Related Experiment Video
Updated: Aug 18, 2025
08:43
A Unified Methodological Framework for Vestibular Schwannoma Research
Published on: June 20, 2017
7.5K
Joint Vestibular Schwannoma Enlargement Prediction and Segmentation Using a Deep Multi-task Model
Kai Wang1, Nicholas A George-Jones2,3, Liyuan Chen1
1The Department of Radiation Oncology, University of Texas Southwestern Medical Center, Dallas, Texas, USA.
The Laryngoscope
|December 10, 2022
Summary
A deep-learning multi-task model accurately segments vestibular schwannoma (VS) and predicts tumor enlargement. This AI tool shows potential for improved VS patient management and treatment planning.
Area of Science:
- Neurosurgery
- Radiology
- Artificial Intelligence
Background:
- Vestibular schwannoma (VS) management requires accurate tumor segmentation and prediction of growth.
- Current methods for tumor enlargement prediction (TEP) and tumor segmentation (TS) can be time-consuming and may lack precision.
Purpose of the Study:
- To develop and evaluate a deep-learning-based multi-task (DMT) model for simultaneous TS and TEP in VS patients.
- To utilize initial contrast-enhanced T1-weighted (ceT1) MRIs for joint prediction and segmentation.
Main Methods:
- A DMT model was developed for joint TS and TEP using retrospective ceT1 MRIs from 103 VS patients.
- The model was trained and evaluated using expert-contoured tumor volumes and a 20% volumetric growth threshold for enlargement.
Main Results:
- The DMT model achieved a median segmentation dice coefficient of 84.20% and an AUC of 0.77 for TEP.
- Performance was significantly better than separate single-task models for both TS and TEP.
- Segmentation was comparable to state-of-the-art methods, while TEP outperformed radiomics-based prediction.
Conclusions:
- The proposed DMT model demonstrates high learning efficiency and promising performance for VS TS and TEP.
- This AI-driven approach has the potential to enhance clinical decision-making and improve VS patient management.
