Deep Learning Techniques for Ear Diseases Based on Segmentation of the Normal Tympanic Membrane

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Otolaryngology

Background:

• Otitis media is a prevalent global infection.
• Telemedicine advancements necessitate improved diagnostic tools for ear conditions.
• Current diagnostic strategies for tympanic membrane (TM) abnormalities lack sufficient explainability for clinical use.

Purpose of the Study:

• To develop an explainable deep learning model for diagnosing ear diseases using otoscopic images.
• To enhance the diagnostic accuracy and screening of otologic conditions.

Main Methods:

• A deep convolutional neural network (CNN) model, Mask R-CNN, was employed to segment normal TM substructures.
• A fully connected neural network evaluated combinations of five TM substructures (malleus, umbo, cone of light, pars flaccida, annulus) for disease detection.

Main Results:

• The algorithm achieved a high area under the curve (0.911) for detecting TM abnormalities using a combination of the malleus, cone of light, and umbo.
• This combination outperformed individual substructure analysis (AUC 0.737-0.873).
• The model demonstrated explainability and effectiveness in screening abnormal TMs.

Conclusions:

• The automated algorithm significantly improves diagnostic accuracy for normal versus abnormal TMs.
• This tool can facilitate timely referrals, enhancing patient quality of life in primary care.
• Explainable AI holds promise for advancing otologic diagnostics.