PurposeThis study aimed to develop a deep learning model incorporating the pulp-to-tooth area ratio (PTR) for estimating age from periapical radiographs of upper central incisors in a Thai population.Materials and MethodsA total of 2,041 periapical radiographs, representing 3,108 upper central incisors from individuals aged 10.00 to 84.25 years, were analyzed. Root and root canal segmentation masks were generated using Labelbox. The dataset was randomly divided into training (2,175 teeth), validation (466 teeth), and test (467 teeth) sets. Model development proceeded in 3 steps: (1) DeepLabv3+ was trained to segment root and root canal regions. (2) ResNet-50 was trained to estimate age using both segmented images and PTR values derived from the masks. (3) The validation set guided model selection, and the final model was evaluated on the test set using mean absolute error (MAE).ResultsThe segmentation model (DeepLabv3+) achieved a mean intersection over union (mIoU) of 81.85%. The age estimation model (ResNet-50) yielded an overall MAE of 6.14 years. For age-specific evaluation, test data were grouped into 4 categories: 10–19 years, 20–39 years, 40–59 years, and ≥60 years. The MAE was lowest in the 10–19 years group (3.74 years) and progressively increased with age: 6.07, 6.58, and 8.40 years, respectively.ConclusionThis study demonstrated that integrating deep learning models with PTR measurement offers a promising method for dental radiographic age estimation using a single tooth.

PurposeThis study investigated the effect of increasing slice thickness and applying sharpening filters on the diagnosis of bone graft loss in regions adjacent to zirconia implants using cone-beam computed tomography (CBCT).Materials and MethodsTwelve zirconia implants were inserted into mandibles, followed by bone graft application. Graft loss was then simulated in half of the samples. CBCT scans were obtained using 2 devices (OP300 and Eagle 3D) with parameters set at 90 kVp, 8 mA, and a 133 µm voxel size. Five examiners assessed the scans under different conditions: original thickness, 1 mm, 2 mm, no filter, Sharpen 1×, and Sharpen 2×. The area under the receiver operating characteristic curve (AUC), sensitivity, and specificity were calculated and compared using 2-way analysis of variance (α=5%).ResultsFor the OP300 device, increasing slice thickness to 1 mm impaired AUC, sensitivity, and specificity metrics (P<0.05). Overall, filter application reduced this impairment (P<0.05). For the Eagle device, slice thickness did not affect diagnostic metrics, but specificity values increased with sharpening filter application on scans at original thickness or with 1 mm slices (P<0.05).ConclusionFor 1 device, a 1 mm increase in slice thickness impaired the diagnosis of bone graft loss in areas adjacent to zirconia implants. However, the application of sharpening filters helped mitigate this impairment.

PurposeImplant identification is a pressing concern in dental implantology, and artificial intelligence (AI) has been evaluated for this purpose. YOLO, a state-of-the-art object detection model, is suitable for medical imaging; therefore, this study assessed YOLOv11—the latest iteration—for identifying 10 implant types in Indian clinical settings and compared its accuracy to that of dental professionals.Materials and MethodsA dataset of 3,161 radiographs, comprising both periapical and panoramic images of 10 implant types, was annotated and used to train and test YOLOv11. Training was performed on Google Colab using an NVIDIA Tesla T4 GPU (16 GB VRAM). A random sample of 200 radiographs was selected from the test dataset and presented to 50 dental practitioners for implant identification. Their responses were analysed and compared, using the chi-square test for statistical significance.ResultsYOLOv11 achieved precision of 0.87, recall of 0.85, an F1-score of 0.86, and an mAP50 of 0.899. The model achieved excellent classification accuracy for Adin (95%), MIS (94%), Bego (92%), ITI (96%), and Bicon (97%). Moderate accuracy was noted for Noris (82%), Osstem (85%), AlphaBio (88%), Dentium (77%), and Bioline (75%). YOLOv11 demonstrated higher overall accuracy and consistency than dental professionals. Dentists’ accuracy ranged from 27% to 49%, whereas that of YOLOv11 ranged from 92% to 100%.ConclusionYOLOv11 recognised most implant classes with over 90% accuracy, surpassing traditional manual techniques in implant detection. Although the model is dependable and efficient, certain aspects require improvement. The study also emphasises the significance of a region-specific approach for clinical relevance.

