Classical approaches and new deep learning trends to assist in accurately and efficiently diagnosing ear disease from otoscopic images

Rural communities in Australia have limited access to Ear, Nose and Throat (ENT) specialists, resulting in a lack of expertise to provide a diagnosis of complex and chronic ear diseases. This literature review examines previous attempts at creating a computer-aided tool to accurately diagnose ear disease and gaps in the literature. A systematic search was conducted to identify relevant papers and the latest best trends in technology. Four papers showed significant results in ear disease detection with deep learning models providing the best performance. Some studies using larger datasets consisting of endoscopic images obtained accuracies of over 90%. No adequate model was found that used otoscopic images with a sensitivity of over 90%. Endoscopic images provide better quality images, making it unclear how the models would perform on otoscopic images. Advanced techniques such as Transformers have not yet been tested in ear disease detection and could help improve model accuracy.