Diagnosing ear infections in children is difficult; a new AI tool at UPMC could help

Using artificial intelligence technology, doctors may soon get a helping hand with diagnosing pediatric ear infections.

A new software tool developed by UPMC and Pitt scientists analyzes video captured of a child’s ear canal via an auriscope connected to a smartphone camera to help determine if the patient has an ear infection.

The AI technology could help doctors with the challenging task of diagnosing ear infections — a job that’s difficult when examining wiggly, fussy young children, said Dr. Judith Martin, professor of pediatrics at Pitt.

“It’s a challenge that people don’t necessarily appreciate how difficult it is,” Martin said. “Toddlers are impossible to wrangle, let alone try to come at them with a device when they’re not feeling well. To look into their ear to see the tympanic membrane, the eardrum, is definitely a challenge.”

The tool is based on a model built by the doctors using 1,151 ear canal videos taken from 635 children who visited outpatient UPMC pediatric offices between 2018 and 2023.

The researchers “taught” the software to analyze ear canals like a doctor by looking at features of the tympanic membrane, or eardrum, and identifying characteristics of an ear infection, or acute otitis media.

“We trained the neural network to think like we tend to think as we make that diagnosis,” said Dr. Alejandro Hoberman, senior author on the study. Hoberman is a professor of pediatrics, director of the division of general academic pediatrics at Pitt’s School of Medicine and president of UPMC Children’s Community Pediatrics.

The two AI models the team created had a high level of accuracy in determining whether a patient had an ear infection. Both had sensitivity and specificity values of greater than 93%, meaning that they had low rates of false negatives and false positives, the team said.

According to Hoberman, previous studies have had diagnostic accuracy ranges from 30% to 84%, depending on type of health care provider, level of training and age of the children being examined.

“These findings suggest that our tool is more accurate than many clinicians,” Hoberman said. “It could be a game changer in primary health care settings to support clinicians in stringently diagnosing acute otitis media and guiding treatment decisions.”

Use in diagnosis

Hoberman clarified that the tool is not a ‘generative AI’ technology that could have problems with misinterpreted data. Instead, it is based on data collected by the doctors — and doctors will always have the final say on whether to diagnose a child with an ear infection.

“The only thing that the AI generates is the prediction of the features — the bulging of the tympanic membrane, and a prediction of the diagnosis based on those features,” Hoberman said. “This is using decision support to help the doctors think about it.”

By comparing a video taken through an auriscope used to examine the ear with the database, the software can help determine whether an ear infection is the cause of a baby’s pain, or whether another issue — such as fluid behind the ear — could be the culprit.

Hoberman noted that the peak time of ear infections is between 6 months to 18-24 months.

“They’re babies, usually fussy and uncomfortable, and having a cold,” he said. “You have to get a good grip to be able to see the eardrum, and they have tiny ear canals, where a little bit of wax can get in the way.”

Dr. Kultar Shergill, pediatrician at Alle Kiski Pediatrics, explained younger children’s Eustachian tubes, which connect the ears to the sinuses, have more trouble draining fluid than in older children and adults. Because of this, they tend to have more ear pain and ear infections.

Accurate diagnosis is important to effectively treat the problem, he noted. Different ear problems have different treatments. Issues with the outer ear, such as ‘swimmer’s ear,’ can require antibiotic ear drops and not respond to antibiotic pills. Pain could also even be caused by foreign bodies, such as beads, that get stuck in the ear.

“For ear infections, you have to look in the ear to diagnose — it’s not something you can just diagnose on the phone,” he said.

Young children also cannot always articulate what part of the body is making them feel poorly, making diagnosis even more difficult, Martin noted.

The diagnostic tool would help doctors gather additional information and bolster their ability to accurately determine which cases are infections and which cases could have other causes.

“A device such as this is certainly a huge assistance. Sometimes that glimpse you get of the eardrum is so brief that you have to be completely paying attention,” she said. “Because of the movement of the child, the window of opportunity can be very small. A device such as this takes the video image and interprets it for you.”

Misdiagnosing ear infections is a constant concern because it could lead to excessive prescription of antibiotics, something doctors try to avoid.

“Twelve to 27 percent of unnecessary antibiotic prescription is related to the overdiagnosis of ear infections (when it is actually) fluid behind the ear,” Hoberman said. “(The tool) decreases underdiagnosis, which would result in inadequate care, and overdiagnosis, which would result in unnecessary antibiotic prescriptions.”

Educational potential

The tool is being used at 10 pediatric offices in conjunction with a CDC study, Hoberman said. But to get it into the hands of more doctors, the team will need to license and certify it as a medical software device, and find a commercial partner to manufacture it.

Once it’s available, Hoberman hopes the tool can also be used to teach pediatricians how to better diagnose ear infections.

“It takes many years and a lot of looking at ears to be able to have the skills to make an accurate diagnosis, because there’s a lot of nuances. It’s a skill that’s difficult to teach,” said UPMC’s Martin. “A tool such as this is really critical for helping to instruct the next generation of clinicians to be able to learn how to make this diagnosis.”

Shergill agreed that the tool seems useful, especially for teaching new pediatricians how to recognize problems inside the ear.

“The landmarks are not easily discernible. To look at a photo from a book and project that onto a real life image of the ear can be difficult,” he said. “A camera would be extremely helpful for learning.”

Though the tool utilizes advanced technology, Martin is confident that it will not “replace” doctors.

“It’s not the only tool,” she said. “The decision-making process is far too complex at this point, and there are other things that go into your assessment of this child, not just a 2 second clip of what the eardrum looks like. It’s meant to assist, not to replace.”