Lee J et al. (2018) [27] |
CNN |
Dental caries |
Periapical radiographs |
600 |
Mean AUC—0.890 |
4 Dentists |
Deep CNN showed a considerably good performance in detecting dental caries in periapical radiographs. |
Casalegno et al. (2019) [29] |
CNN |
Dental caries |
Near-infrared transillumination imaging |
217 |
ROC of 83.6% for occlusal caries; ROC of 84.6% for proximal caries |
Dentists with clinical experience |
CNN showed increased speed and accuracy in detecting dental caries |
Cantu et al. (2019) [30] |
CNN |
Dental caries |
Bitewing radiographs |
141 |
Accuracy 0.80; sensitivity 0.75%; specificity 0.83%; |
4 experienced dentists |
AI model was more accurate than dentists |
Radke et al. (2003) [45] |
ANN |
Disk displacement |
Frontal plane jaw recordings from chewing |
68 |
Accuracy 86.8%, specificity 100%, sensitivity 91.8% |
None |
The proposed model has an acceptable level of error and an excellent cost/benefit ratio. |
Park YH et al. (2021) [31] |
ML |
Early childhood caries |
Demographic details, oral hygiene management details, maternal details |
4195 |
AUROC between 0.774 and 0.785 |
Traditional regression model |
Both ML-based and traditional regression models showed favorable performance and can be used as a supporting tool. |
Kuwana et al. (2021) [39] |
CNN |
Maxillary sinus lesions |
Panoramic radiographs |
1174 |
Diagnostic accuracy, sensitivity, and specificity were 90–91%, 81–85% and 91–96% for maxillary sinusitis and 97–100%, 80–100% and 100% for maxillary sinus cysts. |
None |
The proposed deep learning model can be reliably used for detecting the maxillary sinuses and identifying lesions in them. |
Murata et al. (2018) [37] |
CNN |
Maxillary sinusitis |
Panoramic radiographs |
120 |
Accuracy 87.5%; sensitivity 86.7%; specificity 88.3% |
2 experienced radiologists, 2 dental residents |
The AI model can be a supporting tool for inexperienced dentists |
Kim et al. (2019) [38] |
CNN |
Maxillary sinusitis |
Water’s view radiographs |
200 |
AUC of 0.93 for temporal; AUC of 0.88 for geographic external |
5 radiologists |
the AI-based model showed statistically higher performance than radiologists. |
Hung KF et al. (2022) [40] |
CNN |
maxillary sinusitis |
Cone-beam computed tomography |
890 |
AUC for detection of mucosal thickening and mucous retention cyst was 0.91 and 0.84 in low dose, and 0.89 and 0.93 for high dose |
None |
The proposed model can accurately detect mucosal thickening and mucous retention cysts in both low and high-dose protocol CBCT scans. |
Danks et al. (2021) [34] |
DNN symmetric hourglass architecture |
Periodontal bone loss |
Periapical radiographs |
340 |
Percentage Correct Keypoints of 83.3% across all root morphologies |
Asymmetric hourglass architecture, Resnet |
The proposed system showed promising capability in localizing landmarks and periodontal bone loss and performed 1.7% better than the next best architecture. |
Chang et al. (2020) [36] |
CNN |
Periodontal bone loss |
Panoramic radiographs |
340 |
Pixel accuracy of 0.93; Jaccard index of 0.92; dice coefficient values of 0.88 for localization of periodontal bone. |
None |
The proposed model showed high accuracy and excellent reliability in the detection of periodontal bone loss and classification of periodontitis |
Ozden et al. (2015) [18] |
ANN |
Periodontal disease |
Risk factors, periodontal data, and radiographic bone loss |
150 |
Performance of SVM & DT was 98%; ANN was 46% |
SVM &DT |
SVM and DT showed good performance in the classification of periodontal disease while ANN had the worst performance |
Devito et al. (2008) [26] |
ANN |
Proximal caries |
Bitewing radiograph |
160 |
ROC curve area of 0.884 |
25 examiners |
ANN could improve the performance of diagnosing proximal caries. |
Dar-Odeh et al. (2010) [20] |
ANN |
Recurrent aphthous ulcers |
Predisposing factor and RAU status |
96 |
Accuracy of prediction for network 3 & 8 is 90%; 4,6 & 9 is 80%; 1& 7 is 70%; 2 & 5 is 60% |
None |
the ANN model seemed to use gender, hematologic and mycologic data, tooth brushing, fruit, and vegetable consumption for the prediction of RAU. |
Hung M et al. (2019) [28] |
CNN |
Root caries |
Data set |
5135 |
Accuracy 97.1%; Precision 95.1%; sensitivity 99.6%; specificity 94.3% |
Trained medical personnel |
Shows good performance and can be clinically implemented. |
Iwasaki et al. (2015) [47] |
BBN |
Temporomandibular disorders |
Magnetic resonance imaging |
590 |
Of the 11 BBN algorithms used path conditions using resubstitution validation and 10—fold cross-validation showed an accuracy of >99% |
necessary path condition, path condition, greedy search-and-score with Bayesian information criterion, Chow-Liu tree, Rebane-Pearl poly tree, tree augmented naïve Bayes model, maximum log-likelihood, Akaike information criterion, minimum description length, K2 and C4.5 |
The proposed model can be used to predict the prognosis of TMDs. |
Orhan et al. (2021) [49] |
ML |
Temporomandibular disorders |
Magnetic resonance imaging |
214 |
The performance accuracy for condylar changes and disk displacement are 0.77 and 0.74 |
logistic regression (LR), random forest (RF), decision tree (DT), k-nearest neighbors (KNN), XGBoost, and support vector machine (SVM) |
The proposed model using KNN and RF was found to be optimal for predicting TMJ pathologies |
Diniz de lima et al. (2021) [50] |
ML |
Temporomandibular disorders |
Infrared thermography |
74 |
Semantic and radiomic-semantic associated ML feature extraction methods and MLP classifier showed statistically good performance in detecting TMDs |
KNN, SVM, MLP |
ML model associated with infrared thermography can be used for the detection of TMJ pathologies |
Bas B et al. (2012) [46] |
ANN |
TMJ internal derangements |
Clinical symptoms and diagnoses |
219 |
Sensitivity and specificity for unilateral and anterior disk displacement with and without reduction were 80% & 95% and 69% & 91%; for bilateral and anterior disk displacement with and without reduction were 37% &100% and 100% & 89% respectively. |
Experienced surgeon |
The developed model can be used as a supportive diagnostic tool for the diagnoses of subtypes of TMJ internal derangements |
Choi et al. (2021) [48] |
CNN |
TMJ osteoarthritis |
Panoramic radiographs |
1189 |
Accuracy of 0.78, the sensitivity of 0.73, and specificity of 0.82 |
Oral and maxillofacial radiologist |
The developed model showed performance equivalent to experts and can be used in general practices where OMFR experts or CT is n |
Fukuda et al. (2019) [33] |
CNN |
Vertical root fracture |
Panoramic radiograph |
60 |
The precision of 0.93; Recall of 0.75 |
2 Radiologists and 1 Endodontist |
The CNN model was a promising supportive tool for the detection of vertical root fracture. |