OBJECTIVE:To develop a machine learning model using computed tomography (CT) images for preoperative differential diagnosis of ameloblastoma (AM) and odontogenic keratocyst (OKC), and to compare its performance with that of an oral and maxillofacial radiologist. STUDY DESIGN:This retrospective study analyzed CT images from 154 patients (70 AM, 84 OKC). Seventeen clinical and imaging features were extracted and compiled into a CSV file. The data were randomly divided into five subsets for five-fold cross-validation. In each fold, 80% of the data were used for training and 20% for testing. Two machine learning models-Prediction One and Random Forest-were trained on the training data and evaluated on the test data. Diagnostic performance was assessed using receiver operating characteristic (ROC) curves, and the AUC were calculated. The performance of each model was compared with that of a radiologist blinded to clinical data. Variable importance was also analyzed to identify key diagnostic features. RESULTS:Prediction One achieved an AUC of 0.95. Random Forest achieved an AUC of 0.96. Both models significantly outperformed the radiologist (AUC: 0.80, P < .05). The most influential diagnostic features were root resorption and locularity. CONCLUSION:Machine learning models accurately distinguished AM from OKC and may assist preoperative diagnosis and planning.

OBJECTIVES:In temporomandibular joint osteoarthritis, condylar erosion is considered a sign of disease progression. Few studies have clarified the extent to which clinical and imaging findings are associated with erosion. The purpose of this study was to develop a machine learning model to predict condylar erosion and identify the parameters contributing to prediction. METHODS:We enrolled 197 patients (394 joints) in this study. Condylar erosion was determined using magnetic resonance imaging (MRI) and panoramic radiographs. Clinical data included age, sex, duration of symptoms, maximum mouth opening, joint sounds, temporomandibular joint pain, and masticatory muscle pain. MRI findings were evaluated for disc shape and position, signal changes at the posterior disc attachment (PDA), joint effusion, bone edema, condylar position, and limited movement. Erosion prediction was performed using clinical and MRI data with Prediction One and random forest models. Area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity were calculated. Parameters contributing to prediction were analyzed. RESULTS:Prediction One showed AUC 0.845, accuracy 0.779, sensitivity 0.777, and specificity 0.780. Random forest had AUC 0.836, accuracy 0.759, sensitivity 0.777, and specificity 0.753. The top five parameters contributing to prediction in Prediction One were signal changes at PDA, disc position, bone edema, disc shape, and joint effusion; in random forest, these were disc position, joint effusion, signal changes at PDA, bone edema, and disc shape. CONCLUSIONS:We developed a highly accurate model for predicting condylar erosion and identified those parameters contributing to the prediction.

OBJECTIVE:Predicting the surgical time required for mandibular third molar extraction is challenging because patient-, tooth-, and operator-related factors interact in complex ways. This study aimed to develop an interpretable machine-learning model using patient demographics, panoramic radiographic features, and surgeon-related variables. STUDY DESIGN:We retrospectively reviewed 713 mandibular third molar extractions at a single university hospital. Models were developed using Prediction One, a no-code machine-learning platform. Because continuous-time prediction showed limited performance, we reframed the primary task as binary classification at clinically relevant thresholds (≥31, ≥46, and ≥61 minutes). Performance was benchmarked against three traditional models and feature contribution scores were calculated to assess variable importance. RESULTS AND DISCUSSION:The ≥31-minute model showed the best performance (accuracy, 0.75; area under the curve, 0.80), and outperformed the comparator models. Surgeon experience and specialist certification were the strongest contributors, followed by root-related radiographic morphology. In contrast, indices such as the Pell and Gregory classification and the number of roots, showed limited predictive value. CONCLUSIONS:Minute-by-minute prediction was difficult, but threshold-based prediction was feasible, and we identified factors that contributed substantially to time prediction and those with minimal impact.

