Using machine learning to improve otitis media diagnosis

Acute and chronic otitis media are common reasons for young children to visit the pediatrician. However, misdiagnosing either condition, which may lead to either undertreatment or overtreatment such as unnecessary and inappropriate use of antibiotics, could lead to significant outcomes. A report in Pediatrics examined whether machine learning could help accurately predict middle ear effusion.¹

The investigators used a neural network that was trained to classify images as either “abnormal” (effusion present) or “normal” (no effusion). They used images of tympanic membranes from children who had been taken to an operating room with the intention of either performing a myringotomy and possible tube placement to treat either otitis media with effusion or recurrent acute otitis media.

The average training time in the neural network model was 76.0 seconds. The investigators found that the model had an average image classification accuracy of 83.8% (95% CI: 82.7–84.8). Further bolstering the accuracy of the model, they found that it produced an area under the receiver operating characteristic curve performance of 0.93 (95% CI: 0.91–0.94) as well as an F1-score of 0.80 (95% CI: 0.77–0.82).

The investigators noted that the accuracy found in the model, which used a small data set of images, was higher than the performance of human-experts using otoscopy, which had been seen in a number of previous studies. They concluded that using an artificial intelligence-assisted diagnosis algorithm could improve point-of-care diagnostic accuracy of acute or chronic otitis media, which could improve outcomes for children.

Contemporary Pediatrics reached out to editorial advisory board member Andrew J. Schuman, MD, FAAP, to ask his thoughts on using machine learning for otitis media diagnosis. He commented, "in general diagnosis of otitis media is often straightforward. Artificial intelligence might be more important for the home diagnosis of otitis media via telehealth devices, once technologies have been developed that have been shown reliable - ie interpretation of images of the tympanic memberane via machine learning."

Reference

1. Crowson M, Hartnick C, Diercks G, et al. Machine learning for accurate intraoperative pediatric middle ear effusion diagnosis. Pediatrics. March 17, 2021. Epub ahead of print. doi:10.1542/peds.2020-034546