Related Experiment Video
Updated: Jul 2, 2026
03:58
Enhancing Electrode Location Assessment in Cochlear Implantation via Computed Tomography Image Fusion
Published on: January 17, 2025
Artificial Intelligence-Assisted Inner Ear Computed Tomography Analysis: Radiomics-Based Comparison of Affected and
Joochan Choi1, Woogsang Sunwoo2, Kwanggi Kim3,4,5
1Department of Biomedical Engineering, College of Health & Sci., Gachon University, 1342, Seongnam-Daero, Sujeong-Gu, Gyeonggi-Do, 13120, Republic of Korea.
Journal of Imaging Informatics in Medicine
|July 1, 2026
Summary
Computed tomography (CT) radiomics of the inner ear did not reveal structural differences between affected and unaffected sides in idiopathic sudden sensorineural hearing loss (ISSNHL). Automated 3D segmentation was accurate, but CT-derived features lacked discriminatory value for ISSNHL.
Area of Science:
- Radiology and Medical Imaging
- Otolaryngology
- Artificial Intelligence in Medicine
Background:
- Idiopathic sudden sensorineural hearing loss (ISSNHL) is a complex condition with unclear pathophysiology.
- Computed tomography (CT) and radiomics offer potential for non-invasive assessment of inner ear structures.
- Automated segmentation models can enhance the efficiency and accuracy of radiomic feature extraction.
Purpose of the Study:
- To evaluate if CT-based radiomic features of the inner ear can differentiate the affected side from the contralateral normal-hearing side in patients with ISSNHL.
- To assess the utility of a fully automated three-dimensional (3D) deep learning segmentation model for inner ear analysis in ISSNHL.
- To explore subtle structural differences potentially associated with ISSNHL using advanced imaging techniques.
Main Methods:
- Retrospective analysis of 420 temporal bone CT scans from 318 patients, with a focus on 42 inner ear volumes from 21 unilateral ISSNHL patients.
- Training and validation of a SwinUNETR-based 3D segmentation model for automated inner ear delineation.
- Extraction of 1316 radiomic features (original, wavelet, Laplacian of Gaussian) using PyRadiomics, followed by statistical analysis (Welch's t-test, Mann-Whitney U test) with FDR correction.
Main Results:
- The automated 3D segmentation model achieved high and stable performance in delineating inner ear structures on CT scans.
- No statistically significant differences in CT-based radiomic features were found between ISSNHL-affected and normal-hearing inner ear sides after correction.
- Dimensionality reduction techniques (PCA, UMAP) showed no distinct clustering, and key shape features exhibited overlapping distributions.
Conclusions:
- CT-derived morphological radiomic features do not appear to identify measurable structural differences between affected and contralateral sides in ISSNHL.
- The findings suggest that the pathophysiology of ISSNHL may be primarily functional or microstructural, rather than gross morphological changes detectable by CT radiomics.
- While automated 3D segmentation is effective, CT radiomics has limited discriminatory value for ISSNHL; alternative imaging biomarkers or deep learning features may be needed for etiological assessment.
