Table 5. The table shows relevant review findings of deep learning algorithms for different imaging modalities.
| Authors | Deep learning architectures | Detection/Application | Metrics |
|---|---|---|---|
| Imaging modality: Periapical X-rays | |||
| (Prajapati, Nagaraj & Mitra, 2017) | CNN and transfer learning | Dental caries, periapical infection, and periodontitis | Accuracy:- 0.8846 |
| (Yang et al., 2018) | Conventional CNN | Automated clinical diagnosis | F1 score 0.749 |
| (Zhang et al., 2018) | CNN (label tree with cascade network structure) | Teeth detection & classification | Precision:- 0.958, Recall:- 0.961 F-score :- 0.959 |
| (Choi, Eun & Kim, 2018) | Conventional CNN | Caries detection | F1max:- 0.74 with FPs:- 0.88 |
| (Lee et al., 2018b) | GoogLeNet Inception v3 CNN network | Caries and Non-caries | Premolar accuracy (premolar):- 0.89, Accuracy (molar):- 0.88, and Accuracy:- 0.82, AUC (premolar):- 0.917, AUC (molar):- 0.890, and an AUC (Both premolar and molar):- 0.845 |
| (Lee et al., 2018a) | CNN (VGG-19) | Periodontally compromised teeth (PCT) | For premolars, the total diagnostic Accuracy(premolars):- 0.810, Accuracy(molars):- 76.7% |
| (Geetha, Aprameya & Hinduja, 2020) | Back‑propagation neural network | Caries detection | Accuracy:- 0.971, FPR:- 0.028, ROC :- 0.987, PRC :- 0.987 with learning rate:- 0.4, momentum:- 0.2 |
| Imaging modality: Panoramic X-rays | |||
| (Oktay, 2017) | AlexNet | Teeth detection and classification | Accuracy (tooth detection):- 0.90 Classification accuracy: Molar :-0.9432, Premolar:- 0.9174, Canine & Incissor:- 0.9247 |
| (Chu et al., 2018) | Deep octuplet Siamese network (OSN) | Osteoporosis analysis | Accuracy:- 0.898 |
| (Wirtz, Mirashi & Wesarg, 2018) | Coupled shape model + neural network | Teeth detection | Precision:- 0.790, Recall:- 0.827 Dice coefficient:- 0.744 |
| (Jader et al., 2018) | Mask R-CNN model | Teeth detection | Accuracy:- 0.98, F1-score:- 0.88, precision:- 0.94, Recall:- 0.84, and Specificity:- 0.99 |
| (Lee et al., 2019) | Mask R-CNN model | Teeth segmentation for diagnosis and forensic identification | F1 score:- 0.875, Precision:- 0.858, Recall:- 0.893, Mean‘IoU’:- 0.877 |
| (Kim et al., 2019) | DeNTNet (deep neural transfer Network) | Bone loss detection | F1 score:- 0.75, Accuracy:- 0.69. |
| (Tuzoff et al., 2019) | R-CNN | Teeth detection and numbering | Tooth detection (Precision:- 0.9945 Sensitivity:- 0.9941) Tooth numbering (Specificity:- 0.9994, Sensitivity = 0.9800) |
| (Fukuda et al., 2019) | DetectNet with DIGITS version 5.0 | Vertical root fracture | Recall:- 0.75, Precision:- 0.93 F-measure:- 0.83 |
| (Murata et al., 2019) | AlexNet | Maxillary sinusitis | Accuracy:- 0.875, Sensitivity:- 0.867, Specificity:- 0.883, and AUC:- 0.875. |
| (Kats et al., 2019) | ResNet-101 | Plaque detection | Sensitivity:- 0.75, Specificity:- 0.80, Accuracy:- 0.83, AUC:- 0.5 |
| (Singh & Sehgal, 2020) | 6-Layer DCNN | Classification of molar, premolar, canine and incisor | Accuracy (augmented database):- 0.95, Accuracy (original database):- 0.92 |
| (Muramatsu et al., 2020) | CNN (Resnet 50) | Teeth detection and classification | Tooth detection sensitivity:- 0.964 Average classification accuracy (single model):- 0.872, (multisized models):- 0.932 |
| (Banar et al., 2020) | Conventional CNN | Teeth detection | Dice score:- 0.93, accuracy:- 0.54, a MAE:- 0.69, and a linear weighted Cohen’s kappa coefficient:- 0.79. |
| Imaging modality: Bitewing X-rays | |||
| (Srivastava et al., 2017) | Fully convolutional neural network FCNN | Detection of dental caries | Recall:- 80.5, Precision:- 61.5, F-score:- 70.0 |
| Imaging modality: CT & CBCT | |||
| (Miki et al., 2017a) | AlexNet architecture | A total of seven-Tooth-type classification (canine, molar, premolar, etc.) | Accuracy:- 0.91 |
| (Miki et al., 2017b) | AlexNet | Teeth detection and classification | Detection accuracy:- 0.774, Classification accuracy:- 0.771 |
| (Hatvani et al., 2018) | Subpixel network + U-Net architecture | Teeth resolution enhancement | Mean of difference (area mm2):- 0.0327 Mean of difference(micrometer):- 114.26 Dice coefficient:- 0.9101 |
| (Torosdagli et al., 2018) | CNN (a long short-term memory (LSTM) network) | Anatomical Landmarking | DSC:- 0.9382 |
| (Egger et al., 2018) | CNN (VGG16, FCN) | Mandible detection | Accuracy:- 0.9877, Dice coefficient:- 0.8964 and Standard deviation:- 0.0169 |
| (Hiraiwa et al., 2019) | AlexNet and GoogleNet | Classification of root morphology (Single or extra) | Diagnostic accuracy:- 0.869 |
| Imaging modality: Hybrid dataset | |||
| (Wang et al., 2016) | U-net architecture (Ronneberger, Fischer & Brox, 2015) | Landmark detection in cephalometric radiographs and Dental structure in bitewing radiographs. | F-score => 0.7 |
| (Lee, Park & Kim, 2017) | LightNet and MatConvNet | Landmark detection | N.A |
| (Karimian et al., 2018) | Conventional CNN | Caries detection | Sensitivity:- 97.93%~99.85% Specificity:- 100% |
| Imaging modality: Color images/Oral images | |||
| (Rana et al., 2017) | Conventional CNN | Detection of inflamed and healthy gingiva | precision:- 0.347, Recall: 0.621, AUC:- 0.746 |
| Image type not defined | |||
| (Imangaliyev et al., 2016) | Conventional CNN | Dental plaque | F1-score:- 0.75 |