Deep learning-based severity grading of Meniere's disease using 2D MRI.

Area of Science:

• Medical Imaging Analysis
• Deep Learning in Healthcare
• Inner Ear Disorders

Background:

• Meniere's disease (MD) is a debilitating inner ear condition with challenging diagnosis.
• Current diagnostic methods for MD lack accuracy and consistency.
• Advanced deep learning segmentation tools are needed for reliable MD assessment.

Purpose of the Study:

• To develop and evaluate the multi-stage severity assessment system (MSAS).
• MSAS is a deep learning framework for precise segmentation and severity stratification of MD.
• Utilizes 2D magnetic resonance imaging (MRI) for Meniere's disease assessment.

Main Methods:

• Developed MSAS using pixel-level manual segmentation on 189 patient MRI scans (development) and validated on 70 patients (external test set).
• Integrated sequence-level prediction and slice-level segmentation with techniques including YOLO-V5 and Grad-CAM.
• Evaluated performance using intersection over union (IoU), dice coefficient, accuracy, mean average precision (mAP), and area under the curve (AUC).

Main Results:

• MSAS achieved high overall accuracy (0.971) and AUC (0.995).
• Demonstrated strong vestibular region detection (mAP 0.887 internal, 0.877 external) and segmentation (dice 0.940 internal, 0.941 external).
• Slice attention maps and Grad-CAM provided enhanced interpretability for clinical decision-making.

Conclusions:

• Deep learning techniques show significant potential for improving Meniere's disease diagnosis and severity grading.
• The MSAS framework offers substantial clinical utility in MD management.
• The study advances the understanding of Meniere's disease pathophysiology through advanced imaging analysis.