Table 3.

ANN selected studies.

Authors Name and Year	Methods	Results	Authors Suggestions/Conclusions
Faria et al., (2021) [41]	Custom-made ANN	Detect accuracy = 98.8%, predict accuracy = 99.2%, AUC= 0.9886, 0.9869	This approach may be beneficial for detecting and predicting the RRC’s development in other photos.
Li et al., (2021) [42]	DeepLabv3+, Xception and MobileNetV2	AUC = 0.7, precision = 0.606, recall = 0.415, mIOU = 0.650	Small dataset was used and data augmentation cannot overcome all biases present in small dataset.
Geetha et al., (2020) [43]	Customized BPNN	Accuracy = 97.1%, false positive (FP) rate = 2.8%, ROC area = 0.987, PRC area = 0.987	High quality datasets and improved algorithm can demonstrate good results towards dental practice.
Zanella-Calzada et al., (2018) [44]	Customized ANN	Accuracy = 0.69, AUC values = 0.69 and 0.75	This model can help dentists by providing an easy, free and fast tool for the diagnosis of DC.
Rochman et al., (2018) [46]	ELM	Low error rate = 0.0426	ELM is a powerful predictive tool.
Li et al., (2018) [47]	HMI and ELM	Sensitivities of incisors, canine, premolar, and molars were 78.25 ± 6.02%, 78.00 ± 5.99%, 79.25 ± 7.91%, and 78.75 ± 5.17%	Compared to the ANN approach, this method had a greater classification.
Lu et al., (2018) [48]	PCA and ELM	Accuracy = 79.75%	They are not able to detect the correct name for each landmark, especially for the teeth with similar teeth anatomy.
Li et al., (2018) [49]	GLCM, ELM	Sensitivity= 72%, specificity= 70%, accuracy= 71%	This method is more sensitive and accurate than the wavelet energy and naïve Bayes classifier.