Periapical Radiographic Images Research Articles

Deep-learning performance in identifying and classifying dental implant systems from dental imaging: a systematic review and meta-analysis.

Deep learning (DL) offers promising performance in computer vision tasks and is highly suitable for dental image recognition and analysis. We evaluated the accuracy of DL algorithms in identifying and classifying dental implant systems (DISs) using dental imaging. In this systematic review and meta-analysis, we explored the MEDLINE/PubMed, Scopus, Embase, and Google Scholar databases and identified studies published between January 2011 and March 2022. Studies conducted on DL approaches for DIS identification or classification were included, and the accuracy of the DL models was evaluated using panoramic and periapical radiographic images. The quality of the selected studies was assessed using QUADAS-2. This review was registered with PROSPERO (CRDCRD42022309624). From 1,293 identified records, 9 studies were included in this systematic review and meta-analysis. The DL-based implant classification accuracy was no less than 70.75% (95% confidence interval [CI], 65.6%-75.9%) and no higher than 98.19 (95% CI, 97.8%-98.5%). The weighted accuracy was calculated, and the pooled sample size was 46,645, with an overall accuracy of 92.16% (95% CI, 90.8%-93.5%). The risk of bias and applicability concerns were judged as high for most studies, mainly regarding data selection and reference standards. DL models showed high accuracy in identifying and classifying DISs using panoramic and periapical radiographic images. Therefore, DL models are promising prospects for use as decision aids and decision-making tools; however, there are limitations with respect to their application in actual clinical practice.

Establishing a novel deep learning model for detecting peri-implantitis

Background/purpose: The diagnosis of peri-implantitis using periapical radiographs is crucial. Recently, artificial intelligence may apply in radiographic image analysis effectively. The aim of this study was to differentiate the degree of marginal bone loss of an implant, and also to classify the severity of peri-implantitis using a deep learning model. Materials and methods: A dataset of 800 periapical radiographic images were divided into training (n = 600), validation (n = 100), and test (n = 100) datasets with implants used for deep learning. An object detection algorithm (YOLOv7) was used to identify peri-implantitis. The classification performance of this model was evaluated using metrics, including the specificity, precision, recall, and F1 score. Results: Considering the classification performance, the specificity was 100%, precision was 100%, recall was 94.44%, and F1 score was 97.10%. Conclusion: Results of this study suggested that implants can be identified from periapical radiographic images using deep learning-based object detection. This identification system could help dentists and patients suffering from implant problems. However, more images of other implant systems are needed to increase the learning performance to apply this system in clinical practice.

Role of endodontic treatment related signs for the forensic odontology practice considering real-life quality of comparable radiographs

Background. Real life quality of periapical radiographic images widely varies, while also periapical radiographs not always have been taken at the same angulation, especially in multirooted tooth cases, which in turn potentially may limit the value of obtained information received during endodontic treatment for the further comparative dental identification.
 Objective. To objectify possibility of using real-life quality periapical radiographs obtained during endodontic treatment for the comparative dental identification.
 Materials and Methods. Research was organized in the form of cross-sectional hospital-based study. Copies of ten pairs of selected radiographs were distributed for the Set 1 (simulation of ante-mortem images), which contained 10 radiographs of before root canal treatment, and Set 2 (simulation of post-mortem images), which contained 10 radiographs of after endodontic interventions. Set 1 and Set 2 images were randomized, and structured via Google Form, and five teaching assistants from the Department of Restorative Dentistry were asked to match the images through provided Google Form. 
 Results. The range of 50-80% correct pre- and post-treatment images matching levels were registered among five operators. The mean correct matching level reached 68.0 ± 7.2%. Experience within endodontics seems to be valuable regressor regarding outcome performance (p < 0.05), while additional Radiology specialization did not significantly improved performance level for matching pre- and post-treatment images (p > 0.05). Operator graded 5 out of 10 images pair as relatively easy to match, 2 images – as moderately difficult to match, and 3 images – as hard to match (Fig. 3), while their agreement on the obtained above-mentioned scores reached 0.8.
 Conclusion. Significance of endodontics within forensic odontology field in most cases dictated by using the results of periapical radiology as evidences of ante-mortem origin, while also considering the fact of root canal morphology uniqueness and individualized endodontic treatment pattern consistency both ante- and post-mortem even under the influence of various disturbing factors. On the other hand identification role of endodontic signs may be diminished within the absence of proper quality radiographs, or when such were provided at critically different angulations, while also when endodontic-related evidences could not be supported by dental records.

