Artificial Intelligence in Otoscopy for Tympanic Membrane Pathologies: A State-of-the-Art Scoping Review.

Abstract

This paper aimed to conduct a systematic search within a scoping review framework and provide a critical narrative appraisal of the current literature on artificial intelligence (AI) applications for tympanic membrane (TM) image analysis in the diagnosis of ear infections, with a focus on methodological rigor, dataset characteristics, and clinical applicability. A comprehensive search was performed across various online databases (Scopus, PubMed, Google Scholar, Embase, IEEE Xplore, and SPIE Digital Library) to identify studies published up to June 30, 2024. Studies employing AI for the analysis or classification of TM images in ear disease diagnostics were included. Inclusion and exclusion criteria were predefined. Data extraction covered dataset properties, imaging modalities, AI models, validation strategies, and performance metrics. The level of evidence, potential sources of bias, and methodological rigor were assessed qualitatively. Eighty-three studies met inclusion criteria, encompassing 90 experimental results. Considerable heterogeneity was observed regarding dataset size, image quality, device types, labeling protocols, and AI models. Most studies relied on small, private datasets, limiting model generalizability. While deep learning methods-especially convolutional neural networks-achieved high classification accuracies, methodological limitations included inconsistent data augmentation, imbalanced datasets, and lack of external validation. Few studies employed multidisciplinary teams or standardized protocols. Despite notable advances in AI-driven TM image analysis, direct comparison of studies remains challenging due to substantial heterogeneity and methodological shortcomings. A meta-analysis or systematic review was not performed, as quantitative synthesis would be misleading due to substantial data and labeling heterogeneity, wide methodological diversity, and a widespread lack of external validation across the included studies. Future research should prioritize larger, diverse, and well-annotated datasets, adoption of standardized protocols, and interdisciplinary collaboration to improve the clinical translation and reliability of AI-based diagnostic tools for ear infection management.

Abstract as supplied by the indexing source. © the authors / publisher.