This study aimed to evaluate the clinical feasibility of dental magnetic resonance imaging (MRI) using a microscopy coil by comparing imaging positions in healthy volunteers. Twenty-six healthy volunteers underwent dental MRI in supine and prone positions using a 47 mm microscopy coil on a 3.0T system. T1-weighted (T1W), T2-weighted (T2W), and proton density-weighted (PDW) sequences were acquired. Participant-reported burden was assessed using a 10-point scale. Image quality was evaluated using a 4-point scale for sharpness, artifact, perceived signal-to-noise ratio (SNR), and overall quality. SNR and contrast-to-noise ratio (CNR) were calculated from 1 mm2 regions of interest in dental pulp, inferior alveolar neurovascular bundle, and bone marrow. Statistical analyses included the Wilcoxon signed-rank test, Fisher’s exact test, and paired t-test. Participant-reported burden was lower in the supine position (1.7 ± 1.1) than in the prone (4.7 ± 2.1). Visual assessments demonstrated superior image quality in supine across all sequences. The proportion of non-diagnostic images was higher in the prone position: T1W (50

Purpose:The aim of this research was to develop a prediction model for diagnosis of Sjögren's syndrome using radiomics and machine learning techniques applied to computed tomography images of the parotid glands and to assess its efficacy by temporal validation. Materials and Methods:In total, 132 parotid glands from 66 subjects (33 patients with Sjögren's syndrome and 33 controls) were analyzed. Radiomics features were extracted from manually segmented parotid glands using 3D Slicer. The volume data for 108 parotid glands were chronologically assigned to the training dataset, and the features extracted were imported into Prediction One (Sony Network Communications Inc, Tokyo, Japan). A prediction model was automatically generated. The area under the curve (AUC), accuracy, precision, recall, and F-value were calculated for internal validation. Temporal validation was performed using 24 images of the parotid glands obtained later. Results:A total of 129 radiomics features were extracted, including 18 first-order, 14 shape, and 75 texture features. The internal validation test showed high performance, with an AUC of 0.92, accuracy of 0.88, precision of 0.90, recall of 0.85, and an F-value of 0.88. Temporal validation testing also showed high performance, with an AUC of 0.96. accuracy of 0.88, precision of 0.85, recall of 0.92, and an F-value of 0.88. Conclusion:The prediction model effectively differentiated Sjögren's syndrome using radiomics and machine learning. Use of Prediction One significantly streamlined the workflow, including analysis of radiomics, creation of the prediction model, and evaluation of performance, while substantially reducing the time required.

Magnetic resonance imaging (MRI) has been employed to obtain high-resolution images of dental structures using intraoral and dedicated coils; however, reports on microscopy coils are limited. Furthermore, no reports have detailed the imaging conditions for the clinical application of T1-weighted (T1W), T2-weighted (T2W), and proton density-weighted (PDW) sequences in dental MRI. This study investigated the optimal imaging conditions for clinical dental MRI. Phantoms simulating the dental pulp and bone marrow were constructed, and imaging was performed using a 3T-MRI system and a microscopy coil with varying T1W, T2W, and PDW parameters to determine the trends in change. Subsequently, we imaged the left mandibular first molar region of 21 healthy volunteers using clinically feasible parameters. Tooth visibility was assessed for the T1W, T2W, and PDW images, while contrast ratio (CR) and signal-to-noise ratio (SNR) were calculated and statistically analyzed. Results showed that PDW significantly outperformed T1W and T2W in terms of tooth visibility, CR, and SNR. Under the specified imaging conditions, PDW was optimal for tooth and periodontal tissue morphology evaluation. A statistically significant difference in CR and tooth visibility evaluation was observed between T1W and T2W in the dental pulp. This statistically significant difference suggested that T2W can be used to evaluate the dental pulp, and T1W can be used to evaluate the inferior alveolar nerve and bone marrow properties.

Objective. This study aimed to develop an ultrasound image-based radiomics model for diagnosing cervical lymph node (LN) metastasis in patients with head and neck squamous cell carcinoma (HNSCC) that shows higher accuracy than previous models. Study Design. A total of 537 LN (260 metastatic and 277 nonmetastatic) from 126 patients (78 men, 48 women, average age 63 years) were enrolled. The multivariate analysis software Prediction One (Sony Network Communications Corporation) was used to create the diagnostic models. Furthermore, three machine learning methods were adopted as comparison approaches. Based on a combination of texture analysis results, clinical information, and ultrasound findings interpretated by specialists, a total of 12 models were created, three for each machine learning method, and their diagnostic performance was compared. Results. The three best models had area under the curve of 0.98. Parameters related to ultrasound findings, such as presence of a hilum, echogenicity, and granular parenchymal echoes, showed particularly high contributions. Other significant contributors were those from texture analysis that indicated the minimum pixel value, number of contiguous pixels with the same echogenicity, and uniformity of gray levels. Conclusions. The radiomics model developed was able to accurately diagnose cervical LN metastasis in HNSCC. (Oral Surg Oral Med Oral Pathol Oral Radiol 2025;139:760-769)

