Hybrid geometric and evolutionary optimization framework for radiosurgery planning: isocenter selection and angle optimization for multiple brain metastases.

Abstract

This study aims to develop and evaluate the performance of Collimator Angles Optimization Software version 2 (CAOS-v2) for stereotactic radiosurgery (SRS) planning in patients with multiple brain metastases. Treatment planning was performed in Eclipse v13.6 with RapidArc on the TrueBeam STx. The primary objective is to reduce the volume of normal brain tissue receiving 12 Gy (V12Gy) while maintaining overall plan quality. CAOS-v2 determines the isocenter location using the Smoothed Weiszfeld Algorithm to compute the geometric median of the target centers, in contrast to traditional centroid-based placement. The software subsequently employs a genetic algorithm to optimize couch-collimator angle pairs, reducing the impact of the island blocking problem. Twenty-six clinical cases were retrospectively replanned using CAOS-v2 with a single-fraction prescription dose of 18 Gy applied to all treatment plans. Plans were evaluated using Conformity Index (CI), Gradient Index (GI), Gradient Measure (GM), organs at risk (OARs), and the V12Gy of healthy brain tissue. Application of CAOS-v2 resulted in a reduction of V12Gy by 17.6-38.1% compared to the conventional planning approach. The CI remained stable, indicating consistent dose uniformity across treatment plans. No increase in dose to OARs was observed. Additionally, both the GI and GM significantly decreased, indicating improved dose fall-off characteristics attributable to CAOS-v2. Integration of CAOS-v2 into the planning workflow significantly reduces normal brain dose in multi-metastatic SRS without compromising conformity or gradient performance, supporting its clinical applicability for automated treatment parameter optimization.