Multi-criteria optimization algorithms for high dose rate brachytherapy

Authors: Cui, Songye
Advisor: Beaulieu, LucDesprés, Philippe
Abstract: The overall purpose of this thesis is to use the knowledge of radiation physics, computer programming and computing hardware to improve cancer treatments. In particular, designing a treatment plan in radiation therapy can be complex and user-dependent, and this thesis aims to simplify current treatment planning in high dose rate (HDR) prostate brachytherapy. This project was started from a widely used inverse planning algorithm, Inverse Planning Simulated Annealing (IPSA). In order to eventually lead to an ultra-fast and automatic inverse planning algorithm, three multi-criteria optimization (MCO) algorithms were implemented. With MCO algorithms, a desirable plan was selected after computing a set of treatment plans with various trade-offs. In the first study, an MCO algorithm was introduced to explore the Pareto surfaces in HDR brachytherapy. The algorithm was inspired by the MCO feature integrated in the Raystation system (RaySearch Laboratories, Stockholm, Sweden). For each case, 300 treatment plans were serially generated to obtain a uniform approximation of the Pareto surface. Each Pareto optimal plan was computed with IPSA, and each new plan was added to the Pareto surface portion where the distance between its upper boundary and its lower boundary was the largest. In a companion study, or the second study, a knowledge-based MCO (kMCO) algorithm was implemented to shorten the computation time of the MCO algorithm. To achieve this, two strategies were implemented: a prediction of clinical relevant solution space with previous knowledge, and a parallel computation of treatment plans with two six-core CPUs. As a result, a small size (14) plan dataset was created, and one plan was selected as the kMCO plan. The planning efficiency and the dosimetric performance were compared between the physician-approved plans and the kMCO plans for 236 cases. The third and final study of this thesis was conducted in cooperation with Cédric Bélanger. A graphics processing units (GPU) based MCO (gMCO) algorithm was implemented to further speed up the computation. Furthermore, a quasi-Newton optimization engine was implemented to replace simulated annealing in the first and the second study. In this way, one thousand IPSA equivalent treatment plans with various trade-offs were computed in parallel. One plan was selected as the gMCO plan from the calculated plan dataset. The planning time and the dosimetric results were compared between the physician-approved plans and the gMCO plans for 457 cases. A large-scale comparison against the physician-approved plans shows that our latest MCO algorithm (gMCO) can result in an improved treatment planning efficiency (from minutes to 9:4 s) as well as an improved treatment plan dosimetric quality (Radiation Therapy Oncology Group (RTOG) acceptance rate from 92.6% to 99.8%). With three implemented MCO algorithms, this thesis represents a sustained effort to develop an ultra-fast, automatic and robust inverse planning algorithm in HDR brachytherapy.
Document Type: Thèse de doctorat
Issue Date: 2019
Open Access Date: 8 November 2019
Permalink: http://hdl.handle.net/20.500.11794/37180
Grantor: Université Laval
Collection:Thèses et mémoires

Files in this item:
Description SizeFormat 
35615.pdf7.08 MBAdobe PDFThumbnail
View/Open
All documents in CorpusUL are protected by Copyright Act of Canada.