Deep convolutional neural networks (CNNs) are advancing rapidly in medical research, demonstrating promising results in diagnosis and prediction within radiology and pathology. This study evaluates the efficacy of deep learning algorithms for detecting and diagnosing dental caries using cone-beam computed tomography (CBCT) with the Mask R-CNN architecture while comparing various hyperparameters to enhance detection. A total of 2,128 CBCT images were divided into training and validation and test datasets in a 7:1:1 ratio. For the verification of tooth recognition, the data from the validation set were randomly selected for analysis. Three groups of Mask R-CNN networks were compared: A scratch-trained baseline using randomly initialized weights (R group); A transfer learning approach with models pre-trained on COCO for object detection (C group); A variant pre-trained on ImageNetfor for object detection (I group). All configurations maintained identical hyperparameter settings to ensure fair comparison. The deep learning model used ResNet-50 as the backbone network and was trained to 300epoch respectively. We assessed training loss, detection and training times, diagnostic accuracy, specificity, positive and negative predictive values, and coverage precision to compare performance across the groups. Transfer learning significantly reduced training times compared to non-transfer learning approach (p < 0.05). The average detection time for group R was 0.269 ± 0.176 s, whereas groups I (0.323 ± 0.196 s) and C (0.346 ± 0.195 s) exhibited significantly longer detection times (p < 0.05). C-group, trained for 200 epochs, achieved a mean average precision (mAP) of 81.095, outperforming all other groups. The mAP for caries recognition in group R, trained for 300 epochs, was 53.328, with detection times under 0.5 s. Overall, C-group demonstrated significantly higher average precision across all epochs (100, 200, and 300) (p < 0.05). Neural networks pre-trained with COCO transfer learning exhibit superior annotation accuracy compared to those pre-trained with ImageNet. This suggests that COCO’s diverse and richly annotated images offer more relevant features for detecting dental structures and carious lesions. Furthermore, employing ResNet-50 as the backbone architecture enhances the detection of teeth and carious regions, achieving significant improvements with just 200 training epochs, potentially increasing the efficiency of clinical image interpretation.
Medical titanium alloys are widely used in surgery, orthopedics, stomatology and other medical specialties because of their good biocompatibility. In traditional rigid internal fixation applications, titanium alloy strips or plates must be bent to fit the supported surface. Currently, the common practice is to bend titanium alloy bars in three degrees of freedom manually. However, it is difficult to ensure bending accuracy and achieve the best shape. In this study, we introduce a forming robot for internally fixed titanium alloy strips (FRIFTAS). The forming robot is a device that automatically reshapes the titanium strip with various specifications according to medical needs. Here, the design of mechanical and electrical systems and the development of the overall system are described to illustrate how a FRIFTAS is structured and designed. Three bending experimental tests are conducted. In the bending experiments, the robot bends an initial strip on the roll, pitch and yaw directions independently. The results show that the robot can bend a section of the alloy strip at the desired angle. Then, an overall reshaped titanium alloy strip experiment is discussed. The results show that the titanium-strip-forming robot is capable of automatically reshaping an internally fixed titanium strip. The proposed robot can perform the numerically controlled bending of the medical titanium strip according to physicians’ personalized requirements for surgery to improve the accuracy and efficiency of preoperative preparation and provide a better postoperative appearance and more effectively functioning treatment scheme for patients.
Tooth drilling is an essential dental procedure, but the varying mechanical properties of enamel and dentin make it difficult for robotic systems to maintain a constant drilling force. Fluctuations in force can increase the risk of tissue damage. This study compares four control strategies for position-controlled robotic arms to determine the best method for constant-force dental drilling under variable tooth hardness. A position-controlled arm with a six-axis force sensor was used for two experiments: (1) static tooth constant-force tracking and (2) dynamic tooth drilling. The target force was set at 5 N. The four control strategies tested were: Original Motion control (OM), TCP/CAN Direct-Drive control (TDDM), Natural Logarithm Function Extrapolation control (NLFE), and Force Adaptive Model-predictive Embedded control (FAME). Performance was measured using dynamic response indices (rise time, settling time, overshoot) and steady-state stability metrics (root mean square error, mean absolute error, peak-to-peak force fluctuation). All four strategies achieved basic force-tracking capability; however, their stability and disturbance resistance differed significantly. In static tests, FAME demonstrated the best stability with the smallest overshoot (13.28%) and the least difference between root mean square error and mean absolute error, indicating balanced force output. In dynamic drilling conditions with vibration disturbances, FAME achieved the shortest rise time (0.76 s) and maintained force within the predefined clinical tolerance range (5 ± 1.1 N) after convergence. In contrast, OM and NLFE had larger oscillations, while TDDM had frequent fluctuations. Overall, FAME showed the best stability, convergence, and resistance to disturbances in both static and dynamic tooth drilling. This approach enhances constant-force control under varying tooth hardness, making it a promising solution for safe and precise robotic dental procedures. FAME may improve force stability during robotic dental drilling, thereby reducing force fluctuation and supporting safer future clinical applications.
Background: The outbreak of COVID-19 has been defined by the World Health Organization as a pandemic, and containment depends on traditional public health measures. However, the explosive growth of the number of infected cases in a short period of time has caused tremendous pressure on medical systems. Adequate isolation facilities are essential to control outbreaks, so this study aims to quickly estimate the demand and number of isolation beds. Methods: We established a discrete simulation model for epidemiology. By adjusting or fitting necessary epidemic parameters, the effects of the following indicators on the development of the epidemic and the occupation of medical resources were explained: (1) incubation period, (2) response speed and detection capacity of the hospital, (3) disease healing time, and (4) population mobility. Finally, a method for predicting the number of isolation beds was summarized through multiple linear regression. This is a city level model that simulates the epidemic situation from the perspective of population mobility. Results: Through simulation, we show that the incubation period, response speed and detection capacity of the hospital, disease healing time, degree of population mobility, and infectivity of cured patients have different effects on the infectivity, scale, and duration of the epidemic. Among them, (1) incubation period, (2) response speed and detection capacity of the hospital, (3) disease healing time, and (4) population mobility have a significant impact on the demand and number of isolation beds (P <0.05), which agrees with the following regression equation: N = P × (−0.273 + 0.009I + 0.234M + 0.012T1 + 0.015T2) × (1 + V).
Mandibular tumors and radical oral cancer surgery often cause bone dysmorphia and defects. Most patients present with noticeable mandibular deformations, and doctors often have difficulty determining their exact mandibular morphology. In this study, a deep convolutional generative adversarial network (DCGAN) called CTGAN is proposed to complete 3D mandibular cone beam computed tomography (CBCT) data from CT data. After extensive training CTGAN was tested on 6 mandibular tumor cases, resulting in 3D virtual mandibular completion. We found that CTGAN can generate mandibles with different levels and rich morphology, including positional and angular changes and local patterns. The completion results are shown as tomographic images combining generated and natural areas. The 3D generated mandibles have the anatomical morphology of the real mandibles and transition smoothly to the portions without disease, showing that CTGAN constructs mandibles with the expected patient characteristics and is suitable for mandibular morphological completion. The presented modeling principles can be applied to other areas for 3D morphological completion from medical images. Clinical trial registration: This study is not a clinical trial. Patient data were only used for testing in a virtual environment. The use of digital data used in this study was ethically approved.