Comparing variable neighbourhood search algorithms for the direct aperture optimisation in radiotherapy

Area of Science:

• Medical Physics
• Radiation Oncology
• Computational Optimization

Background:

• Intensity Modulated Radiation Therapy (IMRT) aims to maximize tumor cell eradication while minimizing damage to surrounding organs at risk.
• Conventional IMRT planning involves sequential optimization and sequencing, often leading to suboptimal plans due to unaddressed physical and delivery constraints.
• Direct Aperture Optimization (DAO) addresses these limitations by simultaneously optimizing aperture configurations and radiation intensities, considering delivery constraints.

Purpose of the Study:

• To propose and compare two novel Variable Neighbourhood Search (VNS) based algorithms: Variable Neighbourhood Descent (VND) and reduced Variable Neighbourhood Search (rVNS).
• To evaluate the effectiveness of VND and rVNS in generating clinically appropriate IMRT treatment plans.
• To assess the performance of the proposed algorithms against established DAO methods, specifically matRad's tool.

Main Methods:

• Development and implementation of Variable Neighbourhood Descent (VND), a deterministic VNS variant exploring diverse neighborhood structures.
• Development and implementation of reduced Variable Neighbourhood Search (rVNS), integrating predefined neighborhood moves without a transition rule.
• Application of both algorithms to prostate cancer cases and comparison of results using dosimetric indicators for target coverage and organ-at-risk sparing.

Main Results:

• Both proposed algorithms, VND and rVNS, achieved highly competitive results in prostate cancer treatment planning.
• The rVNS algorithm demonstrated significant improvements, requiring 62.75% fewer apertures and achieving a 63.93% reduction in beam-on time compared to the sequential approach.
• Clinical quality assessments using dosimetric indicators showed comparable or superior performance in target coverage and organ-at-risk sparing against matRad's DAO tool.

Conclusions:

• The proposed VND and rVNS algorithms offer efficient and effective solutions for Direct Aperture Optimization in IMRT.
• rVNS significantly reduces treatment delivery time and complexity by minimizing apertures and beam-on time.
• These advanced optimization techniques hold promise for improving the clinical applicability and efficiency of IMRT for cancer patients.