Artificial intelligence outperforming human doctors in a type of medical diagnosis

A machine learning model (a modality of artificial intelligence) has diagnosed pediatric ear infections more accurately than human doctors, and has done so 30 percent faster than them.

The model, called OtoDX, was more than 95 percent accurate in diagnosing an ear infection on a set of 22 test images, compared with 65 percent accuracy achieved by a group of doctors who had otolaryngologists, pediatricians, and general practitioners, and who reviewed the same images.

When tested on a dataset of more than 600 inner ear images, OtoDX had a diagnostic accuracy of more than 80 percent, which represents a significant jump over the average physician accuracy reported in the medical literature.

The team of Dr. Matthew Crowson, an otolaryngologist and artificial intelligence researcher, from the Massachusetts Eye and Ear Hospital, United States, details these comparative tests in the academic journal Otolaryngology Head & Neck Surgery, under the title “Human vs Machine, Validation of a Deep Learning Algorithm for Pediatric Middle Ear Infection Diagnosis”.

The model uses a type of machine learning called “deep learning” and was prepared from hundreds of medical photos of children before they underwent surgery at the hospital for recurrent ear infections or fluid in the ears.

OtoDx is currently being used in a prototype device combined with a smartphone app. The device acts like a “mini-otoscope” that would attach to a phone’s camera and allow doctors to take photos of the inside of a child’s ear, upload them directly to the app, and receive a diagnostic reading in seconds. (Photo: Mass Eye and Ear)

The results achieved by OtoDX are an important step towards developing a diagnostic tool that may one day be used in clinics to assist clinicians during patient evaluations. An AI-based diagnostic tool can offer clinicians a way to back up their diagnosis or help them figure out difficult cases.

“Ear infections are incredibly common in children, and yet they are often misdiagnosed, leading to delays in care or unnecessary prescriptions of antibiotics,” explains Dr. Crowson. “This model will not replace clinicians’ judgment, but it can serve to complement their expertise and help them be more confident in their treatment decisions.” (Font: NCYT by Amazings)