This study investigated deep learning (DL) systems for diagnosing carotid artery calcifications (CAC) on panoramic radiographs. To this end, two DL systems, one with preceding and one with simultaneous area detection functions, were developed to classify CAC on panoramic radiographs, and their person-based classification performances were compared with that of a DL model directly created using entire panoramic radiographs. A total of 580 panoramic radiographs from 290 patients (with CAC) and 290 controls (without CAC) were used to create and evaluate the DL systems. Two convolutional neural networks, GoogLeNet and YOLOv7, were utilized. The following three systems were created: (1) direct classification of entire panoramic images (System 1), (2) preceding region-of-interest (ROI) detection followed by classification (System 2), and (3) simultaneous ROI detection and classification (System 3). Person-based evaluation using the same test data was performed to compare the three systems. A side-based (left and right sides of participants) evaluation was also performed on Systems 2 and 3. Between-system differences in area under the receiver-operating characteristics curve (AUC) were assessed using DeLong’s test. For the side-based evaluation, the AUCs of Systems 2 and 3 were 0.89 and 0.84, respectively, and in the person-based evaluation, Systems 2 and 3 had significantly higher AUC values of 0.86 and 0.90, respectively, compared with System 1 (P < 0.001). No significant difference was found between Systems 2 and 3. Preceding or simultaneous use of area detection improved the person-based performance of DL for classifying the presence of CAC on panoramic radiographs.

OBJECTIVES:This study aimed to identify the most effective diagnostic assistance system for assessing the relationship between mandibular third molars (M3M) and mandibular canals (MC) using panoramic radiographs. METHODS:In total, 2,103 M3M were included from patients in whom the M3M and MC overlapped on panoramic radiographs. All M3M were classified into high-risk and low-risk groups based on the degree of contact with the MC observed on computed tomography. The contact classification was evaluated using four machine learning models (Prediction One software, AdaBoost, XGBoost, and random forest), three convolutional neural networks (CNNs) (EfficientNet-B0, ResNet18, and Inception v3), and three human observers (two radiologists and one oral surgery resident). Receiver operating characteristic curves were plotted; the area under the curve (AUC), accuracy, sensitivity, and specificity were calculated. Factors contributing to prediction of high-risk cases by machine learning models were identified. RESULTS:Machine learning models demonstrated AUC values ranging from 0.84 to 0.88, with accuracy ranging from 0.81 to 0.88 and sensitivity of 0.80, indicating consistently strong performance. Among the CNNs, ResNet18 achieved the best performance, with an AUC of 0.83. The human observers exhibited AUC values between 0.67 and 0.80. Three factors were identified as contributing to prediction of high-risk cases by machine learning models: increased root radiolucency, diversion of the MC, and narrowing of the MC. CONCLUSION:Machine learning models demonstrated strong performance in predicting the three-dimensional relationship between the M3M and MC.

Purpose: This study showed long-term clinical results and quantitative dose-volume evaluation of image-guided high-dose-rate interstitial brachytherapy (IG HDR-BT) with custom-made surface applicators for lower lip cancer as monotherapy. Material and methods: Patients with localized lower lip cancer, who received IG HDR-BT with custom-made surface applicators as monotherapy at the NHO Osaka National Hospital between February 2012 and January 2015 were enrolled in this study. One to three applicators were implanted interstitially, and two to six were placed on tumor surface. Planning-aimed dose (PAD) was 54 Gy, 48 Gy for a recurrent case, and irradiation was delivered at 6 Gy/fraction, twice a day. Dosimetric goal was to achieve D90: clinical target volume (CTV) > PAD without exces-sive dose to the mandible. A lead shield was placed between the gingiva and lower lip during irradiation to reduce the dose to the mandible in all but one edentulous patient. A gauze with 2% lidocaine was inserted intra-orally to re-duce the dose to the upper lip and maxilla. Results: Six patients (T1 : T2 : T3 = 3 : 2 : 1), including one recurrent case, were enrolled in this study. The CTV was contoured on computed tomography and in two cases, magnetic resonance imaging was used as a reference for contouring. The median follow-up was 69.5 months. The primary tumor was controlled in all cases. No serious late ad-verse reactions were observed. The median D90 (CTV) and V100 (PAD) were 108.9% PAD and 99.3% CTV, respectively. The median D0.1cm3 (mandible) was 3.2 Gy per fraction. Conclusions: IG HDR-BT with custom-made surface applicators for lower lip cancer as monotherapy showed an excellent CTV dose and acceptable doses to the mandible, with good long-term clinical results

