Otoscopic diagnosis using computer vision: An automated machine learning approach.

Area of Science:

• Otolaryngology
• Artificial Intelligence
• Medical Imaging

Background:

• Access to otolaryngology services is limited by specialist shortages and long wait times, particularly in rural areas.
• Primary care providers often lack specialized tools for timely otologic diagnosis.
• AI offers potential solutions to bridge diagnostic gaps and improve patient care.

Purpose of the Study:

• To develop an automated machine learning (ML) computer vision algorithm for otoscopic diagnosis.
• To achieve diagnostic accuracy exceeding that of trained physicians.
• To enable primary care providers to perform timely referrals and triage for otologic conditions.

Main Methods:

• A dataset of 1,366 preprocessed otoscopic images was utilized.
• Images were annotated with 14 distinct otologic diagnoses.
• A multilabel classifier algorithm was trained using Google Cloud Vision AutoML, with performance evaluated against physician diagnoses.

Main Results:

• The AI algorithm demonstrated an average precision of 90.9% and recall of 86.1%.
• The overall accuracy of the algorithm was 88.7%, significantly higher than the average physician accuracy of 58.9%.
• The algorithm correctly diagnosed 79 out of 89 images in the test set.

Conclusions:

• An automated ML computer vision algorithm for otoscopic diagnosis can achieve accuracy comparable to or exceeding that of trained physicians.
• This AI tool has the potential to enhance diagnostic capabilities in primary care settings.
• Artificial intelligence is poised to significantly transform medical practice, necessitating physician adaptation and guidance.