PurposeThis study analyzed cone-beam computed tomography images of 27 patients with multiple myeloma at different disease stages to identify jawbone destruction patterns and assess their associations with clinical data.Materials and MethodsIn this cross-sectional study, 2 trained examiners performed standardized, consensus-based image analyses. Lesions were classified into 4 distinct bone destruction patterns: diffuse, multilocular, unilocular, and punched-out. Clinical data were collected from medical records.ResultsThe sample included 51.8% male and 48.2% female patients, predominantly between 42 and 60 years old. All cases exhibited diffuse bone destruction affecting both jaws. Multilocular and unilocular patterns were observed in 51.9% and 29.6% of cases, respectively, while no punched-out lesions were identified. The unilocular pattern was significantly associated with cases classified as International Staging System stage I and Durie-Salmon stage IIIA.ConclusionAmong the studied cases of multiple myeloma, the most frequently observed bone destruction patterns were diffuse and multilocular. The absence of punched-out lesions may be attributable to the use of 3-dimensional imaging. A clear association was identified between the unilocular pattern and disease staging.

PurposeThis study assessed the effectiveness of an educational video in reducing tongue positioning errors on panoramic radiographs.Materials and MethodsAn educational video instructing patients on proper tongue positioning during panoramic radiograph acquisition was sent via WhatsApp (WhatsApp Inc., Menlo Park, CA, USA) at the time of appointment scheduling. Patients were instructed to view the video again before their appointment. Collected data included patients’ sex, age, scheduling method, educational background, the necessity for panoramic radiograph retake, and the reason for retake. The frequency of retakes due to tongue positioning errors was compared with retrospective data from patients who did not receive the video, resulting in the evaluation of 1,088 panoramic radiographs. Descriptive data analyses were conducted, and simple and multiple logistic regression models were applied with a significance level of 5%.ResultsOf the 1,088 panoramic radiographs evaluated, 69 displayed tongue positioning errors. Of these, 53 radiographs were from patients without access to the educational video, whereas only 16 were from patients who had received the video (P<0.05). Patients without video access were 2.07 times more likely to exhibit tongue positioning errors than those who had access. The other variables assessed (sex, age, scheduling method, and educational background) did not significantly influence tongue positioning errors (P>0.05).ConclusionProviding patients with an educational video on proper tongue positioning significantly reduced tongue positioning errors on panoramic radiographs.

PurposeThis study was performed to assess the relationship of the third molars with the mandibular canal as a predictor of inferior alveolar nerve (IAN) sensory disturbances using panoramic radiography (PR) and cone-beam computed tomography (CBCT).Materials and MethodsA systematic search was conducted of 4 databases—PubMed, Scopus, Web of Science, and Google Scholar—for the period from 1985 to 2024. In the retrieved articles, the outcome of interest was the relationship of the mandibular canal with the third molars on PR and CBCT scans. The risk of bias was assessed using the Newcastle-Ottawa Scale, and quantitative meta-analysis was performed using STATA. A random-effects restricted maximum likelihood model was employed for the meta-analysis, and the I2 statistic was used to assess heterogeneity.ResultsA total of 1,635 articles were initially retrieved. After a rigorous selection process, 20 studies were included in the qualitative synthesis, and 8 were selected for the meta-analysis. The findings indicated that CBCT yielded higher prevalence rates for root darkening, root deflection, interruption of the white line, diversion of the mandibular canal, and narrowing of the mandibular canal (theta values: 49.962, 4.76, 8.09, 2.229, and 4.708, respectively) compared with PR (theta values: 1.363, 1.605, 6.322, 0.655, and 1.449, respectively).ConclusionCBCT was more accurate than PR in investigating predictors of IAN paresthesia in mandibular third molar surgery. Considering the higher prevalence of paresthesia in the presence of root darkening, CBCT may be highly efficient in detecting this parameter and thus aiding in the prevention of paresthesia.

