OREHAS: A fully automated deep-learning pipeline for volumetric endolymphatic hydrops quantification in MRI

TL;DR

The paper addresses the challenge of objectively quantifying endolymphatic hydrops (EH) from routine inner-ear MRI by introducing OREHAS, a fully automatic, end-to-end pipeline that processes 3D-SPACE-MRC and 3D-REAL-IR images to compute per-ear ELR. The pipeline combines three modules—EarGate for slice classification, AuriBox for inner-ear localization, and EHMasker for sequence-specific segmentation—followed by volumetric ELR calculation, all trained with sparse slice annotations. External validation demonstrates that OREHAS closely matches manual ground truth and outperforms proprietary syngo.via in REAL-IR measurements, highlighting improved accuracy and physiological realism of EH quantification. The approach offers an open-source, modular framework with potential to recalibrate clinical thresholds and enable large-scale studies, while reducing operator dependence and ensuring reproducibility. These results establish a robust foundation for automated EH analysis and integration into clinical workflows as a syngo.via OpenApp.

Abstract

We present OREHAS (Optimized Recognition & Evaluation of volumetric Hydrops in the Auditory System), the first fully automatic pipeline for volumetric quantification of endolymphatic hydrops (EH) from routine 3D-SPACE-MRC and 3D-REAL-IR MRI. The system integrates three components -- slice classification, inner ear localization, and sequence-specific segmentation -- into a single workflow that computes per-ear endolymphatic-to-vestibular volume ratios (ELR) directly from whole MRI volumes, eliminating the need for manual intervention. Trained with only 3 to 6 annotated slices per patient, OREHAS generalized effectively to full 3D volumes, achieving Dice scores of 0.90 for SPACE-MRC and 0.75 for REAL-IR. In an external validation cohort with complete manual annotations, OREHAS closely matched expert ground truth (VSI = 74.3%) and substantially outperformed the clinical syngo.via software (VSI = 42.5%), which tended to overestimate endolymphatic volumes. Across 19 test patients, vestibular measurements from OREHAS were consistent with syngo.via, while endolymphatic volumes were systematically smaller and more physiologically realistic. These results show that reliable and reproducible EH quantification can be achieved from standard MRI using limited supervision. By combining efficient deep-learning-based segmentation with a clinically aligned volumetric workflow, OREHAS reduces operator dependence, ensures methodological consistency. Besides, the results are compatible with established imaging protocols. The approach provides a robust foundation for large-scale studies and for recalibrating clinical diagnostic thresholds based on accurate volumetric measurements of the inner ear.

Figures (6)

• Figure 1: Overview of the EH‐Ratio analysis pipeline. In both training/evaluation (top, peach) and inference (bottom, blue) phases, SPACE‐MRC and REAL‐IR volumes are sequentially processed by three deep learning modules: (1) EarGate performs slice selection to retain inner ear anatomy: (2) AuriBox localizes the region of interest through bounding box detection, and (3) EHMasker generates segmentation masks for the vestibular and endolymphatic spaces. The reconstructed 3D masks are subsequently used for volumetric quantification of the EH ratio.
• Figure 2: Representative axial slices from a single subject: (a) 3D-SPACE-MRC image showing the delineated vestibular cavity (hyperintense, black contour); (b) 3D-REAL-IR image depicting the endolymphatic compartment (hypointense, white contour).
• Figure 3: Representative AuriBox detections across both modalities, showing precise ear localization. (Left) SPACE-MRC: predicted bounding boxes for left (blue) and right (cyan) ears; (Right) REAL-IR: bounding boxes aligned with manual annotations.
• Figure 4: Qualitative segmentation examples for both modalities. Panels A–B: 3D-SPACE-MRC; Panels C–D: 3D-REAL-IR. The model prediction (green) and the ground-truth annotation (blue) are overlaid on the MRI slice; the overlap region is shown in yellow. These examples illustrate the close agreement between predicted masks and expert annotations across modalities.
• Figure 5: Comparison of Endolymph-to-Vestibule Ratio (ELR) values for the five fully annotated patients. Results are shown per ear for three sources: OREHAS predictions (blue), manual ground truth (orange), and syngo.via software outputs (green).