Efficient and accurate identification of ear diseases using an ensemble deep learning model

Area of Science:

• Otolaryngology
• Medical Imaging
• Artificial Intelligence

Background:

• Early ear disease detection is crucial but often limited by expert availability and diagnostic accuracy.
• Deep learning offers a promising solution for automated, real-time ear disease diagnosis.

Purpose of the Study:

• To develop and evaluate a deep learning model for the automatic, real-time diagnosis of common ear diseases using otoscope images.
• To improve diagnostic accuracy and accessibility in clinical settings, especially where specialists are unavailable.

Main Methods:

• Trained nine deep convolution neural networks on 20,542 clinical endoscopic ear images.
• Classified eight common ear diseases including normal, Cholestestoma, Chronic suppurative otitis media, and Otomycosis.
• Utilized ensemble classifiers with transfer learning models (DensNet-BC169, DensNet-BC1615) for optimized performance.

Main Results:

• Achieved an average diagnostic accuracy of 95.59% using ensemble classifiers.
• The model demonstrated high accuracy attributed to a large, diverse dataset and training under various conditions.
• Selected DensNet-BC169 and DensNet-BC1615 transfer learning models for their balance of accuracy and training time.

Conclusions:

• The developed deep learning model provides a highly accurate and efficient tool for real-time ear disease diagnosis.
• This automated strategy is valuable for early detection and treatment, particularly in resource-limited clinical environments.
• The study highlights the significant potential of AI in advancing otolaryngological diagnostics and patient care.