Convolutional-neural-network-based radiographs evaluation assisting in early diagnosis of the periodontal bone loss via periapical radiograph

Background/PurposeThe preciseness of detecting periodontal bone loss is examiners dependent, and this leads to low reliability. The need for automated assistance systems on dental radiographic images has been increased. To the best of our knowledge, no studies have quantitatively and automatically staged periodontitis using dental periapical radiographs. The purpose of this study was to evaluate periodontal bone loss and periodontitis stage on dental periapical radiographs using deep convolutional neural networks (CNNs). Materials and methods336 periapical radiographic images (teeth: 390) between January 2017 and December 2019 were collected and de-identified. All periapical radiographic image datasets were divided into training dataset (n = 82, teeth: 123) and test dataset (n = 336, teeth: 390). For creating an optimal deep CNN algorithm model, the training datasets were directly used for the segmentation and individual tooth detection. To evaluate the diagnostic power, we calculated the degree of alveolar bone loss deviation between our proposed method and ground truth, the Pearson correlation coefficients (PCC), and the diagnostic accuracy of the proposed method in the test datasets. ResultsThe periodontal bone loss degree deviation between our proposed method and the ground truth drawn by the three periodontists was 6.5 %. In addition, the overall PCC value of our proposed system and the periodontists’ diagnoses was 0.828 (P < 0.01). The total diagnostic accuracy of our proposed method was 72.8 %. The diagnostic accuracy was highest for stage III (97.0 %). ConclusionThis tool helps with diagnosis and prevents omission, and this may be especially helpful for inexperienced younger doctors and doctors in underdeveloped countries. It could also dramatically reduce the workload of clinicians and timely access to periodontist care for people requiring advanced periodontal treatment.

Artificial intelligence in diagnosing dens evaginatus on periapical radiography with limited data availability

This study aimed to develop an artificial intelligence (AI) model using deep learning techniques to diagnose dens evaginatus (DE) on periapical radiography (PA) and compare its performance with endodontist evaluations. In total, 402 PA images (138 DE and 264 normal cases) were used. A pre-trained ResNet model, which had the highest AUC of 0.878, was selected due to the small number of data. The PA images were handled in both the full (F model) and cropped (C model) models. There were no significant statistical differences between the C and F model in AI, while there were in endodontists (p = 0.753 and 0.04 in AUC, respectively). The AI model exhibited superior AUC in both the F and C models compared to endodontists. Cohen’s kappa demonstrated a substantial level of agreement for the AI model (0.774 in the F model and 0.684 in C) and fair agreement for specialists. The AI’s judgment was also based on the coronal pulp area on full PA, as shown by the class activation map. Therefore, these findings suggest that the AI model can improve diagnostic accuracy and support clinicians in diagnosing DE on PA, improving the long-term prognosis of the tooth.

The diagnostic performance of the Java-based software colorization tool in the diagnosis of dental caries: an in vitro study.

This in vitro study aimed to compare the diagnostic outcomes of caries detection in premolar and molar teeth on digital periapical radiographs viewed as grayscale images and in the Java-based software (JBS) pseudocolor spectrum. The study comprised 92 premolar or molar human teeth (n = 46 with caries and n = 46 without clinical caries). For each tooth, digital periapical radiographic images (DPRIs), JBS pseudocolor images, and stereomicroscope images were collected, and caries presence and depth were recorded. In JBS pseudocolor images, regions of interest (ROIs) were detected, and ROI values for caries area (ROI1), non-caries area (ROI2), mesial area (ROI3), and distal area (ROI4) were obtained. The degree of inter-observer agreement was determined using the kappa coefficient, and the sensitivity and specificity of the methods were assessed using ROC analysis. While neither method was consistent with stereomicroscopy in determining caries depth (P < .005), for both observers the kappa value of JBS pseudocolor images were slightly higher than those for DPRIs. ROI1 was found to be significantly lower than ROI2 (P < .001), whereas no difference was found between ROI3 and ROI4 (P = .877). For both observers, the sensitivity and specificity obtained from JBS pseudocolor were above 89%. JBS pseudocolor images can be used for caries detection and offer high sensitivity and specificity.

