AI-Powered Cavity Detection: Leveraging Transfer Learning in Dental Images

Abstract

Dental cavities are a prevalent concern regarding one's dental health; untreated dental cavities can lead to pain, infection, and tooth loss. The conventional approaches employed in the diagnosis are radiographic evaluation and visual examination, which are limited by subjectivity and the risk of user error. This work demonstrates the diagnosis of dental cavities in dental images through automation with artificial intelligence (AI) using deep learning and transfer learning techniques. The proposed model is based upon an ensemble YOLO architecture for real-time detection of cavities and then leverages test-time augmentation to improve accuracy. The identified dental pathology is subsequently classified into observable changes with and without cavitation using transfer learning. Performance assessments using metrics demonstrated improved accuracy and efficiency in diagnosis. The challenge of the model differentiating cavities exposes the need for more precise optimization, despite the model's superior performance in caries detection. This work illustrates the positive implications AI-dental diagnosis systems can have on patient care, potentially decreasing diagnostic errors and improving early detection. Future work can strategically focus on expanding datasets and refining AI models to assist in the accurate and reliable identification of dental disease.