Towards Decentralized Dental Diagnostics: Federated Learning for Caries Detection

Dental caries is a prevalent oral health issue requiring early detection via X-rays for timely treatment. Traditional machine learning (ML) relies on centralized data, raising privacy concerns in medical imaging. Hospitals and clinics store de-centralized datasets, complicating AI-driven diagnostics without breaching privacy rules. This study employs Federated Learning (FL) to train a deep learning (DL) model across institutions while preserving privacy. For this task, we propose an FL framework using DenseNet201. The model classifies dental X-rays into two groups: those with caries and those without. It uses class weight computation and a dynamic learning rate scheduler to get better results. FL trains local models on decentralized hospital data, aggregating parameters via Weighted FedAvg (W-FedAvg) for balanced client contributions. Results demonstrate 70% classification accuracy while ensuring privacy, highlighting its medical imaging potential.