Objectives The purpose of this study was to generate radiographs including dentigerous cysts by applying the latest generative adversarial network (GAN; StyleGAN3) to panoramic radiography.Methods A total of 459 cystic lesions were selected, and 409 images were randomly assigned as training data and 50 images as test data. StyleGAN3 training was performed for 500 000 images. Fifty generated images were objectively evaluated by comparing them with 50 real images according to four metrics: Fr & eacute;chet inception distance (FID), kernel inception distance (KID), precision and recall, and inception score (IS). A subjective evaluation of the generated images was performed by three specialists who compared them with the real images in a visual Turing test.Results The results of the metrics were as follows: FID, 199.28; KID, 0.14; precision, 0.0047; recall, 0.00; and IS, 2.48. The overall results of the visual Turing test were 82.3%. No significant difference was found in the human scoring of root resorption.Conclusions The images generated by StyleGAN3 were of such high quality that specialists could not distinguish them from the real images.

The objective of this study was to enhance the visibility of soft tissues on cone-beam computed tomography (CBCT) using a CycleGAN network trained on CT images. Training and evaluation of the CycleGAN were conducted using CT and CBCT images collected from Aichi Gakuin University (α facility) and Osaka Dental University (β facility). Synthesized images (sCBCT) output by the CycleGAN network were evaluated by comparing them with the original images (oCBCT) and CT images, and assessments were made using histogram analysis and human scoring of soft-tissue anatomical structures and cystic lesions. The histogram analysis showed that on sCBCT, soft-tissue anatomical structures showed significant shifts in voxel intensity toward values resembling those on CT, with the mean values for all structures approaching those of CT and the specialists’ visibility scores being significantly increased. However, improvement in the visibility of cystic lesions was limited. Image synthesis using CycleGAN significantly improved the visibility of soft tissue on CBCT, with this improvement being particularly notable from the submandibular region to the floor of the mouth. Although the effect on the visibility of cystic lesions was limited, there is potential for further improvement through refinement of the training method.

Surgeons routinely interpret preoperative radiographic images for estimating the shape and position of the tooth prior to performing tooth extraction. In this study, we aimed to predict the difficulty of lower wisdom tooth extraction using only panoramic radiographs. Difficulty was evaluated using the modified Parant score. Two oral surgeons (a specialist and a clinical resident) predicted the difficulty level of the test data. This study also aimed to evaluate the performance of a deep learning model in predicting the necessity for tooth separation or bone removal during wisdom tooth extraction. Two convolutional neural networks (AlexNet and VGG-16) were created and trained using panoramic X-ray images. Both surgeons interpreted the same images and classified them into three groups. The accuracies for humans were 54.4% for both surgeons, 57.7% for AlexNet, and 54.4% for VGG-16. These results indicate that accurately predict the difficulty of wisdom teeth extraction using panoramic radiographs alone is challenging. However, AlexNet and VGG-16 had sensitivities of more than 90% for crown and root separation. The predictive ability of our proposed model is equivalent to that of an oral surgery specialist, and a recall value > 90% makes it suitable for screening in clinical settings.

