Development of Deep Learning Based Model for Efficient Diagnosis of Tympanic Membrane Infection and Referrals using Otoscopic Images

Infections in the tympanic membrane (in middle ear) is a serious middle ear ailment that mostly occurs in infants and younger children. It generally occurs as a result of mucosal disorders that causes impact on middle ear mucosa. Therefore, it requires on-time treatment and therapeutics. This work investigates the use of Inception v4 model for precise detection of some specific disorders in middle ear by taking inputs of otoscopy images. Acute Otitis Media (AOM), Acute Suppurative Otitis Media (ASOM), Chronic Suppurative Otitis Media (CSOM), Myringosclerosis (MS), and Normal Tympanic Membrane (TM) are the five different image classifications that make up the dataset, which was gathered and validated from AIIMS Raipur Hospital. Every category includes carefully chosen images for training and testing and their numbers depending on the rarity of the disease. While segregating different OM types and MS from normal TMs, binary classification models achieved remarkable accuracy, outperforming 95%. A noteworthy accuracy of 96.88% was attained by the model using AOM vs Normal dataset. In addition to this, multi-class classification model successfully classified TMs into AOM, ASOM, CSOM, MS and Normal TM with an accuracy of 93.75 %