|
Author and year | Study design | Groups | Application | Assessment method | Follow-up period | Outcome |
Study | Control |
|
Abdalla-Aslan et al. [5] 2020 | Cohort study | Machine learning computer vision algorithms | NA | OD | Automatic algorithm was used to detection and classification restoration while vector machine algorithm with error-correcting output codes was applied for cross-validation | NA | Machine learning demonstrated excellent performance in detecting and classifying dental restorations on panoramic images |
Bouchahma et al. [6] 2019 | Clinical trial | CNN | NM | OD and endodontics | Prediction of three types of treatments; fluoride, filling, and root canal treatments. The model was trained to learn on dataset of 200 X-ray images of patients’ teeth collected | NM | DL overall accuracy was 87%. The best prediction was the fluoride treatment with 98%, followed by RCT detection 88% and filling 77% |
Kuwada et al. [7] 2020 | Clinical trial | DetectNet, AlexNet, and VGG-16 | NM | OD | 400 images were randomly selected as training data, and 100 as validating and testing data. The remaining 50 images were used as new testing data. Recall, precision, and F-measure were used for detection of impacted teeth | NM | DetectNet and AlexNet appear to have potential use in classifying the presence of impacted supernumerary teeth in the maxillary incisor region on PR, while VGG-16 showed lower values |
Lee et al. [8] 2020 | Clinical trial | CNN on 20 automated 20 tooth segments | Oral radiologist manually performed individual tooth annotation on the PA | OD and forensic dentistry | 846 images with tooth annotations from 30 PA were used for training, and 20 as the validation and test sets. A fully deep learning method using the mask R-CNN model was implemented through a fine-tuning process to detect and localize the tooth structures | NM | It achieved high performance for automation of tooth segmentation on dental panoramic images. The proposed method might be applied in the first step of diagnosis automation and in forensic identification |
Ekert et al. [9] 2019 | Clinical trial | CNN to detect AL | Six independent examiners detect AL | Endodontics and OD | NN was trained and validated via 10 times repeated group shuffling. Results were compared with the majority vote of 6 examiners who detected ALs on an ordinal scale | NM | A moderately deep CNN showed satisfying discriminatory ability to detect ALs on panoramic radiographs |
Saghiri et al. [10] 2012 | Clinical trial | ANN | Endodontist’s opinion | Endodontics | Working length was determined and confirmed radiographically by endodontists and compared with ANN, and stereomicroscope as a gold standard after tooth extraction in cadaver | NM | ANN was more accurate than endodontists’ determinations when compared with measurements by using the stereomicroscope |
Arisu et al. [11] 2018 | Clinical trial | ANN | NM | Restorative dentistry | Obtained measurements and data were fed to an ANN to establish the correlation between the inputs; composite shade curing units and outputs; tooth number | NM | ANN showed that the light-curing units and composite parameter had the most significant effect on the bottom to top Vickers hardness ratio of the composites |
Yamaguchi et al. [12] 2019 | Clinical trial | 12 dislodge CAD/CAM composite resin crowns with DL | 12 trouble-free CAD/CAM composite resin crowns | Restorative dentistry | Convolution neural network (CNN) technique was used to predict debonding of composite crowns using 2D images captured from 3D stereolithography models | NM | Deep learning with CNN model showed good performance in terms of dislodgement predictability of composite crowns through 3D stereolithography models |
Otani et al. [13] 2015 | Experimental study | Ten veneer preparation with a robotic arm | Ten conventional veneers prepared by a clinician | Restorative dentistry | Accuracy and precision of veneer preparation were compared for all sites and separately for each tooth surface (facial, finish line, incisal) through 3D images and computation | NM | The robotic arm was able to prepare the tooth model as accurately as the control. However, a better finish line accuracy and precision was showed by the robotic arm |
Wang et al. [14] 2014 | Experimental study | Automatic laser ablation system for tooth crown preparation | NM | Prosthodontics | A layer-by-layer ablation method is developed to control the laser focus during the crown preparation | NM | The movement range and the resolution of the robotic system meet the satisfying requirements of typical dental operations for clinical crown preparation |
Takahashi et al. [15] 2020 | Experimental study | CNN | NM | Prosthodontics and OD | 1184 images of dental arches were classified into four arch types. A CNN method to classify images was developed using tensor flow and Kera’s deep learning libraries | NM | The results of this study suggest that dental arches can be classified and predicted using a CNN |
Patcas et al. [16] 2019 | Cohort study | CNN was applied in posttreatment photographs of 146 orthognathic patients | Pretreatment photographs of 146 patients | Orthodontics | CNN-based technique was used to compare facial attractiveness and apparent age of patients through pre- and posttreatment photographs | NA | Artificial intelligence can be used to detect facial attractiveness scores and apparent age in orthognathic surgery patients |
Li et al. [17] 2020 | Clinical trial | 50 oral images and 274 anterior through automated photo integrating system | Manual segmentation system | Esthetic dentistry | The facial and intraoral key points were detected by an automatic algorithm and compared with manual segmentation on standard photographs | NM | The proposed automated system can eliminate the need for dentists to employ a laborious image integration process and has potential for broad applicability in the field of esthetic dentistry |
Li et al. [44] 2015 | Experimental study | BPNN and GA neural network | Traditional neural network | Esthetic dentistry | The weighs and threshold values of GA and BPNN were compared for assistance in tooth color matching in dentistry | NM | GA and BP have practical application and can make teeth color matching objective and accurate |
