Method for simultaneous selection of treatment isocenters and margins for polymetastatic extracranial stereotactic ablative radiotherapy
Jordan M. Slagowski, Gemma A. Davies, Adam Bayliss, Laura C. Bennett, Mustafa M. Basree, Michael F. Bassetti, Carri K. Glide‐HurstAbstract
Background
Stereotactic ablative radiotherapy (SABR) represents a new era of treatment for polymetastatic extracranial disease but introduces unique challenges in isocenter selection and margin optimization to balance treatment efficiency and normal tissue sparing.
Purpose
Develop an end‐to‐end method to simultaneously cluster M tumor targets and optimize isocenter number (N) and position to minimize the additional margins required to maintain target coverage as tumor‐to‐isocenter distance increases.
Methods
K‐means clustering identified candidate isocenters ranging from one to one per target. Margins required for 95% coverage probability were determined based on tumor‐to‐isocenter distance, accounting for translational (5 mm) and rotational (1°, 2°, 3°) uncertainties modeled as three‐dimensional normal distributions. K‐means total margin volume was compared versus isotropic 5‐ and 10‐mm margins and benchmarked against a derivative‐free hybrid optimization method. The method was evaluated on 20 clinical lung cases with 2–21 targets per patient, target‐to‐target distances of 3.8–32.4 cm, and volumes of 0.07–41.05 cm 3 .
Results
The total margin volume determined by k‐means showed no significant differences ( p = 0.94) relative to the hybrid optimization method, with median differences of 0.03% (0.01 cm 3 ), 0.14% (0.06 cm 3 ), and 0.19% (0.15 cm 3 ) for 1°, 2°, and 3° rotational uncertainties, respectively. Significant ( p < 0.0167) differences in total margin were observed versus fixed margins of 5 or 10 mm. Increasing the number of isocenters for a given patient reduced the median total margin volume by 31.0% for N = 2 versus N = 1 and 15.7% for N = 3 versus N = 2, but diminishing returns were observed as additional isocenters were added ( N = 4 to N = 6: −6.6%, −5.4%, and −3.7%, respectively). Median run time was 0.19 s for k‐means versus 306.9 s for the reference method.
Conclusion
K‐means clustering offers an efficient method for selecting the number and locations of isocenters to reduce the total margin volume in multi‐target extracranial SABR.