Table 5.

Advanced AI applications in pediatric dentistry.

Authors	Summarized Abstract	Methods Used	Results	Conclusions
You [48]	-AI model detected dental plaque on primary teeth with high accuracy.	-A pediatric dentist manually identified plaque on photos before and after applying a plaque-disclosing agent. Compared AI and manual diagnostic methods on an additional 102 intraoral photos.	-Mean intersection over union for detecting plaque was 0.726 ± 0.165. -AI model had higher MIoU compared to the dentist.	-AI model performs well in detecting dental plaque on primary teeth. -AI technology has the potential to enhance pediatric oral health.
Bilgir [47]	-AI system detected and numbers teeth on panoramic radiographs successfully.	-Developed AI (CranioCatch, Eşkişehir, Turkey) to detect and number teeth, tested on 249 panoramic radiographs.	-AI system successfully detected and numbered teeth on panoramic radiographs. -Estimated sensitivity, precision, and F-measure were 0.9559, 0.9652, 0.9606.	-AI can support clinicians, potentially replacing human observers in the future.
Duman [50]	-CNN-based AI model detected taurodontism in panoramic radiography effectively.	-Utilized 434 anonymized panoramic radiographs for developing automatic taurodont tooth segmentation models with a Pytorch-implemented U-Net.	-Sensitivity, precision, and F1-score values CNN system for taurodont tooth segmentation achieved results close to expert level in detecting taurodontism.	-CNN identifies taurodontism with results close to expert level.
Çalışkan [46]	-Deep learning for submerged primary tooth classification and detection.	-Employed Faster R-CNN architecture for detecting and classifying submerged molars. -Process involved defining molar boundaries on radiographs.	-Deep CNN showed high specificity in detecting submerged molars on radiographs. -AI approaches like deep CNN were promising for interpreting dental images.	-Further studies needed to enhance sensitivity for clinical application.
Ahn [45]	-Developed deep learning models for mesiodens classification in panoramic radiographs.	-Developed using SqueezeNet, ResNet-18, ResNet-101, and Inception-ResNet-V2.	-ResNet-101 and Inception-ResNet-V2 had accuracy over 90%. -SqueezeNet showed relatively inferior results in mesiodens classification.	-Deep learning models can aid in accurate and faster diagnosis.
Zhao [44]	-Developed AI tool for adenoid hypertrophy assessment in children.	-Developed an automated tool for adenoid hypertrophy assessment using a convolutional neural network, assessing regions defined by four key landmarks.	-High sensitivity, specificity, and accuracy for adenoid hypertrophy assessment. -Area under the receiver operating characteristic curve was 0.987.	-Automated system accurately assesses adenoid hypertrophy from lateral cephalograms.
Koopaie [49]	-Study compared cystatin S levels in early childhood caries patients. -Conducted a cross-sectional, case–control study.	-Collected unstimulated whole saliva samples using suction and measured cystatin S concentrations via ELISA. -Machine learning used to predict ECC based on cystatin S.	-Salivary cystatin S levels were significantly lower in early childhood caries. -Machine learning methods improved ECC prediction with cystatin S levels.	-Machine learning models enhance ECC prediction using cystatin S and demographics.