Purpose: The aims of this study were to create a deep learning model to distinguish between nasopalatine duct cysts (NDCs), radicular cysts, and no-lesions (normal) in the midline region of the anterior maxilla on panoramic radiographs and to compare its performance with that of dental residents. Materials and Methods : One hundred patients with a confirmed diagnosis of NDC (53 men, 47 women; average age, 44.6 +/- 16.5 years), 100 with radicular cysts (49 men, 51 women; average age, 47.5 +/- 16.4 years), and 100 with normal groups (56 men, 44 women; average age, 34.4 +/- 14.6 years) were enrolled in this study. Cases were randomly assigned to the training datasets (80%) and the test dataset (20%). Then, 20% of the training data were randomly assigned as validation data. A learning model was created using a customized DetectNet built in Digits version 5.0 (NVIDIA, Santa Clara, USA). The performance of the deep learning system was assessed and compared with that of two dental residents. Results: The performance of the deep learning system was superior to that of the dental residents except for the recall of radicular cysts. The areas under the curve (AUCs) for NDCs and radicular cysts in the deep learning system were significantly higher than those of the dental residents. The results for the dental residents revealed a significant difference in AUC between NDCs and normal groups. Conclusion: This study showed superior performance in detecting NDCs and radicular cysts and in distinguishing between these lesions and normal groups.

The purpose of this study is to develop two-step deep learning models that can automatically detect implant regions on panoramic radiographs and identify several types of implants. A total of 1,574 panoramic radiographs containing 3675 implants were included. The implant manufacturers were Kyocera, Dentsply Sirona, Straumann, and Nobel Biocare. Model A was created to detect oral implants and identify the manufacturers using You Only Look Once (YOLO) v7. After preparing the image patches that cropped the implant regions detected by model A, model B was created to identify the implant types per manufacturer using EfficientNet. Model A achieved very high performance, with recall of 1.000, precision of 0.979, and F1 score of 0.989. It also had accuracy, recall, precision, and F1 score of 0.98 or higher for the classification of the manufacturers. Model B had high classification metrics above 0.92, exception for Nobel’s class 2 (Parallel). In this study, two-step deep learning models were built to detect implant regions, identify four manufacturers, and identify implant types per manufacturer.

The objectives of this study were to create a mutual conversion system between contrast-enhanced computed tomography (CECT) and non-CECT images using a cycle generative adversarial network (cycleGAN) for the internal jugular region. Image patches were cropped from CT images in 25 patients who underwent both CECT and non-CECT imaging. Using a cycleGAN, synthetic CECT and non-CECT images were generated from original non-CECT and CECT images, respectively. The peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM) were calculated. Visual Turing tests were used to determine whether oral and maxillofacial radiologists could tell the difference between synthetic versus original images, and receiver operating characteristic (ROC) analyses were used to assess the radiologists' performances in discriminating lymph nodes from blood vessels. The PSNR of non-CECT images was higher than that of CECT images, while the SSIM was higher in CECT images. The Visual Turing test showed a higher perceptual quality in CECT images. The area under the ROC curve showed almost perfect performances in synthetic as well as original CECT images. In conclusion, synthetic CECT images created by cycleGAN appeared to have the potential to provide effective information in patients who could not receive contrast enhancement.

The present study aimed to assess the consistencies and performances of deep learning (DL) models in the diagnosis of condylar osteoarthritis (OA) among patients with dentofacial deformities using panoramic temporomandibular joint (TMJ) projection images. A total of 68 TMJs with or without condylar OA in dentofacial deformity patients were tested to verify the consistencies and performances of DL models created using 252 TMJs with or without OA in TMJ disorder and dentofacial deformity patients; these models were used to diagnose OA on conventional panoramic (Con-Pa) images and open (Open-TMJ) and closed (Closed-TMJ) mouth TMJ projection images. The GoogLeNet and VGG-16 networks were used to create the DL models. For comparison, two dental residents with < 1 year of experience interpreting radiographs evaluated the same condyle data that had been used to test the DL models. On Open-TMJ images, the DL models showed moderate to very good consistency, whereas the residents’ demonstrated fair consistency on all images. The areas under the curve (AUCs) of both DL models on Con-Pa (0.84 for GoogLeNet and 0.75 for VGG-16) and Open-TMJ images (0.89 for both models) were significantly higher than the residents’ AUCs (p < 0.01). The AUCs of the DL models on Open-TMJ images (0.89 for both models) were higher than the AUCs on Closed-TMJ images (0.72 for both models). The DL models created in this study could help residents to interpret Con-Pa and Open-TMJ images in the diagnosis of condylar OA.

