Author: Xianjin Dai, PhD, Wu Liu, Eric Nguyen, Lei Xing, Lewei Zhao 👨🔬
Affiliation: Department of Radiation Oncology, University of California, Los Angeles, Department of Radiation Oncology, Stanford University 🌍
Purpose: Recent developments in hybrid quantum solvers, which combine quantum and classical processing, enable greater flexibility to tackle problems outside the limited scope of pure quantum computing. Here, we assess the feasibility of using a hybrid quantum annealer to optimize eye plaque brachytherapy (EPB) treatment planning.
Methods: We first formulate EPB as an integer programming problem, where the seed activity in a given slot can be encoded as integer variables with discretized levels. We used D-Wave’s Constrained Quadratic Model (CQM) hybrid quantum solver to optimize seed activities to fulfill two overarching objectives: (1) ensure that voxels within the tumor volume and those comprising the 2D retinal surface margin receive at least the prescription dose, and (2) minimize the dose to critical organs-at-risk, including the lens, optic nerve, fovea, retina, and muscle. We used I-125 eye plaque treatment plan data from eleven patients for evaluation, extracted from the treatment planning system, Plaque Simulator v6.9.3. We compare the clinical quality of differential loading plans generated by CQM hybrid solver and a traditional simulated annealing (SA) solver with the clinical uniform loading plan in terms tumor coverage and OAR sparing.
Results: CQM, SA and clinic plans all achieve nearly 100% PTV V100 for 10 of 11 patients. Among the 10 comparable patients, the CQM demonstrates significant advantages, achieving approximately 23% lower maximum doses to the fovea, optic nerve, and lens compared to the clinical plan, while the SA delivers 30-45% higher maximum doses to these structures compared to the clinical plan.
Conclusion: Our study demonstrates that the hybrid quantum solver is able to generate high quality differential seed loading plans in EPB using combinatorial optimization, whereas classical simulated annealing was challenged by this task.