Table 7.
Summary of related studies for AI application in oral pathology.
| 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 |
|---|---|---|---|---|---|---|---|---|---|
| Disease diagnosis |
Orhan et al., 2021 [36] | ML (KNN and Random Forest (RF)) | Magnetic Resonance Imaging | Temporomandibular disorders | Pathologists | NA | NA | Accuracy: Condylar changes—0.77, Disk displacement—0.74 | The model was found to be optimal in predicting temporomandibular disorders |
| Hung et al., 2022 [55] | Three step CNN based on V-Net and SVR | CBCT images | 445 images, Train: 311 images, Valid: 62 images, Test: 249 images 311/62/249 | Maxillary sinusitis | NA | NA | AUC: Mucosal thickening—0.91, Mucous retention cyst—0.84 | The model helped detect and segment mucosal thickening and mucosal retention cyst using low-dosed CBCT scans | |
| Kuwana et al., 2021 [56] | CNN (DetectNet) | Panoramic radiographs | 1174 images | Maxillary sinus lesions | NA | NA | Maxillary sinusitis: Accuracy—0.90–0.91, Sensitivity—0.81–0.85, Specificity—0.91–0.96, Maxillary sinus cysts: Accuracy—0.97–1.00, Sensitivity—0.80–1.00, Specificity—1.00 | The model was helpful in detecting maxillary sinus lesions | |
| Choi et al., 2021 [58] | CNN (ResNet) | Panoramic radiographs | 1,189 images, Training: 951 images, Testing: 238 images 951/238 | Temporomandibular joint disorders (TMJ) osteoarthritis | Oral and maxillofacial radiologist (OMFR) | Five-fold | Temporal: AUC - 0.93, Geographical external: AUC—0.88 | The model achieved significantly higher diagnostic performance compared to that of radiologists | |
| Kim et al., 2019 [42] | CNN | Water’s view radiographs | 200 images | Maxillary sinusitis | Five radiologists | NA | Accuracy: Upper dental model: 0.99, Lower dental model: 0.987 | The label free mesh simplification method was helpful in preserving the teeth boundary information using 3D dental model | |
| Murata et al., 2019 [57] | AlexNet CNN | Panoramic radiographs | 120 images | Maxillary sinusitis | Two radiologists, Two dentists | NA | Accuracy: 0.875, Sensitivity: 0.867, Specificity: 0.883 | The model shows diagnostic performance similar to the radiologists and superior to the resident dentists | |
| Jeyaraj et al., 2019 [59] | Partitioned CNN (GoogleNet Inception V3) | Hyperspectral images | 600 images | Oral Cancer | Expert oncologist | Seven-fold | Benign tissue Accuracy—0.914, Malign tissue Accuracy—0.945 | The model helped predict cancerous or benign tumor and has the potential to be applied as a workbench for automated classification | |
| Disease prognosis |
Iwasaki et al. 2015 [27] | Bayesian Belief Network (BNN) | Magnetic Resonance Imaging | 590 images | Temporomandibular joint disorders (TMJ) | NA | Ten-Fold | Accuracy: 0.99 | The model has the potential to determine the progression of TMD in terms of bone changes, disc displacement and bony space and disc affect with encouraging diagnostic performance |
| Bas et al., 2012 [28] | Back Propagation ANN | Electronic records | 219 records | Clinical symptoms (Temporomandibular joint disorders (TMJ)) | Experienced oral and maxillofacial surgeon | NA | Unilateral with and without reduction: Sensitivity—0.80 & 0.95, Specificity—0.69 & 0.91 | The model was helpful in diagnosing the preliminary subtypes of TMJ and can be useful in the decision-making process |