The anatomy of the matched atlas case is then linked to the query case via deformable registration in MIM Maestro™, which provides fine local-regional matching. Following the same anatomical deformation, the expert dose in atlas is warped onto the query case, creating the goal dose conforming the query case's target. DVH objectives were sampled from the goal dose to guide automatic IMRT treatment planning in Eclipse™ for the query case. Dosimetry comparison between dose-guided automatic plans (DAPs) and clinical plans for query cases are reported.
Results: Generating goal dose is highly efficient with customized workflows in MIM™. The deformation registration provides a realistic goal dose for the query cases in terms of dose falloff at the PTV-OAR junctions and PTV conformity. The automatic planning takes ~2.5 min (~70 iterations) without human intervention. Compared to clinical plans, DAPs improved the PTV conformity index from 1.031±0.044 for expert plans to 1.005±0.019 (p=0.04). Other dosimetric parameters are similar between DAP and clinical plans (p>0.1): homogeneity indices are 0.073±0.010 and 0.079±0.015; Bladder-gEUDs are 3917±352 cGy and 3934±354 cGy; Rectum-gEUDs are 3923±256 cGy and 3932±323 cGy, respectively.
Conclusion: Dose-guided automatic treatment planning is feasible and efficient. Atlas-based patient-specific DVH objectives can effectively guide the optimizer to achieve similar or better plan quality compared to clinical plans.
(The authors thank Adam Neff from MIM Software Inc. for technical supports.)
Funding Support, Disclosures, and Conflict of Interest: The study is partially supported by NIH grant and a Varian master research grant.