Edinger [30] 2004 | Clinical trial | ROSY, a robot-like electronic simulator | NM | Prosthodontics | Accuracy of the simulator was measured for all directions in space by registering eccentric jaw positions on both sides of 10 subjects | NM | Its accuracy may render it suitable for clinical applications |
Meissner et al. [31] 2006 | Clinical trial | Automated smart ultrasonic calculus detection system | NM | Periodontics | The detection device is based on a conventional dental piezoelectric ultrasonic hand piece with a conventional scaler insert | NM | It was able to distinguish between different tooth surfaces in vitro independently from tip movements |
Meissner et al. [32] 2005 | Clinical trial | A novel calculus recognition device applied on 70 extracted teeth | NM | Periodontics | Impulse generator, coupled to a conventional piezo-driven ultrasonic scaler, sends signals to the cementum via the tip of an ultrasound device | NM | This system is able to function correctly, independent of the lateral forces and the tip angle of the instrument |
Devito et al. [33] 2008 | Clinical trial | Multilayer perceptron neural network | Twenty-five dental specialists with 20 years’ experience | OD | Evaluation of proximal caries on radiographic through ANN | NM | AI improves the radiographic diagnosis of proximal caries by 39.4% |
Kositbowornchai et al. [34] 2006 | Clinical trial | Learning vector quantization (LQV, NN) | NM | Restorative dentistry and OD | Tooth sections and microscopic examinations were used to confirm the actual dental caries status | NM | AI plays a useful and supporting in making dental caries diagnosis |
Patcas et al. [35] 2019 | Clinical trial | Ten images evaluated by CNN model | Ten images were analyzed by laypeople, orthodontists, and oral surgeon on a visual analogue scale | Orthodontics | Decision on profile and frontal images of cleft patients were compared between CNN technique and conventional rater group to evaluate facial attractiveness | NM | AI can be a helpful tool to describe facial attractiveness and overall analysis were comparable with the rater groups |
Lee et al. [36] 2018 | Clinical trial | CNN | Four calibrated board-certified dentists | OD and restorative dentistry | A pretrained GoogleNet Inception v3 CNN network was used for preprocessing and transfer learning | NM | CNN provides considerably good performance in detecting dental caries in PR |
Vranckx et al. [37] 2020 | Clinical trial | CNN and ResNet-101 | Manual measurements by 2 observers | OD | CNN and ResNet-101 jointly predicted the molar segmentation maps and an estimate of the orientation lines | NM | Fast, accurate, and consistent automated measurement of molar angulations on dental PR |
Lee et al. [38] 2020 | Clinical trial | Fifty cases of class2 TMJOA | Fifty cases of normal TMJ | OMFS | The condylar head was classified into 2 categories and tested by making 300 images | NM | AI can be used to support clinicians with diagnosis and decision making for treatments of TMJOA |
Hung et al. [39] 2019 | Clinical study | Machine learning method ANN was used on bitewing radiograph | Training group consisting of conventional radiograph analysis | Gerontology | Support vector machine (ANN) was used to detect root caries on radiograph by determining AUC | NM | Support vector machine showed 97.1% accuracy, 95.1% precision, 99.6% sensitivity, and 94.3% specificity for root caries detection |
Cui et al. [27] 2020 | Cohort study | CDS model applied to 3559 patient records | Two prosthodontists’ opinion | OMFS | CDS model was used to predict the outcome of teeth extraction through electronic dental records | NA | The machine learning CDS was an efficient tool to predict teeth extraction outcome |
Sornam and Prabhakaran [40] 2019 | Clinical study | LB-ABC with BPNN | BPNN classifier | Restorative dentistry | The BPNN classifier is compared with the LB-ABC-based BPNN classifier for dental caries classification | NM | The learning rate generated by the LB-ABC for the BPNN classifier achieved the best training and testing accuracy of 99.16% |
Setzer et al. [41] 2020 | Clinical study | Evaluation of periapical lesion by DL method | Rating by OMF radiologist, an endodontist, and a senior graduate student | Endodontics | The CBCT segmentation was assessed by DL, CNN detection | NM | DL algorithm trained in a limited CBCT environment showed excellent results in lesion detection accuracy |
Cantu et al. [42] 2020 | Clinical study | Caries detection on bitewing radiograph with DL | Opinion of four experienced dentists | OD, OR | CNN (U-Net) and Intersection-over-Union were used to detect caries on radiographs | NM | The deep neural network was accurate than dentists |
Aliaga et al. [45] 2020 | Experimental study | Automatic computation and intelligent image segmentation of 370 radiographs | Expert dentist opinion | OD, OMFS | Automatic computation for analysis of mandibular indices and osteoporosis detection | NM | Automatic computation of mandibular indices and intelligent image segmentation was an efficient and reliable approach for early osteoporosis detection |
Kim et al. [28] 2018 | Case-control study | Machine learning prediction models for BRONJ after extraction in 125 patients with drug use | Conventional methods, serum CTX level | OMFS/OM | Five machine learning methods such as logistic regression model, decision tree, support vector machine, ANN, and random forest were applied to predict BRONJ at extraction sites | NA | Machine learning showed superior performance in predicting BRONJ compared with serum CTX level and drug holiday period |
Dumast et al. [29] 2018 | Case-control study | 17 tested OA subjects evaluated with deep CNN on 3D images | 17 age and sex-matched control subjects without OA | OMFS | Deep neural network classifier of 3D condylar morphology (ShapeVariationAnalyzer, SVA), and a flexible web-based system for data storage, computation and integration (DSCI) of high dimensional imaging, clinical, and biological data | NA | Deep neural network is a useful tool for classification of TMJOA |
Sorkhabi and Khajeh [43] 2019 | Clinical trial | 3D deep CNN and CBCT | Postextraction clinical parameter measurements | OD and implant dentistry | 3D CNN method was used to measure alveolar bone density on CBCT images | 6 months | 3D deep CNN technique can accurately classify alveolar bone. Pattern, which is helpful in dental implant placement and diagnosis |
|