Development of Deep Learning Techniques for Dental Caries Detection: A Systematic Literature Review

Abstrak

Dental caries remains one of the most prevalent oral health problems worldwide, requiring early and accurate detection to prevent extensive damage and reduce treatment costs. Recent advances in artificial intelligence particularly deep learning have led to significant improvements in diagnostic accuracy and consistency compared to traditional visual or radiographic assessments. This systematic literature review evaluates the development of deep learning techniques for dental caries detection based on studies published between 2020 and 2024. Following the PRISMA 2020 guidelines, six eligible studies were identified from Scopus, PubMed, IEEE Xplore, and ScienceDirect databases. The review synthesizes findings related to model architectures, imaging modalities, dataset characteristics, and performance metrics. The results show that models such as CNN, YOLO, U-Net, and EfficientNet consistently demonstrate high accuracy in identifying carious lesions, with bitewing and panoramic radiographs producing the most reliable diagnostic outcomes. However, limitations remain, including dataset variability, limited sample sizes, and reduced sensitivity for early-stage lesions. This review highlights current progress, methodological challenges, and potential research opportunities, emphasizing the need for standardized datasets, improved clinical validation, and stronger multidisciplinary collaboration to support the integration of deep learning into dental diagnostic workflows.