DEEP LEARNING FOR CARIES DETECTION IN DIAGNOCAM IMAGES: A SYSTEMATIC REVIEW OF CURRENT EVIDENCE AND CHALLENGES

Abstract

Dental caries remains a widespread chronic disease, with early detection being vital to prevent lesion progression and tooth loss. Conventional diagnostic methods, such as visual-tactile examination and bitewing radiographs, have limited sensitivity for early lesions, particularly in children. Near-infrared light transillumination (NILT), as used in the DIAGNOcam system, provides a radiation-free and non-invasive alternative for detecting occlusal and proximal caries. Recently, deep learning approaches, especially convolutional neural networks (CNNs), have shown great promise in automating dental image analysis and improving diagnostic accuracy.

This systematic review analyzed studies published from 2012 to 2025 on caries detection using DIAGNOcam images. A total of six clinical and three deep learning studies met the inclusion criteria. Primary outcomes included diagnostic accuracy, sensitivity, and specificity. Clinical studies confirmed DIAGNOcam’s reliability for early lesion detection, while AI-based approaches achieved comparable or superior performance to traditional methods.

However, limited datasets, small sample sizes, and inconsistent annotation standards restrict generalization. Future work should emphasize larger, standardized datasets and interpretable AI models. Combining DIAGNOcam imaging with deep learning may enable earlier, more objective, and patient-friendly caries diagnosis.