Authors | Country | Specialty | Study design | Study theme | Dataset Source | Datatype | No. of annotators | Annotators specialty | Sample size | Task | Types of Neural networks/ Models/algorithms | Performance metrics | Maturity | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
1 | Talaat et al | Egypt [12] | Orthodontics | Validation study | Classification | Local | Intraoral photographs | 2 | Both orthodontists | 700 | Malocclusion detection | CNN (YOLO) | Accuracy: 99.99 Preceison:0.99 Recall:1 F1 score:1 | 2 |
2 | Khazaei et al | Iran [13] | Orthodontics | Validation Study | Classification | Local | Lateral Cephalograms | N/A As sex was input from patient records | NM | 1476 | Gender determination | CNN, Dense Net, ResNet, VGG | DenseNet Accuracy 0.9 Resnet Accuracy 0.62 VGG 16 Accuracy 0.75 | 2 |
3 | Ehtesham et al | Iran [14] | Oral Medicine | Pilot study | Decision Support System | Local | Health records | N/A As the cases and symptoms were retrieved from medical records | NM | 500 training cases, 39 testing cases | Differential diagnosis | ML (Nearest neighbor algorithm) | Metrics were not mentioned explicitly but out of 39 diagnosis there were 9 misdiagnosis | 4 |
4 | Koopaie et al | Iran [15] | Restorative Dentistry | Case-control study | Classification | Local | Health records + cystatin S levels of saliva | 3 | Dentist | 20 | Caries detection | ML, ANN, XGBoost Random Forest SVM | ANN: Acc 90%, Sen 100%, Spec 72%. XGboost: Acc 89%, Sen 93%, Spec 84%. Random Forest: Acc 85%, Sen 93%, Spec 76%. SVM: Acc 85%, Sen 86%, Spec 84%. | 2 |
5 | Adnan et al | Pakistan [1] | Endodontics | Validation study | Instance Segmentation | Local | OPGs | 2 | Endodontist | 40 OPG, 1280 teeth crops | Teeth detection and numbering | CNN | 88.2 | 2 |
6 | Mariam et al | Pakistan [16] | Oral Pathology | Validation study | Diagnostic test accuracy | Local | Histology patches | 2 | Pathologist | 280 | Gaze pattern for annotation (Keratin pearl) | Faster R-CNN with Inceptionv2 architecture [2] YOLOv3 with DarkNet53 architecture [3] YOLOv5 with DarkNet53 architecture. | 85% less time was consumed in gaze annotations | 3 |
7 | Fatima et al | Pakistan [17] | Endodontics | Validation study | Classification and Decision Support System | Local and International | Periapical x rays | 2 | Radiologist and Dentist | 516 | Lesion detection Length measurement/ apical foramen Root morphology Root Fracture Case difficulty | CNN Mask RCNN with back bone Resnet 50 Resnet 101 Mobile Net V2 Proposed Backbone Network | Resnet 50: Pre.82, Rec.83, F1.78 Resnet 101: Pre .81, Rec .84,F1 .74 Mobile net V2: Pre.86, Rec .87,F1 .84 Proposed backbone network: Pre 0.86, Rec 0.89, F1 0.89 | 2 |
8 | Bharathi et al | India [18] | Oral Pathology | Experimental study (why) | Classification | Others (Internet sources, mobile cameras) | Focused photographs | NM | NM | 500 total, 140 diabetic | Diabetes tongue | CNN | 83 | 2 |
9 | Patil e tal | India [19] | Oral and Maxillofacial Surgery | Validation study | Classification | Local | OPGs | 1 | Oral and Maxillofacial surgeon | 509 | Gender determination (Mandible) | ANN and Regression | The discriminant analysis had an overall accuracy of 69.1%, logistic regression showed an accuracy of 69.9%, and ANN exhibited a higher accuracy of 75%. | 2 |
10 | Prasad et al | India [20] | Orthodontics | Validation study | Decision Support System | Local | OPGs, Dental cast, dental clinical data | 10–15 | Orthodontists | 700 training set: test set 70:30 10–15 | Treatment planning (Maxilla, mandible, facial landmarks, age, dentition) | ML predictive models: XGB Classifier Random Forest Classifier Linear SVM Decision Tree Classifier Logistic Regression K-Neighbor Classifier Naive Bayes Classifier | The model showed an overall average accuracy of 84%, with the Decision Tree, Random Forest and XGB classifier algorithms showing the highest accuracy ranging from 87 to 93%. | 2 |
11 | Katyal et al | India [21] | Orthodontics | Validation study | Decision Support System | Local | OPGs | NM | WebCeph | 25 | Cephalometric analysis (Maxilla, mandible, facial landmarks, dentition) | CNN | No statistically sig difference between manual tracing, digital tracing and WebCeph, WebCeph was more efficient | 3 |
12 | Sherly et al | India [22] | Restorative dentistry | Validation study | Classification | NM | OPGs, Clinical pictures | NM | NM | NM | Caries detection | DL (MDP (maximum directional pattern) using CNN | 98.6% | 2 |
13 | Benakatti et al | India [23] | Oral and Maxillofacial surgery | Validation study | Identification | Local | OPGs | NM | NM | NM (only mentioned 80% of data set for training and 20% for testing) | Implant system identification | Support vector machine (SVM), K-Nearest Neighbor (KNN), XG boost, and Logistic Regression Classifiers | Best accuracy: Logistic Regression, Overall 0.67 | 2 |
14 | India [24] | Orthodontics | Validation study | Decision Support System | Local | Dental clinical data | 1 | NM | 1000 | Influence of anatomical risk factors in determining number of traumatized teeth per affected individual | MLP model of ANN | Overbite (100%) as the strongest risk factor for TDI in number of teeth of affected individual | 2 | |
