A Web-Accessible Comprehensive (Weave) Platform for Vestibular Schwannoma Auto-Segmentation and Treatment Follow-up
Abstract
Purpose
To develop a WEb-Accessible comprehensiVE (WEAVE) platform, that combines AI-driven segmentation with a user-friendly interface to automatically segment Vestibular Schwannomas (VS) and track longitudinal volumes for disease assessment, planning, and monitoring.
Methods
The WEAVE platform consists of a front-end web client and a back-end server. The front-end web client is built on Rust and WebAssemly, featuring fast image rendering on web browsers. The back-end server is established on a SQLite database and an auto-segmentation engine. Using nnU-Net as the backbone for auto-segmentation, we trained and validated three models to accommodate various combinations of image modalities available in clinic, including CT-T1c-T2, T1c-T2 and T1c only. Using manual clinical delineation as ground truth, auto-segmentation performance was evaluated on 20 testing cases with multiple metrics including absolute and relative volume differences (AVD/RVD), Dice score, mean surface-to-surface distance (MSSD), and 95th percentile Hausdorff distance (HD95).
Results
The three models demonstrated comparable performance without significant differences. Mean Dice scores ranged from 0.89-0.90 across models. AVD ranged from 0.11-0.13cc, while RVD ranged from 10.70-13.44%. MSSD and HD95 values were (0.14-0.19mm) and (0.74-0.88mm) respectively. Average inference time was approximately 60 seconds per case. The platform successfully enabled web-based interaction, longitudinal tumor volume tracking and flexible visualization options including single and multiple image views and graphical representation of volume changes over time.
Conclusion
This work presents a comprehensive platform that combines automated segmentation with longitudinal tracking to support VS management. The AI models achieved Dice scores comparable to inter-observer variability in manual contouring, indicating clinical adequacy. The tracking capability provides consistency in treatment planning and monitoring and opens the possibility to advance AI-driven segmentation and streamline workflows for other intracranial pathologies.