The approximity of the inferior alveolar nerve (IAN) to the roots of lower third molars (M3) is a risk factor for the occurrence of nerve damage and subsequent sensory disturbances of the lower lip and chin following the removal of third molars. To assess this risk, the identification of M3 and IAN on dental panoramic radiographs (OPG) is mandatory. In this study, we developed and validated an automated approach, based on deep-learning, to detect and segment the M3 and IAN on OPGs. As a reference, M3s and IAN were segmented manually on 81 OPGs. A deep-learning approach based on U-net was applied on the reference data to train the convolutional neural network (CNN) in the detection and segmentation of the M3 and IAN. Subsequently, the trained U-net was applied onto the original OPGs to detect and segment both structures. Dice-coefficients were calculated to quantify the degree of similarity between the manually and automatically segmented M3s and IAN. The mean dice-coefficients for M3s and IAN were 0.947 ± 0.033 and 0.847 ± 0.099, respectively. Deep-learning is an encouraging approach to segment anatomical structures and later on in clinical decision making, though further enhancement of the algorithm is advised to improve the accuracy.

The objective of this study is to assess the classification accuracy of dental caries on panoramic radiographs using deep-learning algorithms. A convolutional neural network (CNN) was trained on a reference data set consisted of 400 cropped panoramic images in the classification of carious lesions in mandibular and maxillary third molars, based on the CNN MobileNet V2. For this pilot study, the trained MobileNet V2 was applied on a test set consisting of 100 cropped PR(s). The classification accuracy and the area-under-the-curve (AUC) were calculated. The proposed method achieved an accuracy of 0.87, a sensitivity of 0.86, a specificity of 0.88 and an AUC of 0.90 for the classification of carious lesions of third molars on PR(s). A high accuracy was achieved in caries classification in third molars based on the MobileNet V2 algorithm as presented. This is beneficial for the further development of a deep-learning based automated third molar removal assessment in future.

The purpose of this study was to present and validate an innovative semi-automatic approach to quantify the accuracy of the surgical outcome in relation to 3D virtual orthognathic planning among patients who underwent bimaxillary surgery. For the validation of this new semi-automatic approach, CBCT scans of ten patients who underwent bimaxillary surgery were acquired pre-operatively. Individualized 3D virtual operation plans were made for all patients prior to surgery. During surgery, the maxillary and mandibular segments were positioned as planned by using 3D milled interocclusal wafers. Consequently, post-operative CBCT scan were acquired. The 3D rendered pre- and postoperative virtual head models were aligned by voxel-based registration upon the anterior cranial base. To calculate the discrepancies between the 3D planning and the actual surgical outcome, the 3D planned maxillary and mandibular segments were segmented and superimposed upon the postoperative maxillary and mandibular segments. The translation matrices obtained from this registration process were translated into translational and rotational discrepancies between the 3D planning and the surgical outcome, by using the newly developed tool, the OrthoGnathicAnalyser. To evaluate the reproducibility of this method, the process was performed by two independent observers multiple times. Low intra-observer and inter-observer variations in measurement error (mean error < 0.25 mm) and high intraclass correlation coefficients (> 0.97) were found, supportive of the observer independent character of the OrthoGnathicAnalyser. The pitch of the maxilla and mandible showed the highest discrepancy between the 3D planning and the postoperative results, 2.72° and 2.75° respectively. This novel method provides a reproducible tool for the evaluation of bimaxillary surgery, making it possible to compare larger patient groups in an objective and time-efficient manner in order to optimize the current workflow in orthognathic surgery.

The implementation of augmented reality (AR) in image-guided surgery (IGS) can improve surgical interventions by presenting the image data directly on the patient at the correct position and in the actual orientation. This approach can resolve the switching focus problem, which occurs in conventional IGS systems when the surgeon has to look away from the operation field to consult the image data on a 2-dimensional screen. The Microsoft HoloLens, a head-mounted AR display, was combined with an optical navigation system to create an AR-based IGS system. Experiments were performed on a phantom model to determine the accuracy of the complete system and to evaluate the effect of adding AR. The results demonstrated a mean Euclidean distance of 2.3 mm with a maximum error of 3.5 mm for the complete system. Adding AR visualization to a conventional system increased the mean error by 1.6 mm. The introduction of AR in IGS was promising. The presented system provided a solution for the switching focus problem and created a more intuitive guidance system. With a further reduction in the error and more research to optimize the visualization, many surgical applications could benefit from the advantages of AR guidance.

Quantitative analysis of the volume and shape of the temporomandibular joint (TMJ) using cone-beam computed tomography (CBCT) requires accurate segmentation of the mandibular condyles and the glenoid fossae. This study aimed to develop and validate an automated segmentation tool based on a deep learning algorithm for accurate 3D reconstruction of the TMJ. A three-step deep-learning approach based on a 3D U-net was developed to segment the condyles and glenoid fossae on CBCT datasets. Three 3D U-Nets were utilized for region of interest (ROI) determination, bone segmentation, and TMJ classification. The AI-based algorithm was trained and validated on 154 manually segmented CBCT images. Two independent observers and the AI algorithm segmented the TMJs of a test set of 8 CBCTs. The time required for the segmentation and accuracy metrics (intersection of union, DICE, etc.) was calculated to quantify the degree of similarity between the manual segmentations (ground truth) and the performances of the AI models. The AI segmentation achieved an intersection over union (IoU) of 0.955 and 0.935 for the condyles and glenoid fossa, respectively. The IoU of the two independent observers for manual condyle segmentation were 0.895 and 0.928, respectively (p<0.05). The mean time required for the AI segmentation was 3.6 s (SD 0.9), whereas the two observers needed 378.9 s (SD 204.9) and 571.6 s (SD 257.4), respectively (p<0.001). The AI-based automated segmentation tool segmented the mandibular condyles and glenoid fossae with high accuracy, speed, and consistency. Potential limited robustness and generalizability are risks that cannot be ruled out, as the algorithms were trained on scans from orthognathic surgery patients derived from just one type of CBCT scanner. The incorporation of the AI-based segmentation tool into diagnostic software could facilitate 3D qualitative and quantitative analysis of TMJs in a clinical setting, particularly for the diagnosis of TMJ disorders and longitudinal follow-up.