Geometric and dosimetric effects of an automated clustering and plan optimization pipeline for multiple brain metastasis radiosurgery

Abstract

Background

Treating multiple brain metastases simultaneously with single-isocenter multi-target (SIMT) stereotactic radiosurgery (SRS) can streamline treatment planning and delivery. However, it can also result in targets at large distances from the isocenter, increasing sensitivity to rotational setup errors and requiring wider collimator leaves. Grouping the targets and treating each group with its own plan retains the practical benefits of SIMT SRS while mitigating the concerns of targets at large distances off-isocenter.

Purpose

This work presents a prototype of a newly integrated treatment planning pipeline that automatically clusters targets into groups and subsequently optimizes a plan for each group simultaneously. Treatment scenarios featuring varying numbers of isocenters were compared.

Methods

Twenty-nine treatment courses for patients receiving SRS for 3–15 brain metastases were retrospectively analyzed. Multiple treatment scenarios were prepared for each patient-course by grouping targets into 1–5 isocenters using k-means clustering. PTV margins were added using two methods, one with a magnitude that varied with distance to the isocenter, and a second with a constant 1-mm magnitude. A non-clinical, command-line version of the Brainlab Elements Multiple Brain Mets SRS software was used to optimize plans within each scenario simultaneously. Cumulative doses were analyzed in terms of geometric values (target distances to isocenter, total PTV volume), dosimetric values (volume of normal brain (brain-GTVs) receiving 5, 8, 10, 12, and 18 Gy (V_5Gy, V_8Gy, V_10Gy, V_12Gy, and V_18Gy)), and logistical values (number of Monitor Units (MUs)). A representative sample of 69% of plans with variable margins was reviewed by clinical SRS experts.

Results

As the number of isocenters increased, distances from the targets to the isocenters decreased, and the proportion of targets within a given distance increased. For example, 36%, 77%, 94%, 99%, and 100% of targets were within 4 cm of the isocenter for 1–5 isocenters, respectively. For the variable margin method, the total PTV volume relative to the single-isocenter scenario had a median decrease of 17.5, 24.3, 31.6, and 37.8% for 2–5 isocenters, respectively. Similarly, the relative change in the normal brain dose metrics decreased with increasing number of isocenters (e.g. V_12Gy decreased 20.1, 26.2, 36.2, and 40.9%, respectively). The relative change in dose metrics at each level was found to have a linear relationship with that of the total PTV volume (e.g. V_12Gy, R² = 0.81). For the constant margin method, total PTV volume remained constant, but normal brain dose metrics still decreased due to the grouping and simultaneous plan optimization alone (e.g. V_12Gy decreased 6.4, 10.5, 15.1, and 17.1%). Regarding logistical values, the number of MUs increased with each additional isocenter for both margin methods, but at a rate slower than proportional to the number of isocenters. Of the evaluated plans, nearly all (97.7%) were clinically acceptable without modification.

Conclusion

The SIMT SRS treatment planning pipeline presented here features automatic grouping of targets followed by automatic dose optimization of up to five isocenters. By quantifying the geometric, dosimetric, and logistical effects of adding isocenters, this pipeline can empower clinicians to make resource-conscious treatment decisions optimal for each patient.