AimWe aim to evaluate the diagnostic performance of a deep learning (DL) system for determining the three-dimensional contact status between the mandibular third molar and canal on panoramic radiography images. MethodsA total of 800 image patches consisting of 400 patches of low- and high-risk groups, each verified by computed tomography (CT) or cone-beam CT for dental use, were cropped from downloaded panoramic images and input into a DL system. Seven hundred of these patches (350 high-risk and 350 low-risk group patches) were randomly assigned to the training and validation datasets, and 100 (50 high-risk and 50 low-risk group patches) were assigned to the test datasets. Using data augmentation for the training datasets, the training process was carried out twice. Receiver operating characteristic (ROC) analysis was used to compare the performance of two kinds of observers (residents and radiologists) with the same test images. The interclass correlation coefficients (ICCs) were determined to evaluate the diagnostic consistency. ResultsThe area under the ROC curves (AUCs) of the DL model, residents, and radiologists were 0.85, 0.55, and 0.81, respectively. Significant differences were observed between the DL model and residents, and between the residents and radiologists. The ICCs of the DL model, residents, and radiologists were 0.69, 0.19, and 0.54, respectively. ConclusionsThe DL model has potential for use in diagnostic support in the evaluation of the three-dimensional contact status between the mandibular third molar and canal on panoramic images.

Objectives: This study aimed to clarify the performance of magnetic resonance imaging (MRI)-based deep learning classification models in diagnosing temporomandibular joint osteoarthritis (TMJ-OA) and to compare the developed diagnostic assistance with human observers. Methods: The subjects were 118 patients who underwent MRI for examination of TMJ disorders. One hundred condyles with TMJ-OA and 100 condyles without TMJ-OA were enrolled. Deep learning was performed with four networks (ResNet18, EfficientNet b4, Inception v3, and GoogLeNet) using five-fold cross validation. Receiver operating characteristics (ROC) curves were drawn for each model and diagnostic metrics were determined. The performances of the four network models were compared using Kruskal-Wallis tests and post-hoc Scheffe tests, and ROCs between the best model and human were compared using chi-square tests, with p < 0.05 considered significant. Results: ResNet18 had areas under the curves (AUCs) of 0.91-0.93 and accuracy of 0.85-0.88, which were the highest among the four networks. There were significant differences in AUC and accuracy between ResNet and GoogLeNet (p = 0.0264 and p = 0.0418, respectively). The kappa values of the models were large, 0.95 for ResNet and 0.93 for EfficientNet. The experts achieved similar AUC and accuracy values to the ResNet metrics, 0.94 and 0.85, and 0.84 and 0.84, respectively, but with a lower kappa of 0.67. Those of the dental residents showed lower values. There were significant differences in AUCs between ResNet and residents (p < 0.0001) and between experts and residents (p < 0.0001). Conclusions: Using a deep learning model, high performance was confirmed for MRI diagnosis of TMJ-OA.

AimTo evaluate the effects of the combined use of segmentation or detection models on the deep learning (DL) classification performance for cyst-like lesions of the jaws on panoramic radiographs. MethodsThe panoramic radiographs of 536 patients with cyst-like lesions of the jaws and 130 patients without cyst-like lesions were used in this study. The radiographs were arbitrarily assigned to training, validation, and test datasets. The following three DL systems were created: System 1 directly classified cyst-like lesions of the jaws on panoramic radiographs using a VGG-16 convolution neural network (CNN), System 2 combined two CNNs to perform a preceding segmentation (U-Net) and then the classification (VGG-16), and System 3 combined two CNNs to perform a preceding detection (YOLO) and then the classification (VGG-16). The classification performance of three systems was evaluated and compared with that of oral and maxillofacial radiologists. ResultsThe classification performance of System 2 was higher than the other DL systems, demonstrating the efficacy of the combined use of DL segmentation and classification models. System 3 followed it. The radiologists showed similar accuracy with System 2 and higher performance than Systems 1 and 3. The three DL systems and the radiologists all showed higher performance for dentigerous and radicular cysts than for ameloblastoma and odontogenic keratocysts, because of bias in the number of cases between categories even if data were collected at two institutions. ConclusionsThe performance of DL classification of cyst-like lesions of the jaws was improved by the addition of a DL segmentation technique.