An artificial intelligence classifier as a screening tool to rule out otitis media in children

Objective: Acute otitis media is the most common bacterial infection among children and a significant global health burden. Despite its high incidence, diagnostic accuracy is poor. The objective of this study was to evaluate whether an artificial intelligence classifier can rule out otitis media in children based on a tympanic membrane image.

Methods: Artificial intelligence analysis of tympanic membrane images was carried out on images gathered as part of a randomized double-blind study. 793 tympanic membrane images were analyzed with an AI classifier. Images were obtained from children aged 6 to 35 months participating in a trial investigating the efficacy of amoxicillin-clavulanate for acute otitis media. The primary outcome was the sensitivity, specificity and accuracy of the classifier.

Results: All four variants of the artificial intelligence classifier showed excellent sensitivity for an abnormal ear (96% to 100%), and areas under the curves were respectively high (0.83-0.92). After a change in image normalization due to an initially poor image quality, the performance of the best variant improved to a specificity of 73%, and sensitivity remained high (92%).

Conclusions: Our study suggests that an artificial intelligence classifier at a primary level can rule out otitis media in children. This may eliminate the need for a physician's visit in the great majority of suspected acute otitis media cases in children with healthy ears. Further research in a parent-led setting is needed to measure the real-world impact of automatic classifiers.