Automated deep learning for classification of dental implant radiographs using a large multi-center dataset

This study aimed to evaluate the accuracy of automated deep learning (DL) algorithm for identifying and classifying various types of dental implant systems (DIS) using a large-scale multicenter dataset. Dental implant radiographs of pos-implant surgery were collected from five college dental hospitals and 10 private dental clinics, and validated by the National Information Society Agency and the Korean Academy of Oral and Maxillofacial Implantology. The dataset contained a total of 156,965 panoramic and periapical radiographic images and comprised 10 manufacturers and 27 different types of DIS. The accuracy, precision, recall, F1 score, and confusion matrix were calculated to evaluate the classification performance of the automated DL algorithm. The performance metrics of the automated DL based on accuracy, precision, recall, and F1 score for 116,756 panoramic and 40,209 periapical radiographic images were 88.53%, 85.70%, 82.30%, and 84.00%, respectively. Using only panoramic images, the DL algorithm achieved 87.89% accuracy, 85.20% precision, 81.10% recall, and 83.10% F1 score, whereas the corresponding values using only periapical images achieved 86.87% accuracy, 84.40% precision, 81.70% recall, and 83.00% F1 score, respectively. Within the study limitations, automated DL shows a reliable classification accuracy based on large-scale and comprehensive datasets. Moreover, we observed no statistically significant difference in accuracy performance between the panoramic and periapical images. The clinical feasibility of the automated DL algorithm requires further confirmation using additional clinical datasets.

Accurate detection for dental implant and peri-implant tissue by transfer learning of faster R-CNN: a diagnostic accuracy study.

The diagnosis of dental implants and the periapical tissues using periapical radiographs is crucial. Recently, artificial intelligence has shown a rapid advancement in the field of radiographic imaging. This study attempted to detect dental implants and peri-implant tissues by using a deep learning method known as object detection on the implant image of periapical radiographs. After implant treatment, the periapical images were collected and data were processed by labeling the dental implant and peri-implant tissue together in the images. Next, 300 images of the periapical radiographs were split into 80:20 ratio (i.e. 80% of the data were used for training the model while 20% were used for testing the model). These were evaluated using an object detection model known as Faster R-CNN, which simultaneously performs classification and localization. This model was evaluated on the classification performance using metrics, including precision, recall, and F1 score. Additionally, in order to assess the localization performance, an evaluation through intersection over union (IoU) was utilized, and, Average Precision (AP) was used to assess both the classification and localization performance. Considering the classification performance, precision = 0.977, recall = 0.992, and F1 score = 0.984 were derived. The indicator of localization was derived as mean IoU = 0.907. On the other hand, considering the indicators of both classification and localization performance, AP showed an object detection level of AP@0.5 = 0.996 and AP@0.75 = 0.967. Thus, the implementation of Faster R-CNN model for object detection on 300 periapical radiographic images including dental implants, resulted in high-quality object detection for dental implants and peri-implant tissues.

Modified pulpotomy procedure in immature permanent teeth with apical periodontitis: a randomised controlled trial

ObjectivesThe objective of this study was to evaluate the effectiveness of a modified pulpotomy (MP) procedure in immature permanent teeth with apical periodontitis (AP).DesignRandomised controlled trial.SettingOne public hospital in Shanghai,...

The outcome of endodontic treatment using an enhanced infection protocol in specialist practice.

Objectives This study aims to determine the outcome of primary root canal treatment with specific enhanced infection control protocol. The secondary aim was to compare percentages of successful outcomes in this study with a previous study undertaken by the same operator using both periapical radiograph (PR) and cone beam computed tomography (CBCT).Materials and methods Root canal treatment of 110 teeth in 95 patients carried out by a single operator using an enhanced infection control procedure (disinfection of gutta percha before obturation, changing of gloves after each intraoperative radiograph and also before the start of the root canal obturation). PR and CBCT scans of 94 teeth in 87 patients were assessed 12 months after completion of primary root canal treatment and compared with their respective pre-treatment (diagnostic) PR and CBCT scans. Healing was assessed by a consensus panel consisting of two calibrated examiners. Comparison of the PR and CBCT images for 'healed' and/or 'healing' outcomes was performed using McNemar's test.Results The 'healed' rate (absence of periapical radiolucency) was 89.4% using PR and 78.7% for CBCT (p <0.046). This rate increased to 95.7% for PR and 92.6% for CBCT when the 'healing' group (reduced size of periapical radiolucency) was included (p <0.046).Conclusion The frequent changing of gloves during the course of endodontic treatment and disinfection of gutta percha points before obturation, together with the use of contemporary rotary files, resulted in a high success rate of primary root canal treatment compared to similar clinical studies conducted previously.

An Audit of the Technical Quality and Iatrogenic Errors of Root Canal Treatment by Undergraduate Dental Students at the University of Sharjah.

The aim of this study was to determine the technical quality of root canal fillings and the presence of iatrogenic errors in the cases treated by undergraduate dental students using radiographic images. This study included 480 root-filled teeth, which were evaluated using intraoral periapical radiographic images. The technical quality of obturation was assessed by observing the length, density, and taperness of the root filling. Ledges, apical transportation, fractured instruments, zipping, and root perforation were recorded as iatrogenic errors. Teeth were classified as acceptable if the technical parameters were adequate and there were no iatrogenic errors. Data were statistically analyzed using Pearson's chi-squared test. The quality of root filling was acceptable in 183 of the 480 teeth. The rate of acceptable quality was higher for the teeth treated by 4th-year students (41.9%) than for those treated by 5th-year students (27.4%; p = 0.004). Length and taperness were adequate in more of the patients treated by the 4th-year students (p<0.05). There was a significant difference in the incidence of ledge formation and apical transportation in relation to the student's level (p<0.05). The technical quality of root fillings performed without iatrogenic errors by undergraduate dental students was declared acceptable in 38.1% of the assessed teeth. There was a noticeable difference in the quality of root canal treatment between the 4th- and 5th-year students. The findings demonstrate that periodic assessment of the technical quality of root filling performed by undergraduate dental students provides feedback on curriculum standards.

A pilot study of a deep learning approach to detect marginal bone loss around implants

BackgroundRecently, there has been considerable innovation in artificial intelligence (AI) for healthcare. Convolutional neural networks (CNNs) show excellent object detection and classification performance. This study assessed the accuracy of an artificial intelligence (AI) application for the detection of marginal bone loss on periapical radiographs.MethodsA Faster region-based convolutional neural network (R-CNN) was trained. Overall, 1670 periapical radiographic images were divided into training (n = 1370), validation (n = 150), and test (n = 150) datasets. The system was evaluated in terms of sensitivity, specificity, the mistake diagnostic rate, the omission diagnostic rate, and the positive predictive value. Kappa (κ) statistics were compared between the system and dental clinicians.ResultsEvaluation metrics of AI system is equal to resident dentist. The agreement between the AI system and expert is moderate to substantial (κ = 0.547 and 0.568 for bone loss sites and bone loss implants, respectively) for detecting marginal bone loss around dental implants.ConclusionsThis AI system based on Faster R-CNN analysis of periapical radiographs is a highly promising auxiliary diagnostic tool for peri-implant bone loss detection.

The Impact of Cone-beam Computed Tomography on Diagnostic Thinking, Treatment Option, and Confidence in Dental Trauma Cases: A Before and After Study

IntroductionThe aim of this study was to identify whether cone-beam computed tomographic (CBCT) imaging affects endodontists' diagnostic thinking, treatment option, and confidence in dental trauma cases. MethodsTwelve endodontists reviewed 15 dental trauma cases with clinical histories and periapical radiography (PR) and answered questions regarding their diagnostic thinking and treatment decisions and their confidence in both. One month later, the same participants reviewed the same cases and answered similar questionnaires with the aid of CBCT imaging. A statistical analysis of their responses was conducted. ResultsDifferences were observed in diagnostic thinking when using PR or CBCT imaging (P < .05), and the use of PR was associated with a higher number of “unsure” answers to questions about diagnoses. After reviewing periapical radiographic images, the approach most frequently endorsed was “don’t start conventional endodontic treatment and order CBCT,” whereas the most common decision made after analyzing the CBCT images was “start conventional endodontic treatment.” Clinical interventions were proposed more often when participants evaluated the clinical case using CBCT imaging than when using PR (P < .05). The participants' degree of confidence in their diagnostic thinking was not different after analysis using PR or CBCT imaging (P > .05). However, there was a difference between PR and CBCT imaging in participants' confidence in their treatment decisions (P < .05). Participants who initially indicated a lack of confidence in their diagnosis or treatment option gained confidence when CBCT images were available. However, the inverse behavior was also observed for participants who were initially confident on the basis of PR (P < .05). Participants who decided to start treatment after CBCT analysis were less confident than when using PR (P < .05). ConclusionsIn dental trauma cases, CBCT imaging influenced participants' diagnostic thinking and choice of treatment modality and affected their confidence in the decision-making process.

Use of the deep learning approach to measure alveolar bone level.

