Deep learning image classification model for pediatric ear disease

Project Summary/Abstract Otitis media is the number one reason children seek medical care in the US with over 30 million visits each year. Otitis media encompasses acute otitis media (AOM, infection) and otitis media with effusion (OME, fluid not infected). Correct diagnosis is important for appropriate treatment. Pediatricians are currently about 50% accurate with diagnosis which often leads to children being treated with antibiotics when they are not needed. The application of artificial intelligence for medical diagnosis is on the rise. Artificial intelligence has been shown to out-perform doctors’ ability to diagnose based on visual exams for other diseases such as skin lesions. Artificial intelligence that can diagnose an image of an eardrum has been demonstrated in proof of concept. However, this technology has not been widely available or applicable to images that can be achieved at home by parents of young children. To address these drawbacks, we are validating an artificial intelligence algorithm that can diagnose ear disease in young children who have ear pain. Our algorithm has been trained with eardrum images that were obtained with equipment available to parents over the counter. Our algorithm will be applicable to images that can be achieved by parents at home. We are carrying out the studies needed for FDA clearance so that it can be commercialized. Aims of this proposal will be to 1) test the algorithm’s performance on images of children who have ear pain; and 2) demonstrate the ability of healthcare providers to use and understand the outputs of the algorithm when it assesses an ear image. The work that will result from this proposal will provide the data needed for the artificial intelligence algorithm to be cleared by FDA for use. After this clearance, the algorithm has tremendous potential of use by telemedicine platforms and healthcare systems.