ToothAid: A system for early detection of oral conditions

Abstract

Due to reliance on radiographic imaging, visual inspection, and limited dental knowledge, early identification of gingivitis, dental
caries, dental plaque and gingivitis remains limited in remote and resource constrained settings, affecting billions of people globally. In order to democratise radiation free oral health screening, this paper proposed ToothAid, an Internet of Things enabled dental diagnostic assistance. The system uses a Raspberry Pi 4 and Camera Module v3 to capture visible light intraoral
pictures. It then uses a two-stage deep learning pipeline that includes a YOLOv8 model for realtime tooth localisation and a convolutional neural network for multiclass illness detection. Effective offline edge inference is made possible by model
quantisation and TensorFlow Lite deployment. ToothAid is a scalable point-of-care system for early dental disease identification, as demonstrated by experimental results that show good precision and recall with low inference latency.