New Pitt app diagnoses ear infections with a smartphone camera
Most parents of young children know that ear infections are incredibly common. What they might not know is that they are often hard to diagnose.
Ear infections are accurately diagnosed 75% of the time (or lower) in pediatric and primary care settings, according to multiple studies.
“It’s a little bit of a black hole in pediatrics,” said Alejandro Hoberman, professor of pediatrics and director of the Division of General Academic Pediatrics at Pitt’s School of Medicine. “To make a proper diagnosis, one needs to take into account features of the eardrum that we get a very quick view of in a wriggling baby.”
In an effort to improve the accuracy of ear infection diagnosis, Hoberman and his colleagues at Pitt and UPMC created a smartphone app that uses artificial intelligence to analyze video frames of the ear takenby a smartphone camera.
The tool, it turns out, was incredibly successful — diagnosing ear infections with almost 94% accuracy. Their results were published Mondayin JAMA Pediatrics.
“It’s impressive,” said Hoberman. “It got to be like 94% accurate, which is better than any studies of people.”
Ear infections are the second most common illness diagnosed in children (first is the common cold), and about 75% of babies are diagnosed with an ear infection before their first birthday, according to Hoberman. Ear infections are treated with antibiotics, but they can easily be mistaken for fluid behind the ear without an ear infection, which doesn’t benefit from antibiotic treatment.
The study trained the smartphone app by building a library of
1,151 videos of the ears of 635 children seen in UPMC offices between 2018 and 2023. For each video, two experts assessed whether the ear was infected or not in order to teach and test the AI model.
The increased accuracy using AI comes in part from the ability of the tool to capture and analyze video footage. It does so using a smartphone camera hooked up to an otoscope.
Hoberman recalled a recent visit where he examined a patient with symptoms of an ear infection. It took him about five minutes to get a good enough look at the ear so that he could make a determination, he said. He then tested the smartphone tool, which made the same diagnosisthat he did in seconds.
“I trained that neural network but it was faster than I was,” he said. “The video is a collection of frames and all you may need is two good frames to make a decision. My eyes are not fast enough to capture those things.”
The app is already being used in about 10 UPMC pediatricians’ offices and is available for use in research, said Hoberman. The university is in talks with multiple commercial partners to develop it further and to push for FDA approval.
Eventually, the tool could possibly be used by parents at home during a virtual doctor’s appointment, he said. The parent could attach their smartphone to an off-the-shelf otoscope available on Amazon, and then transmit the data from their phone to their pediatrician to confirm a diagnosis.
“If the parent can capture the video, the doctor can make a decision instead of dragging a patient to the MedExpress or urgent care,” said Hoberman.
As an added bonus, smartphone footage could be attached to the child’s medical records, or could be shown to parents to help them understand the course of treatment. A parent might push for antibiotics because they believe their child has an ear infection, and video footage could be helpful in convincing them otherwise.
“If you were thinking that your child has an ear infection because they were tugging at their ear,” said Hoberman, “looking at the proof of the pudding that it’s not an ear infection is really cool.”