Development and Evaluation of an Automated Quantum-Inspired Optimization Framework for Partial-Breast IMRT
Abstract
Purpose
To develop a streamlined and automated quantum-inspired optimization framework using mixed-integer quadratic programming (MIQP).
Methods
A quantum-inspired treatment planning framework based on a weighted quadratic cost function with integer beamlet intensities was developed in MATLAB using open-source MatRad functionalities. Custom MATLAB/MatRad scripts automated open pencil-beam fluence from clinically feasible beam angles. The MIQP formulation, encapsulated as a software library (C++), enabled automatic MIQP problem generation, parameter specification (python GUI), optimization (Gurobi), and import of optimized solutions into MatRad for plan evaluation and display. The workflow was applied retrospectively to seven partial-breast patients (30 Gy in 5 fractions). IMRT (step-and-shoot) plans from a clinical planning system and the quantum-inspired solver, both optimized using identical beam angles, were compared using dose–volume histogram (DVH) metrics; target coverage (PTV D95%), lung (V10Gy, V5Gy) and heart dose metrics (V3Gy).
Results
By encapsulating all components within a unified MATLAB-driven workflow, we eliminated manual, multi-environment processing and enabled automated plan optimization and evaluation. Quantum-inspired plans were generated reproducibly across all cases with minimal parameter adjustment with a single case runtime of under 10 seconds. Preliminary DVH comparisons demonstrated comparable target coverage (PTV D95% = 30 Gy for both) and organ-at-risk (OAR) sparing (Heart V3Gy = 1.2 ± 1.7% for quantum-inspired plans vs 1.7 ± 2.2% clinical plans. Ipsilateral lung V10Gy = 1.5 ± 1.0 % vs 2.0 ± 2.2%, contralateral lung V5Gy = 0.3 ± 0.5% vs 0.9 ± 2.3%, respectively). For quantum-inspired plans, observed inter-patient variability in dose metrics reflected differences in target–organ geometry.
Conclusion
This streamlined and scalable software framework using MIQP formulation enables efficient, reproducible evaluation of quantum-inspired optimization methods and provides a foundation for ongoing studies aimed at improving plan quality at higher complexities. Future studies will further integrate software components, expand the patient dataset, explore beam angle optimization, and support IMPT.