Keywords
TinyML
Multi-class classification
Edge impulse
Otoscope camera
How to Cite
Abstract
This study presents a TinyML-based machine learning system for multi-class ear condition classification on an edge device, designed to process images captured from a digital otoscope camera. Utilizing a dataset comprising five categories, Normal, Acute Otitis Media (AOM), Cerumen Impaction (CI), Chronic Otitis Media (COM), and Myringosclerosis (MYS), the proposed system performs near real-time classification directly on a resource-constrained microcontroller. The model was developed and optimized using the Edge Impulse platform and deployed as a quantized TinyML library. The system architecture incorporates lightweight convolutional neural networks (CNNs) and the EON™ Compiler to ensure efficient memory usage while maintaining high diagnostic performance. Experimental results demonstrate a validation accuracy of 97.5% and a testing accuracy of 96.31%, with a peak RAM footprint of only 240.3K and an inferencing latency of 1482 ms. These findings highlight the potential of TinyML for portable, low-power medical applications, providing a foundation for privacy-preserving, GDPR-compliant, on-device diagnostics in auditory healthcare without reliance on cloud infrastructure.
References
Al-Rahim Habib, A., Faruque, M., & Islam, A. M. (2022). Artificial intelligence to classify ear disease from otoscopy: A systematic review and meta-analysis. Clinical Otolaryngology, 47, 1354–1365.
Alaca, Y., & Akmeşe, Ö. F. (2025). Pancreatic tumor detection from CT images converted to graphs using whale optimization and classification algorithms with transfer learning. International Journal of Imaging Systems and Technology, 35, e70040.
Bingol, H. (2022). Classification of OME with eardrum otoendoscopic images using hybrid-based deep models, NCA, and Gaussian method. Traitement du Signal, 39.
Cioflan, C., Fonseca, J., Wang, X., & Benini, L. (2025). Nanohydra: Energy-efficient time-series classification at the edge. arXiv preprint, arXiv:2510.20038.
Demircan, F., Ekinci, M., Cömert, Z., & Gedikli, E. (2025). Enhanced classification of ear disease images using metaheuristic feature selection. Sakarya University Journal of Computer and Information Sciences, 8, 58–75.
Diez, P. L., Sundgaard, J. V., Margeta, J., Diab, K., Patou, F., et al. (2024). Deep reinforcement learning and convolutional autoencoders for anomaly detection of congenital inner ear malformations in clinical CT images. Computerized Medical Imaging and Graphics, 113, 102343.
Edge Impulse. (2025). Edge Impulse documentation. Accessed Sep. 01, 2025.
Fawcett, T. (2006). An introduction to ROC analysis. Pattern Recognition Letters, 27, 861–874.
Han, S., Mao, H., & Dally, W. J. (2015). Deep compression: Compressing deep neural networks with pruning, trained quantization and Huffman coding. arXiv preprint, arXiv:1510.00149.
Heydari, S., & Mahmoud, Q. H. (2025). Tiny machine learning and on-device inference: A survey of applications, challenges, and future directions. Sensors, 25, 3191.
Hizem, M., Bousbia, L., Ben Dhiab, Y., Aoueileyine, M. O. E., & Bouallegue, R. (2025). Reliable ECG anomaly detection on edge devices for Internet of Medical Things applications. Sensors, 25, 2496.
Hymel, S., Banbury, C., Situnayake, D., Elium, A., Ward, C., et al. (2022). Edge Impulse: An MLOps platform for tiny machine learning. arXiv preprint, arXiv:2212.03332.
Kwon, C. K. (2023). Development of embedded machine learning finger number recognition application using Edge Impulse platform. In 2023 Congress in Computer Science, Computer Engineering, & Applied Computing (CSCE) (pp. 2697–2699). IEEE.
Livingstone, D., & Chau, J. (2020). Otoscopic diagnosis using computer vision: An automated machine learning approach. The Laryngoscope, 130, 1408–1413.
Mihigo, I. N., Zennaro, M., Uwitonze, A., Rwigema, J., & Rovai, M. (2022). On-device IoT-based predictive maintenance analytics model: Comparing TinyLSTM and TinyModel from Edge Impulse. Sensors, 22, 5174.
Moreau, L. (2024). Introducing: EON compiler RAM-optimized. Accessed Sep. 01, 2025.
Özdilli, Ö., Şevik, S., Alaca, Y., & Uzunoğlu, Y. (2025). Optimal design of flat plate fin heat sinks using a computational fluid dynamics (CFD) and deep learning (DL)-based ensemble approach with explainable artificial intelligence (XAI) integration. Applied Thermal Engineering, p. 127547.
Rana, A., Dhiman, Y., & Anand, R. (2022). Cough detection system using TinyML. In 2022 International Conference on Computing, Communication and Power Technology (IC3P) (pp. 119–122). IEEE.
Rehman, Z. U., Fauzi, M. F. A., Lokman, F. N. I., Touhami, M., & Saim, L. (2025). Efficient and interpretable otoscopic image classification via distilled CNN with adaptive channel attention. IEEE Access.
Rust, E. (2020). Getting started with Edge Impulse. Accessed Dec. 03, 2025.
Schizas, N., Karras, A., Karras, C., & Sioutas, S. (2022). TinyML for ultra-low power AI and large-scale IoT deployments: A systematic review. Future Internet, 14, 363.
Song, D., Song, I. S., Kim, J., Choi, J., & Lee, Y. (2022). Semantic decomposition and anomaly detection of tympanic membrane endoscopic images. Applied Sciences, 12, 11677.
Tatlı, Y. (2025). Ear pathologies using deep learning on otoscopic images. Uluslararası Sürdürülebilir Mühendislik ve Teknoloji Dergisi, 9, 51–57.
Tjoa, E., & Guan, C. (2020). A survey on explainable artificial intelligence (XAI): Toward medical XAI. IEEE Transactions on Neural Networks and Learning Systems, 32, 4793–4813.
UCI Machine Learning Repository. (2025). Otoscopic image dataset. Kaggle, accessed Dec. 03, 2025.
Wang, A., Chen, H., Liu, L., Chen, K., Lin, Z., et al. (2024). YOLOv10: Real-time end-to-end object detection. Advances in Neural Information Processing Systems, 37, 107984–108011.
Warden, P., & Situnayake, D. (2019). TinyML: Machine Learning with TensorFlow Lite on Arduino and Ultra-Low-Power Microcontrollers. O’Reilly Media.
Xie, Y., & Fang, Q. (2025). An energy-aware generative AI edge inference framework for low-power IoT devices. Electronics, 14, 4086.
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.