Multimodal Artificial Intelligence for the Clinical Assessment of Otitis Media

Acute otitis media (AOM) and otitis media with effusion (OME) can often be recognized from characteristic otoscopic findings. However, patients also present with subtle, overlapping, or resolving findings that are difficult to assign to a single diagnosis from an image alone. If these encounters are excluded because the reference diagnosis is uncertain, artificial intelligence (AI) models may perform well on selected images while providing less help in clinically uncertain cases. This focused narrative review examines whether symptoms, tympanometry, and hearing information can complement otoscopic images and improve the representation of clinically uncertain middle ear disease. Selected studies have combined images with acoustic measurements, pure-tone audiometry, or patient information. Their findings support feasibility, but differences in diagnostic targets and study design limit conclusions about benefit in uncertain AOM and OME. We propose that future systems account for diagnostically challenging cases during development and include them in clinical evaluation, while describing the current ear status alongside any diagnostic label. Large language models may help express supported findings, their clinical interpretation, and remaining uncertainty in readable language. Such explanations require evaluation independent of classification accuracy. The clinical value of multimodal AI should be established by its ability to support appropriate decisions in representative patients, including those with incomplete or discordant findings.