OtoDx prototype device fits over a smartphone camera and diagnoses in seconds.
An AI tool for diagnosing ear infections in children has proven to be 30% more accurate than clinicians’ evaluations.
The ‘mini otoscope’ is attached to a phone’s camera so doctors can snap pictures of the inside of a child’s ear. The images are then uploaded to an app and the resulting diagnosis is produced within seconds.
OtoDx, a machine learning model developed at Mass Eye and Ear in Massachusetts, was more than 95% accurate in its diagnosis compared to 65% accuracy from a group of pediatricians, primary care doctors and ENTs. The findings were published in Otolaryngology-Head and Neck Surgery.
The AI-powered diagnostic tool can be used to assist pediatricians with a second opinion to make informed clinical decisions.
It can be difficult to assess children for ear infections because they’re usually under distress when being examined. If ear infections aren’t treated, kids can experience hearing loss, developmental delays, serious illnesses like meningitis, or even death in some places.
However, if antibiotics are overprescribed, children could develop antibiotic resistance, leading to challenges in fighting off other infections.
“This model won’t replace the judgment of clinicians but can serve to supplement their expertise and help them be more confident in their treatment decisions,” said Matthew Crowson, MD, the study’s lead author and an otolaryngologist at Mass Eye and Ear.
The deep learning model was trained using around 640 images of tympanic membranes from pediatric patients who had fluid in the ears or recurrent ear infections. They were all about to undergo surgery for fluid drainage or tube placements in the ears. The pictures were categorized as ‘infected,’ ‘normal,’ or ‘liquid behind the eardrum.’ The mean diagnostic accuracy was 80.8%.
The classification was more refined than a study done last year when the images were marked as ‘abnormal’ or ‘normal.’
OtoDx is collaborating with Mass General Brigham Innovation to bring the prototype device to clinic and commercialization.