Otitis Net: AI-Driven Scalable Paediatric Otitis Media Diagnosis from Otoscopic Images

Abstract

Otitis media represents a group of inflammatory disorders affecting the middle ear, which can lead to serious complications if not diagnosed in a timely manner. Accurate identification of these conditions remains challenging, highlighting the need for an efficient screening system supported by machine learning techniques. This work proposes an automated diagnostic framework based on deep learning for analyzing otoscopic images, aiming to address the growing prevalence of ear infections, especially among children, where such conditions are a major cause of hearing loss and frequent clinical visits. Despite the availability of digital otoscopic imaging, diagnosis is still largely dependent on manual examination by specialists, which can be subjective, time-intensive, and susceptible to variability, delays, and human error, particularly in low-resource environments. Traditional diagnosis involves distinguishing between conditions such as Acute Otitis Media, Chronic Otitis Media, Myringosclerosis, Cerumen Impaction, and normal cases, often without standardized criteria. To improve reliability and scalability, the proposed system incorporates image preprocessing, deep feature extraction using DenseNet121, and classification through multiple machine learning models including Nearest Centroid classifier (NCC), K-Nearest Neighbors (KNN), and eXtreme Gradient Boosting (XGBoost), along with a proposed ensemble approach that combines a calibrated perceptron and a dense neural network using a voting mechanism. The dataset is split into training and testing sets, and performance is assessed using metrics such as accuracy, precision, recall, F1-score, confusion matrix, ROC curves, and AUC. The experimental findings indicate that the ensemble voting model outperforms individual classifiers, offering a robust, efficient, and scalable solution for automated otitis media diagnosis and supporting clinical decision-making.