Program Information
Integration of Reduced Order Constrained Optimization for IMRT Planning Into the Eclipse Treatment Planning System
S Bakr1,2*, H Nourzadeh1 , S Tuomaala3 , L Happersett2, J Yang2 , R Radke1 , A Jackson2, (1) Rensselaer Polytechnic Institute, Troy, NY, (2) Memorial Sloan-Kettering Cancer Ctr, New York, NY, (3) Varian Medical Systems, New York, NY.
Presentations
PO-BPC-Exhibit Hall-6 (Saturday, March 7, 2015) Room: Exhibit Hall
Purpose: We describe the integration of the reduced order constrained optimization (ROCO) algorithm, a previously developed algorithm for accelerating IMRT optimization, into a commercial treatment planning system (Eclipse v13.5). While previous implementations of ROCO were successfully applied to IMRT planning in different sites, these were demonstrated using an in house, not commercially available treatment planning system.
Methods: ROCO suggests a systematic way to characterize the intensity space in a compact form. This enables forming a computationally tractable constrained optimization, which directly enforces hard clinical constraints on Organs at Risk (OARs) and Planning Target Volumes (PTVs). ROCO starts with sampling of the intensity space by solving a set of unconstrained optimization problems. Principal component analysis (PCA) is used to attain a reduced order representation of fluence profiles, and the dose corresponding to each principal component is computed. Then, a constrained optimization problem is formed and solved over the basis coefficients spanning the reduced size space. The new implementation supports a variety of clinically relevant constraints. It also exploits a previously used iterative scheme to automatically accommodate Dose Volume constraints. The resulting fluence map is imported back into Eclipse for deliverability assessment, final dose calculation, and plan evaluation. ROCO is developed as a stand alone application in .NET framework 4.5. It uses Eclipse scripting API V13.5 to interact with optimization and dose calculation modules. The new pipeline harnesses asynchronous and multithreaded programming techniques as well as highly optimized libraries and solvers.
Results: The ROCO Eclipse implementation is tested on a non nodal prostate case resulting in a clinically acceptable plan according to our institute’s latest clinical protocol.
Conclusion: The new integration adds distinctive features that make it more practical in a clinical setup, where the planners can potentially save hours of tedious trial and error to devise clinically relevant IMRT plans.
Funding Support, Disclosures, and Conflict of Interest: We acknowledge the collaboration of Varian Medical Systems in carrying out this work. This work was supported in part by Grant Number R01CA148876-02 from the 401 National Cancer Institute (NCI).
Contact Email: