Robust Single-Isocenter Selection for Multi-Target SRS Using Weighted Expected Dice Similarity Coefficient
Abstract
Purpose
This work demonstrates the use of a weighted expected Dice similarity coefficient (wDSC) for optimizing isocenter placement in single-isocenter multi-target (SIMT) LINAC-based stereotactic radiosurgery (SRS).
Methods
Fifteen cases of randomly generated and clinical GTV-based spherical targets were evaluated (3-11 targets per case). Geometric-based isocenters using center of surface area (CSA) and geometric mean (GeoM) were calculated. An additional minimum-volume (MinVol) isocenter was computed by identifying the coordinate that minimizes the distance-based margin volume required by all targets under rotational uncertainty. For each target, an expected DSC was computed using Monte Carlo sampling to characterize misalignment under probable setup uncertainty. Expected DSC values were combined using a weighted average that emphasizes smaller targets, and the isocenter was optimized by maximizing the resulting wDSC. Target volumes were rotated by ±1° about each candidate isocenter along the principal axes to model SRS rotational tolerance. The worst-case DSC for each target was recorded, and the minimum value across targets was used to compare isocenter robustness.
Results
Higher wDSC values correspond to more uniform overlap across all targets after perturbations. The wDSC isocenter achieved the highest minimum worst-case DSC in ten cases. In the remaining cases, although the GeoM isocenter attained a higher minimum value, the wDSC isocenter achieved a higher mean worst-case DSC across targets indicating more consistent coverage across all targets. Per-target analysis further showed geometric isocenters preserved high worst-case DSC for large targets (>0.85) but allowed small-target DSC to reach 0.1, indicating elevated risk of geometric miss for small targets despite acceptable coverage of larger targets. In contrast, wDSC isocenter increased the small-target DSC lower bound to 0.45 without compromising large-target DSC.
Conclusion
wDSC-based optimization provides a robust framework for single-isocenter placement in SIMT SRS by improving worst case overlap for small targets without compromising coverage of larger targets.