Smart Otoscope: Real-Time Ear Disorder Classification Using Embedded Machine Learning in Portable Diagnostic Devices

Abstract

The recent rapid development in embedded machine learning and portable medical devices has opened up new frontiers in real-time disease diagnosis, particularly in resource-limited settings. This paper proposes an integrated solution to classifying ear disorders using a CNN-based machine learning model implemented with Python. The pipeline begins with the collection of diverse medical imaging data from Kaggle that targets chest CT scans, EEG signals, and otoscopic images to simulate a multimodal healthcare setting in real life. The otoscopic images are emphasized for the primary task of ear disorder classification. Data preprocessing is imposed rigorously using normalization techniques such as Min-Max scaling and Z-score standardization to ensure consistent input distribution, which is crucial to accelerating model convergence as well as increasing generalization. Next, feature engineering is conducted using polynomial combinations and rolling means to add more pattern discovery in the imaging data. A light-weight CNN model is then constructed to extract spatial features and classify, optimized for potential deployment on low-power embedded systems. The model is contrasted with conventional metrics including accuracy, precision, recall, and F1-score and contrasted with other algorithms including KNN, XGBoost, and hybrid models. The results indicate that the proposed CNN model is superior to comparison methods in classification accuracy at 98.93%, demonstrating its power and practicality for use in portable diagnostic devices capable of enhancing point-of-care decision-making. The system coded in Python entirely provides an effective solution for intelligent, real-time medical diagnosis.