Author: Arezoo Modiri, Robabeh Rahimi, Akira SaiToh, Amit Sawant 👨🔬
Affiliation: Maryland University Baltimore, University of Maryland, University of Maryland in Baltimore, Department of Computer and Information Sciences, Sojo University 🌍
Purpose: It has been a longstanding challenge to optimize the daily schedule of radiation treatment rooms toward minimum patient wait times, efficient use of clinical staff and reduced running cost of gantries. While genetic algorithms have been one of the most successful workflow optimization algorithms, their complexity becomes a computational burden when solved in classical way. We developed the first of its kind tailored variant of a quantum-inspired genetic algorithm for radiation oncology workflow optimization.
Methods: In our design, each quantum-inspired chromosome represented an entire daily schedule consisting of tracks for a single gantry. Each cell possessed two quantum states: (i) a superposition of patient IDs and (ii) a superposition of gantry statuses. The initial generation consisted of 12 randomly generated chromosomes. Simulations were performed on a PC with 16 CPU cores and 32 GB memory. The fitness function was a summation of benefit scores (e.g., completed therapy steps) minus penalty scores (e.g., schedule conflicts). Crossover, mutation, and repair steps were implemented similarly to classical counterparts, with a non-demolition assumption.
Results: For a case study of 3 gantries and 12 patients, the quantum-inspired algorithm achieved the optimization goal in comparable time but with 77% fewer chromosomes than its classical counterpart - 35 versus 150 chromosomes. The surviving, crossover, mutation, and repair ratios were 0.84, 0.27, 0.3 and 0.85, respectively. We showed that the repair ratio has an optimum case-dependent value in the quantum-inspired algorithm in contrast to its continuous positive correlation with fitness growth in the classical algorithm.
Conclusion: We developed a prototype variant of the quantum-inspired genetic algorithm tailored for radiation oncology workflow optimization. The design is extendable to include more variables such as patient room selection. The low quantum memory cost of our algorithm is of practical importance for future (beyond current scope) implementation on quantum computing hardware.