VOLT: a novel open-source pipeline for automatic segmentation of endolymphatic space in inner ear MRI

• Published in: Journal of neurology 2020-12, Vol.267 (Suppl 1), p.185-196
• Main Authors: Gerb, J., Ahmadi, S. A., Kierig, E., Ertl-Wagner, B., Dieterich, M., Kirsch, V.
• Format: Article
• Language: English
• Subjects: Age; Algorithms; Datasets; Edema; Etiology; Gadolinium; Inner ear; Intravenous administration; Magnetic resonance imaging; Medicine; Medicine & Public Health; Neurology; Neuroradiology; Neurosciences; Original Communication; Segmentation; Vertigo
• ISSN: 0340-5354; 1432-1459
• DOI: 10.1007/s00415-020-10062-8

Background Objective and volumetric quantification is a necessary step in the assessment and comparison of endolymphatic hydrops (ELH) results. Here, we introduce a novel tool for automatic volumetric segmentation of the endolymphatic space (ELS) for ELH detection in delayed intravenous gadolinium-enhanced magnetic resonance imaging of inner ear (iMRI) data. Methods The core component is a novel algorithm based on Volumetric Local Thresholding (VOLT). The study included three different data sets: a real-world data set (D1) to develop the novel ELH detection algorithm and two validating data sets, one artificial (D2) and one entirely unseen prospective real-world data set (D3). D1 included 210 inner ears of 105 patients (50 male; mean age 50.4 ± 17.1 years), and D3 included 20 inner ears of 10 patients (5 male; mean age 46.8 ± 14.4 years) with episodic vertigo attacks of different etiology. D1 and D3 did not differ significantly concerning age, gender, the grade of ELH, or data quality. As an artificial data set, D2 provided a known ground truth and consisted of an 8-bit cuboid volume using the same voxel-size and grid as real-world data with different sized cylindrical and cuboid-shaped cutouts (signal) whose grayscale values matched the real-world data set D1 (mean 68.7 ± 7.8; range 48.9–92.8). The evaluation included segmentation accuracy using the Sørensen-Dice overlap coefficient and segmentation precision by comparing the volume of the ELS. Results VOLT resulted in a high level of performance and accuracy in comparison with the respective gold standard. In the case of the artificial data set, VOLT outperformed the gold standard in higher noise levels. Data processing steps are fully automated and run without further user input in less than 60 s. ELS volume measured by automatic segmentation correlated significantly with the clinical grading of the ELS ( p