To quantify expert interobserver variability in longitudinal vestibular schwannoma growth assessment after radiosurgery and evaluate AI segmentation performance relative to human variability.
✦ The floor
Discussion
Signed responses from readers of the wire.
Clinicians should be aware that significant interobserver variability exists when assessing vestibular schwannoma growth post-radiosurgery; AI-assisted segmentation may improve consistency, but adoption into routine practice should await further validation in prospective clinical settings.
Standardising how tumour growth is measured after radiosurgery is critical for avoiding both unnecessary retreatment and missed progression — AI-driven segmentation could reduce the subjectivity currently baked into those decisions.
- 01Experts disagree meaningfully on whether a vestibular schwannoma has grown after radiosurgery, creating inconsistent treatment decisions.
- 02AI segmentation was evaluated as a potential tool to reduce this interobserver variability.
- 03Clinical decision thresholds for 'growth' vary across centres, complicating outcome comparisons.
- 04The study quantifies the magnitude of expert disagreement, providing a benchmark for AI performance.
- 05Findings suggest AI could standardise measurement, though prospective validation is still needed.
There is significant interobserver variability among experts when assessing vestibular schwannoma growth after radiosurgery.
studysupportedAI segmentation performance can be benchmarked against clinician decision thresholds for tumour growth.
studypartially supported- PMID
- 43020972
- DOI
- 10.1007/s10278-026-02516-8.
- Journal
- Journal of Digital Imaging
- Publication type
- research_article
- Evidence level
- 2b
- Population
- Patients with vestibular schwannoma who underwent radiosurgery; expert clinicians assessing tumour images
- Intervention
- AI-based tumour segmentation for vestibular schwannoma post-radiosurgery assessment
- Comparator
- Expert clinician interobserver assessment
Primary outcomes
Degree of interobserver variability in growth assessment among clinical experts; AI segmentation performance relative to clinical decision thresholds