Table 4.

Characteristics of the studies.

Authors/Years	Type of Study	Type of AI	Materials and Methods	Results
Taylor Mason et al., 2023 [63]	Retrospective observational study	ML (LR, RF, SVMs, ANN)	393 patients, a diverse population. Trained LR, RF, SVM, and ANN on 70% of data, and tested on 30%. Evaluated accuracy and precision for extraction decisions.	High accuracy in predicting tooth extraction decisions.
Etemad et al., 2021 [64]	Retrospective observational study	ANN, RF	838 orthodontic patient records. Split into extraction and non-extraction samples. Used 117 clinical and cephalometric variables for ML (RF and MLP) for tooth extraction prediction.	High accuracy in predicting tooth extraction therapy.
Lee et al., 2022 [66]	Retrospective observational study	ML (RF, LR)	196 skeletal class III patients, 136 training, 60 tests. Estimated neural network success rate. Binary classifier for surgical case prediction.	AI is useful for successfully classifying patients up to 90% of candidates for surgery.
Chaiprasittikul et al., 2023 [67]	Retrospective observational study	ANN	Analysis of 538 cephalometric radiographs using Detectron2 and ANN. Developed neural network decision support system for orthognathic surgery prediction.	AI is useful for successfully classifying up to 90% of candidates for surgery.
Prasad et al., 2022 [62]	Retrospective observational study	ML (extreme gradient boosting, RF, decision tree)	Analyzed 700 orthodontic cases with 33 inputs and 11 outputs. Developed ML models and compared their predictions with expert orthodontist decisions.	The overall accuracy of the models was 84%.
Jung et al., 2016 [65]	Retrospective observational study	ANN	Analyzed 156 patients with 12 cephalometric variables, 6 indexes, and 3-bit extraction pattern diagnosis. Created and evaluated ANN.	Effectiveness in assisting professionals in decision-making with success rates of 84–93%.