Performance evaluation results of an artificial neural network developed for the purpose of classifying otoendoscopic images

The article describes an attempt to implement an automated approach in the diagnosis of ear diseases using a convolutional neural network. In the course of the work, a dataset consisting of 8791 images obtained during human otoendoscopic examination was formed, labelled and uploaded. The neural network was trained and tested. To organize the work of the algorithm, a tree of diagnoses was created and classes of images were defined: normal, defect of the unstretched section of the tympanic membrane, adhesive otitis media, foreign body of the external auditory canal, neotympanic membrane, sulfur plug, shunt, exudative otitis media, exostoses and neoplasms of the external auditory canal, diffuse otitis media, defect of the unstretched section of the tympanic membrane. The developed and trained artificial neural network demonstrated an accuracy of 91.2% in recognising different nosological classes related to the middle ear and diseases of external auditory canal. The proposed technology can be further used in medical practice to control and improve the quality of diagnostics of ear pathologies.