15 | Yadalam et al | India [25] | Oral and Maxillofacial surgery | Validation study | Decision Support System | Local | Dental clinical data | NM | NM | 1032,825 (80%) training set, 207 (20%) data authentication, 45 controls | Prediction of postoperative pain after implant surgery (VAS) | ML (Multiple Linear Regression MLR) model | The data set’s output is evaluated using a confusion matrix: the higher the confusion matrix’s diagonal values, the more correct the predictions. The training sample accuracy and the validation sample | 2 |
16 | Shwewood et al | India [26] | Endodontics | Validation | Identification and classification | Local | CBCT | 2 | Endodontists | 135, 100 for training and 35 for testing | identification of C-shaped root canal anatomy | DL/CNN (U-Net, Residual U-Net and Xception U-Net) | Xception U-Net with CLAHE provided the best results the mean sensitivity values were highest for Xception U-Net (0.786 ± 0.0378, diagnostic test grade: 78.6% [good]) followed by residual U-Net (mean sensitivity 5 0.746 ± 0.0391, diagnostic test grade: 74.6% [good]) and U-Net (0.720 ± 0.0495, diagnostic test grade: 72.0% [good]). The mean positive predictive | 2 |
17 | Mallishery et al | India [27] | Endodontics | Validation study | Decision Support System | Local | Radiovisographs, AAE Case difficulty Assessment forms | 2 | Endodontists | 500 | Difficulty of endodontic case | ML (support vector machine (SVM), and deep neural network (DNN)) | 94.96% | 2 |
18 | Moidu et al | India [28] | Endodontics | Validation study | Classification | Local | Periapical x-ray | 3 | Endodontists | 3000 periapical root areas (PRA) on 1950 digital IOPA 450 PRA for model validation 540 PRA on 250 IOPA for testing performance | Periapical lesion | DL (CNN YOLO.v3) | Across all PAI score, 303 PRA (56.11%) exhibited true prediction. True prediction: PAI score 1 90.9% PAI scores 2 and 5: 30% PAI scores 3: 60% PAI score 4: 71% False Prediction: PAI score 2 was interpreted as PAI score 3 (32%) and PAI score 1 (25%). PAI score 5 was interpreted as PAI score 4 (35.8%) and PAI score 3 (20.8%) PAI scores were dichotomized as healthy (PAI scores 1 and 2) and diseased (PAI scores 3, 4, and 5), the model achieved a true prediction of 76.6 and 92%, respectively The CNN model achieved a 92.1% sensitivity, 76% specificity, 86.4% PPV and 86.1% NPV Accuracy:86.3% F1 score: 0.89 Matthews correlation coefficient: 0.71 | 2 |
19 | Ghosh et al | India [29] | Restorative dentistry | Randomized parallel group study | Dental practice management | Local | Others (Scientific papers) | NM | NM | 250 | Improving patient recall rate (Patient recall) | topic modeling using Labeled LDA (Latent Dirichlet Allocation) | improved the patient recall rate from 21.1 to 37.8% (p-value = 0.024). | 4 |
20 | Fidya et al | Indonesia [30] | Orthodontics | Validation study | Classification | NM | Casts | NM | Health records | 150 | Gender Determination | ML Naive Bayes Decision tree MLP | The accuracy rate of the Naive Bayes method was 82%, while that of the decision tree and MLP amounted to 84%. | 2 |
21 | Widyaningrum et al | Indonesia [31] | Periodontics | Validation study | Classification | Local | OPGs | 2 | Periodontist and General Dentist | 1100 | Periodontitis | CNN U-Net, RCNN | U-Net showed the characteristic of semantic segmentation, and Mask R-CNN performed instance segmentation with accuracy, precision, recall, and F1-score values of 95, 85.6, 88.2%, and 86.6%, | 2 |
22 | Mahto et al | Nepal [32] | Orthodontics | Validation study | Decision Support System | local | Lateral Cephalograms | NM | WebCeph | 30 | compare the linear and angular cephalometric measurements obtained from web-based fully automated Artificial Intelligence (AI) driven platform “WebCeph”™ with that from manual tracing | CNN | overall Intraclass correlation coefficient for manual and WebCeph tracing was > 0.9 | 3 |
23 | Thahn et al | Vietnam [33] | Restorative dentistry | Validation study | Classification | Local | Dental clinical data | 1 | Dentist | 1902 | Dental caries | CNN Faster Region-Based Convolutional Neural Networks (Faster R-CNNs), You Only Look Once version 3 (YOLOv3), RetinaNet, and Single-Shot Multi-Box Detector (SSD) | For cavitated caries: YOLOv3:87.4% and Faster R-CNN: 71.4% For visually non-cavitated: YOLOv3: 36.9% and Faster R-CNN: 26% Specificity of all 4 models for cavitated caries: 86% Sensitivity of all 4 models for non-cavitated caries: 71% | 2 |
24 | Tuan et al | Vietnam [34] | Oral Radiology | Validation study | Segmentation | NM | OPGs | NM | NM | 56 | new semi-supervised fuzzy clustering algorithm named as SSFC-FS based on Interactive Fuzzy Satisficing for the dental X-ray image segmentation problem. | ML | Not mentioned | 1 |
25 | Ngoc et almachine | Vietnam [35] | Endodontics | Validation study | Classification | NM | Bitewing | 2 | Dentist | 130 | Periapical lesion diagnosis | CNN Faster R-CNN | Specificity: 97.9% Sensitivity: 89.5% Accuracy: 95.6% | 2 |