Deep convolutional neural networks for early detection of interproximal caries using bitewing radiographs: A systematic review.

OBJECTIVES: To thoroughly review Deep Convolutional Neural Networks for detecting interproximal caries with bitewing radiographs. DATA: Data was collected from studies that utilized Deep Convolutional Neural Networks (DCNN) focused on the analysis of bitewing radiographs taken with intraoral X-ray units. SOURCES: A comprehensive literature search was conducted across various scholarly databases including Google Scholar, MDPI, PubMed, ResearchGate, ScienceDirect, and IEEE Xplore, encompassing 2014 to 2024. The risk of bias assessment utilized the current version of the Quality Assessment Tool for Diagnostic Accuracy Studies (QUADAS-2). STUDY SELECTION: After reviewing 291 articles, 10 studies met the criteria and were analyzed. All 10 studies used bitewing radiographs, focusing on deep learning tasks such as segmentation, classification, and detection. The sample sizes varied widely from 112 to 3,989 participants. Convolutional neural networks (CNNs) were the most commonly used model. According to the QUADAS-2 assessment, only 40% of the studies included in this review were found to have a low risk of bias in the reference standard domain. CLINICAL SIGNIFICANCE: A Deep Convolutional Neural Networks based caries detection system helps in the early identification of caries by analyzing bitewing radiographs and reduces diagnostic errors. By identifying early-stage lesions, patients can undergo minimally invasive treatments instead of more complex procedures, thereby improving patient outcomes in dental care. CONCLUSION: This systematic review provides an overview of various studies that utilize deep learning models to identify interproximal caries lesions in bitewing radiographs. It highlights the efficacy of YOLOv8 in detecting interproximal caries from bitewing radiographs compared to other Deep CNN models.