The goal was to use a deep convolutional neural network to measure the radiographic alveolar bone level to aid periodontal diagnosis. A deep learning (DL) model was developed by integrating three segmentation networks (bone area, tooth, cemento-enamel junction) and image analysis to measure the radiographic bone level and assign radiographic bone loss (RBL) stages. The percentage of RBL was calculated to determine the stage of RBL for each tooth. A provisional periodontal diagnosis was assigned using the 2018 periodontitis classification. RBL percentage, staging, and presumptive diagnosis were compared with the measurements and diagnoses made by the independent examiners. The average Dice Similarity Coefficient (DSC) for segmentation was over 0.91. There was no significant difference in the RBL percentage measurements determined by DL and examiners ( ). The area under the receiver operating characteristics curve of RBL stage assignment for stages I, II, and III was 0.89, 0.90, and 0.90, respectively. The accuracy of the case diagnosis was 0.85. The proposed DL model provides reliable RBL measurements and image-based periodontal diagnosis using periapical radiographic images. However, this model has to be further optimized and validated by a larger number of images to facilitate its application.

A Comparison of Cone Beam Computed Tomography and Periapical Digital Radiography for Evaluation of Root Canal Preparation

The aim of the present study is to compare cone beam computed tomography and periapical digital radiography for the evaluation of root canal preparation. Nine extracted human molars were used in this study. Following access cavity preparation, mesio-buccal roots of maxillary and mesial roots of mandibular molars were prepared and the remaining roots were cut off. Three amalgam cavities were prepared on the coronal part of the teeth and were filled with amalgam to be used as landmarks. Specimens were scanned using cone-beam computed tomography and periapical digital radiograph images were obtained before and after root canal preparation. WaveOne Gold Primary was used for root canal preparation to full working length. Specimens were then scanned using CBCT and a periapical radiograph for the after-instrumentation images. The transportation and centering ratio were measured and calculated on the CBCT and periapical radiographic images. The Bland–Altman method was used for detecting the bias in the evaluation of agreement between the two methods’ measurements. There was agreement between the two methods’ measurements using CBCT scans and periapical digital radiographic images in the evaluation of transportation and centering ratio parameters. The two methods could be used interchangeably in measurements of transportation and calculating the centering ratio.

Root canal curvature as a prognostic factor influencing the diagnostic accuracy of radiographic working length determination and postoperative canal axis modification: an in vitro comparative study

BackgroundRadiographic analysis of tooth morphology is mandatory for accurate calibration of the degree of canal curvature angle and radiographic working length to its real dimensions in case difficulty assessment protocols. This study aimed to determine the impact of the degree of root canal curvature angle on maintaining the real working length and the original canal axis of prepared root canals using a reciprocating rotary instrumentation technique.MethodsRadiographic image analysis was performed on 60 extracted single-rooted human premolar teeth with a moderate canal curvature (10°–25°) and severe canal curvature (26°–70°). Working length and longitudinal canal axis were determined using cone-beam computed tomography (CBCT) and digital periapical radiography. The real canal length was determined by subtracting 0.5 mm from the actual canal length. Root canals were prepared using the WaveOne Gold reciprocating file (Dentsply Maillefer, Ballaigues, Switzerland).ResultsThere was no significant relation of the degree of canal curvature angle to the accuracy of radiographic working length estimated on CBCT and digital periapical radiographic techniques (P > 0.05). Postinstrumentation changes in the original canal axis between moderate and severe canal curvature angles, assessed on CBCT and periapical digital radiographic images were statistically non-significant (P > 0.05).ConclusionsA standardized digital periapical radiographic method performed similarly to the CBCT technique near to its true working length. No significant interaction exists between the diagnostic working length estimation, postoperative root canal axis modification, and the degree of canal curvature angle, using reciprocating rotary instrumentation technique.

Peri-Implant Bone Loss Measurement Using a Region-Based Convolutional Neural Network on Dental Periapical Radiographs.

Determining the peri-implant marginal bone level on radiographs is challenging because the boundaries of the bones around implants are often unclear or the heights of the buccal and lingual bone levels are different. Therefore, a deep convolutional neural network (CNN) was evaluated for detecting the marginal bone level, top, and apex of implants on dental periapical radiographs. An automated assistant system was proposed for calculating the bone loss percentage and classifying the bone resorption severity. A modified region-based CNN (R-CNN) was trained using transfer learning based on Microsoft Common Objects in Context dataset. Overall, 708 periapical radiographic images were divided into training (n = 508), validation (n = 100), and test (n = 100) datasets. The training dataset was randomly enriched by data augmentation. For evaluation, average precision, average recall, and mean object keypoint similarity (OKS) were calculated, and the mean OKS values of the model and a dental clinician were compared. Using detected keypoints, radiographic bone loss was measured and classified. No statistically significant difference was found between the modified R-CNN model and dental clinician for detecting landmarks around dental implants. The modified R-CNN model can be utilized to measure the radiographic peri-implant bone loss ratio to assess the severity of peri-implantitis.

