3D-Based Model Shown Most Effective in Segmenting Inner Ear for Calculation of EH Volume Ratio

CLINICAL QUESTION

How effective is a 3D-based deep neural network model (U-Net) used to segment the inner ear in accurately analyzing and calculating endolymphatic hydrops (EH) volume ratio?

BOTTOM LINE

The 3D U-Net model was shown to be superior in performance to the 2D model in analyzing the EH volume ratio, and the 3D model with data pairing performed better than the model without data pairing.

BACKGROUND: Endolymphatic hydrops is a pathological anatomical feature in which an increase in endolymphatic volume distends the structures enclosing the endolymphatic space. Image-based identification of EH may be key to understanding inner ear illnesses. Given its small anatomical size, however, segmentation of the inner ear and, thus, the endolymphatic space is very difficult.

STUDY DESIGN: Clinical trial

SETTING: Research Institute of Clinical Medicine of Jeonbuk National University–Biomedical Research Institute, Jeonbuk National University Hospital, Jeonju, Republic of Korea

SYNOPSIS: Researchers built a dataset of magnetic resonance cisternography (MRC) and HYDROPS Mi2 stacks labeled with the segmentation of the perilymph fluid space and endolymph fluid space of the inner ear to devise a 3D segmentation deep neural network model. They evaluated magnetic resonance imaging from 129 patients (71 females; mean age 51.2 years): 53 with dizziness and 76 with hearing loss. Precise segmentation was performed, and the EH volume ratio was calculated by dividing the volume of the endolymphatic space in the HYDROPS-Mi2 stack by that of the total fluid space in the MRC stack. The researchers performed experiments under various conditions, including the segmentation of single-task U-Net models without data pairing and the use of 2D U-Net and nnU-Net as backbones. Overall, the 3D model with data pairing showed the best performance. Authors note that, because histological confirmation of EH is impossible in living patients with related diseases, the accumulation of cases in which correlations have been made between EH imaging findings and clinical symptoms, as well as with results from various conventional functional ontological tests, is critical. Study limitations included the small size of the annotated datasets.

CITATION: Yoo T-W, et al. Automated volumetric analysis of the inner ear fluid space from hydrops magnetic resonance imaging using 3D neural networks. Sci Rep. 2024;14:24798. doi: 10.1038/s41598-024-76035-3.