Clinical and MRI factors influencing interobserver delineation variability of brain metastases before stereotactic...

Innovations in systemic therapies have prolonged survival for many cancers, increasing the number of patients living with brain metastases. In stereotactic radiotherapy, steep dose gradients and small target margins mean that delineation differences may affect target coverage and the dose to adjacent normal brain. We aimed to identify lesion- and MRI-related factors associated with interobserver variability in brain metastasis gross tumor volume delineation.

We retrospectively included 149 patients (354 metastases) treated with stereotactic radiotherapy at the University Hospital of Brest (2014-2022). Four radiology residents delineated each metastasis on pretreatment three-dimensional T1-weighted gadolinium-enhanced MRI after completing a structured training set of 40 brain metastases selected outside the study cohort (excluded from all analyses). Lesion characteristics (edema, hemorrhage, necrosis/cystic change, proximity to organs at risk, histology, and volume) and MRI acquisition parameters were collected. Interobserver agreement was quantified using the Dice similarity coefficient and the mean distance to agreement (MDA).

Medians and interquartile ranges (IQRs) for Dice and MDAs across metastases (pairwise values averaged per metastasis) were 0.85 (IQR 0.80-0.89) and 0.34 mm (IQR 0.29-0.44), respectively. Lesions with perilesional edema and a larger volume showed higher overlap agreement (Dice); however, MDA increased with lesion volume, consistent with greater absolute boundary deviations in larger lesions. Several distance-to-OAR variables showed significant univariable associations with agreement metrics; however, these findings were inconsistent across Dice and MDA and were not retained in multivariable analysis. Among MRI parameters, 3 T field strength was associated with a slightly higher agreement; no other parameters showed consistent associations in this exploratory analysis.

Interobserver variability in brain metastasis delineation varies with lesion characteristics and selected MRI acquisition parameters. Standardized contouring guidance and MRI acquisition may reduce variability, supporting quality assurance and peer-review processes for stereotactic radiotherapy.