Artificial Intelligence in Fractured Dental Implant Detection and Classification: Evaluation Using Dataset from Two Dental Hospitals.

Fracture of a dental implant (DI) is a rare mechanical complication that is a critical cause of DI failure and explantation. The purpose of this study was to evaluate the reliability and validity of a three different deep convolutional neural network (DCNN) architectures (VGGNet-19, GoogLeNet Inception-v3, and automated DCNN) for the detection and classification of fractured DI using panoramic and periapical radiographic images. A total of 21,398 DIs were reviewed at two dental hospitals, and 251 intact and 194 fractured DI radiographic images were identified and included as the dataset in this study. All three DCNN architectures achieved a fractured DI detection and classification accuracy of over 0.80 AUC. In particular, automated DCNN architecture using periapical images showed the highest and most reliable detection (AUC = 0.984, 95% CI = 0.900–1.000) and classification (AUC = 0.869, 95% CI = 0.778–0.929) accuracy performance compared to fine-tuned and pre-trained VGGNet-19 and GoogLeNet Inception-v3 architectures. The three DCNN architectures showed acceptable accuracy in the detection and classification of fractured DIs, with the best accuracy performance achieved by the automated DCNN architecture using only periapical images.

A Performance Comparison between Automated Deep Learning and Dental Professionals in Classification of Dental Implant Systems from Dental Imaging: A Multi-Center Study.

In this study, the efficacy of the automated deep convolutional neural network (DCNN) was evaluated for the classification of dental implant systems (DISs) and the accuracy of the performance was compared against that of dental professionals using dental radiographic images collected from three dental hospitals. A total of 11,980 panoramic and periapical radiographic images with six different types of DISs were divided into training (n = 9584) and testing (n = 2396) datasets. To compare the accuracy of the trained automated DCNN with dental professionals (including six board-certified periodontists, eight periodontology residents, and 11 residents not specialized in periodontology), 180 images were randomly selected from the test dataset. The accuracy of the automated DCNN based on the AUC, Youden index, sensitivity, and specificity, were 0.954, 0.808, 0.955, and 0.853, respectively. The automated DCNN outperformed most of the participating dental professionals, including board-certified periodontists, periodontal residents, and residents not specialized in periodontology. The automated DCNN was highly effective in classifying similar shapes of different types of DISs based on dental radiographic images. Further studies are necessary to determine the efficacy and feasibility of applying an automated DCNN in clinical practice.

Evaluation of computer-aided diagnosis system for detecting dental approximal caries lesions on periapical radiographs

Objective: To establish and to evaluate a computer-aided system based on deep-learning for detection and diagnosis of dental approximal caries on periapical radiographs. Methods: One hundred and sixty human premolars and molars extracted for orthodontic or periodontal reasons were obtained from Department of Oral and Maxillofacial Surgery, Affiliated Stomatological Hospital, Fujian Medical University. A total of 160 periapical radiographic images were divided into a training dataset (n=80) and a test dataset (n=80). A deep-learning based computer-aided caries diagnosis system was established and trained. The performances of computer-aided diagnosis system and human observer were compared using receiver operating characteristic (ROC) curves, precision-recall (P-R) curves, the area under the curve (AUC), sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). The AUC values of human observers and caries diagnosis system was compared by using an online statistical tool (SPSSAU 20.0). Chi-square test was used to analyze the differences between human observers and caries diagnosis system (ɑ=0.05). Results: The AUC values of human observers and caries diagnosis system were 0.729 (95%CI: 0.650-0.808) and 0.762 (95%CI: 0.685-0.839), respectively (P>0.05). No significant differences were found for the specificity, PPV and NPV between the caries diagnosis system and human observers (P all>0.05). The caries diagnosis system was significantly more sensitive in detecting dental proximal caries than human observers (P<0.05). For the diagnosis of level-1 caries (caries limited to outer 1/2 of enamel), the sensitivity of human observers and computer-aided detection system were 27% and 77%, respectively (P<0.05). Conclusions: The computer-aided diagnosis system provided similar accuracy as human observers and significantly better sensitivity than human observers, especially for shallow caries in enamel.