Related Experiment Video
Updated: Sep 2, 2025
A Unified Methodological Framework for Vestibular Schwannoma Research
Published on: June 20, 2017
Fully Automated 3D Vestibular Schwannoma Segmentation with and without Gadolinium-based Contrast Material: A
Olaf M Neve1, Yunjie Chen1, Qian Tao1
1Department of Otorhinolaryngology and Head & Neck Surgery (O.M.N., N.P.d.B., J.C.J., E.F.H.), Division of Image Processing, Department of Radiology (Y.C., Q.T., B.P.F.L., M.S.), and Department of Radiology (S.R.R., W.G., M.C.K., B.M.V.), Leiden University Medical Center, Otorhinolaryngology H5-P, PO Box 9600, 2300 RC Leiden, the Netherlands; and Knowledge Driven AI Lab, Delft University of Technology, Delft, the Netherlands (Q.T.).
This study developed a convolutional neural network (CNN) for automated vestibular schwannoma measurements on MRI scans. The CNN achieved accurate tumor detection and delineation, comparable to human experts.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Medicine
- Neurosurgery
Background:
- Vestibular schwannomas require accurate measurement for diagnosis and treatment planning.
- Manual delineation of tumors on MRI can be time-consuming and subject to inter-observer variability.
- Automated segmentation tools can potentially improve efficiency and consistency.
Purpose of the Study:
- To develop and validate automated measurement of vestibular schwannomas using convolutional neural networks (CNNs).
- To assess the performance of CNN models on both T1- and T2-weighted contrast-enhanced MRI scans.
- To compare automated delineations with human expert assessments.
Main Methods:
- Retrospective analysis of MRI data from 214 patients across 37 centers.
- Training and validation of two CNN models (T1 and T2) using fivefold cross-validation.
- Quantitative evaluation using Dice index, Hausdorff distance, and surface-to-surface distance (S2S), alongside an observer study.
Main Results:
- The T1-weighted CNN model demonstrated state-of-the-art performance with a mean S2S distance < 0.6 mm.
- The T2-weighted CNN model also achieved high accuracy with a mean S2S distance < 0.6 mm.
- Observer studies showed the automated tool was similar to human delineations in 85%-92% of cases.
Conclusions:
- CNN models accurately detect and delineate vestibular schwannomas on contrast-enhanced T1- and T2-weighted MRI.
- The models effectively distinguish between intrameatal and extrameatal tumor components.
- Automated segmentation offers a reliable and efficient alternative to manual measurements.
