Table 6.
Summary of related studies for AI application in prosthetic/restorative dentistry.
| AI Application | Author, Year (Ref) | Architecture | Data Modality | Dataset Size Split (Train/Val/Test or Train/Test) | Study Factor | Reference Standard (Ground Truth) | Validation Scheme | Results (Performance Metrics/Values) | Conclusion |
|---|---|---|---|---|---|---|---|---|---|
| Treatment planning | Lee et al., 2022 [39] | Decision Tree | Electronic records | 94 clinical cases | Tooth prognosis | Three prosthodontists | NA | Accuracy: 0.841 | The model was helpful in determining tooth prognosis for effective treatment planning |
| Disease detection | Abdalla-Aslan et al., 2020 [40] | Cubic SVM based algorithm | Panoramic radiographs | 83 images | Dental restorations | Experienced practitioners | Five-fold | For detection Sensitivity: 0.94, For classification Sensitivity: 0.98 | The model has the potential to detect and classify dental restorations to promote patient’s health |
| Disease diagnosis | Lee et al. 2020 [54] | Fine Tuned and pretrained Deep CNN | Panoramic and periapical radiographs | 10,770 images, Train: 6462 images, Valid: 2154 images, Test: 2154 images 6462/2154/2154 | Dental implants | Periodontists | Ten-fold | AUC:0.971 | The model was helpful in the identification and classification of dental implants with performance similar to that of periodontist |
| Takahashi et al., 2021 [71] | Deep CNN (ResNet152) pretrained with ImageNet weights | Oral photographs | 1,184 images, maxilla: 748 images, mandible: 436 images | Partially edentulous arches | Clincian | NA | Maxilla: Accuracy: 0.995, Recall: 1.00, Precision: 0.25, AUC: 0.99, Mandible Accuracy: 0.997, Precision: 0.25, Recall: 1.00, AUC: 0.98 | The method was helpful in classification of dental arches and can be effective in designing removable partial dentures | |
| Disease segmentation | Xu et al., 2018 [72] | Two Hierarchical CNNs | 3D dental images | 1200 images, Train: 1000, Valid: 50, Testing: 150 1000/50/150 | Preserve teeth boundary | NA | NA | Accuracy: Upper dental model: 0.99, Lower dental model: 0.987 | The label-free mesh simplification method helped preserve the teeth boundary information using the 3D dental model. |
| Treatment Planningand Prognosis |
Cui et al., 2020 [25] | Triple Classification algorithm (Extreme Gradient Boost (XGBoost)) | Electronic health records | 4135 records | Tooth extraction therapy | Two prosthodontists | Five-fold | Binary Classification Accuracy: 0.962, Precision: 0.865, Recall: 0.830, Triple Classification Accuracy: 0.924, Precision: 0.879, Recall: 0.836 | The model was helpful in predicting tooth extracting therapy with performance superior to that of prosthodontists |
| Javed et al. 2020 [26] | ANN | Electronic records | 45 records of children | Occlusal caries lesions | NA | Leave one out | Regression co-efficient: 0.99 | The model was helpful in occlusal dentinal caries lesions and the study proposes an iOS app for meticulous prediction of caries |