Given the growing use of cone-beam computed tomography (CBCT) scans, this study assessed radiation exposure from these scans in the context of national guidelines and recommended dose limits. The current literature was reviewed to quantify the benefit of opportunistic diagnosis of carotid artery calcification relative to the potential risk of radiation-induced cancer. The average radiation from CBCT at its largest field of view and highest resolution possible amounts to a reasonable but still low ionizing radiation exposure. This exposure is comparable to 22 days of background radiation and is notably lower than the radiation exposure from medical CT scans. According to the risk assessment analysis, the risk of stroke events involving internal and external carotid artery calcification (CAC) was 202 and 67 per 100,000 individuals, respectively. In contrast, the estimated risk of radiation-induced cancer associated with CBCT was notably lower, at 0.6 per 100,000. The present study advocates for a comprehensive assessment of CBCT scans encompassing the areas of the internal and external carotid arteries by a knowledgeable professional, given the potential advantages of early detection of vascular abnormalities. Dental professionals who take scans involving these areas need to be mindful of reporting these findings and refer patients to their primary care physician for further investigation.

The purpose of this study was to classify mandibular molar furcation involvement (FI) in periapical radiographs using a deep learning algorithm. Full mouth series taken at East Carolina University School of Dental Medicine from 2011-2023 were screened. Diagnostic-quality mandibular premolar and molar periapical radiographs with healthy or FI mandibular molars were included. The radiographs were cropped into individual molar images, annotated as " healthy" or " FI," and divided into training, validation, and testing datasets. The images were preprocessed by PyTorch transformations. ResNet-18, a convolutional neural network model, was refined using the PyTorch deep learning framework for the specific imaging classification task. CrossEntropyLoss and the AdamW optimizer were employed for loss function training and optimizing the learning rate, respectively. The images were loaded by PyTorch DataLoader for efficiency. The performance of ResNet-18 algorithm was evaluated with multiple metrics, including training and validation losses, confusion matrix, accuracy, sensitivity, specificity, the receiver operating characteristic (ROC) curve, and the area under the ROC curve. After adequate training, ResNet-18 classified healthy vs. FI molars in the testing set with an accuracy of 96.47%, indicating its suitability for image classification. The deep learning algorithm developed in this study was shown to be promising for classifying mandibular molar FI. It could serve as a valuable supplemental tool for detecting and managing periodontal diseases.

This study was conducted to compare the objective image quality of radiographs acquired with a handheld X-ray device to those obtained with a wall-mounted X-ray device. Brightness, noise, uniformity, and contrast were evaluated. To assess the first 3 parameters, radiographs of an acrylic block were acquired with an unused photostimulable phosphor (PSP) plate from the VistaScan system (Mini Easy, Dürr Dental, Bietigheim-Bissingen, Germany). Initially, 6 radiographs were taken with a Focus X-ray wall-mounted device (Instrumentarium, Tuusula, Finland) operating at 60 kVp, 7 mA, and 0.125 s. Another 6 radiographs were captured using an Eagle handheld X-ray device (Alliage, São Paulo, Brazil) at 60 kVp, 2.5 mA, and 0.35 s. The means and standard deviations of the gray values for all radiographs were calculated using ImageJ (National Institutes of Health, Bethesda, MD, USA). For contrast assessment, radiographs of an aluminum step wedge were obtained using the same PSP plate, X-ray devices, and acquisition parameters. The percentage of contrast variation was determined. The impacts of the devices on image quality were compared using the Student t-test, with a significance level of 5% (P<0.05). Compared with the wall-mounted device, the handheld device produced radiographs with higher brightness and noise, as indicated by mean values of 6.57 (0.49) and 3.49 (0.02), respectively. Furthermore, it demonstrated lower uniformity and contrast, with respective means of 3.75 (0.02) and 35.48 (0.09) (P<0.05). Radiographs obtained using a handheld X-ray device exhibit lower theoretical image quality than those acquired with a wall-mounted device.