V2-Net: An Attention-guided Volumetric Regression Network for Tooth Landmark Localization on CT Images with Metal Artifacts.

For virtual surgical planning in orthognathic surgery, marking tooth landmarks on CT images is an important procedure. However, the manual localization procedure of tooth landmarks is time-consuming, labor-intensive, and requires expert knowledge. Also, direct and automatic tooth landmark localization on CT images is difficult because of the lower resolution and metal artifacts of dental images. The purpose of this study was to propose an attention-guided volumetric regression network (V2-Net) for accurate tooth landmark localization on CT images with metal artifacts and lower resolution. V2-Net has an attention-guided network architecture using a coarse-to-fine-attention mechanism that guided the 3D probability distribution of tooth landmark locations within anatomical structures from the coarse V-Net to the fine V-Net for more focus on tooth landmarks. In addition, we combined attention-guided learning and a 3D attention module with optimal Pseudo Huber loss to improve the localization accuracy. Our results show that the proposed method achieves state-of-the-art accuracy of 0.85 ± 0.40 mm in terms of mean radial error, outperforming previous studies. In ablation studies, we observed that the proposed attention-guided learning and a 3D attention module improved the accuracy of tooth landmark localization in CT images of lower resolution and metal artifacts. Furthermore, our method achieved 97.92% in terms of the success detection rate within the clinically accepted accuracy range of 2.0 mm.

Comparison of 2D, 2.5D, and 3D segmentation networks for maxillary sinuses and lesions in CBCT images.

BACKGROUND: The purpose of this study was to compare the segmentation performances of the 2D, 2.5D, and 3D networks for maxillary sinuses (MSs) and lesions inside the maxillary sinus (MSL) with variations in sizes, shapes, and locations in cone beam CT (CBCT) images under the same constraint of memory capacity. METHODS: The 2D, 2.5D, and 3D networks were compared comprehensively for the segmentation of the MS and MSL in CBCT images under the same constraint of memory capacity. MSLs were obtained by subtracting the prediction of the air region of the maxillary sinus (MSA) from that of the MS. RESULTS: The 2.5D network showed the highest segmentation performances for the MS and MSA compared to the 2D and 3D networks. The performances of the Jaccard coefficient, Dice similarity coefficient, precision, and recall by the 2.5D network of U-net + + reached 0.947, 0.973, 0.974, and 0.971 for the MS, respectively, and 0.787, 0.875, 0.897, and 0.858 for the MSL, respectively. CONCLUSIONS: The 2.5D segmentation network demonstrated superior segmentation performance for various MSLs with an ensemble learning approach of combining the predictions from three orthogonal planes.

Automatic classification of 3D positional relationship between mandibular third molar and inferior alveolar canal using a distance-aware network.

The purpose of this study was to automatically classify the three-dimensional (3D) positional relationship between an impacted mandibular third molar (M3) and the inferior alveolar canal (MC) using a distance-aware network in cone-beam CT (CBCT) images. We developed a network consisting of cascaded stages of segmentation and classification for the buccal-lingual relationship between the M3 and the MC. The M3 and the MC were simultaneously segmented using Dense121 U-Net in the segmentation stage, and their buccal-lingual relationship was automatically classified using a 3D distance-aware network with the multichannel inputs of the original CBCT image and the signed distance map (SDM) generated from the segmentation in the classification stage. The Dense121 U-Net achieved the highest average precision of 0.87, 0.96, and 0.94 in the segmentation of the M3, the MC, and both together, respectively. The 3D distance-aware classification network of the Dense121 U-Net with the input of both the CBCT image and the SDM showed the highest performance of accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve, each of which had a value of 1.00. The SDM generated from the segmentation mask significantly contributed to increasing the accuracy of the classification network. The proposed distance-aware network demonstrated high accuracy in the automatic classification of the 3D positional relationship between the M3 and the MC by learning anatomical and geometrical information from the CBCT images.

Ceph-Net: automatic detection of cephalometric landmarks on scanned lateral cephalograms from children and adolescents using an attention-based stacked regression network.

BACKGROUND: The success of cephalometric analysis depends on the accurate detection of cephalometric landmarks on scanned lateral cephalograms. However, manual cephalometric analysis is time-consuming and can cause inter- and intra-observer variability. The purpose of this study was to automatically detect cephalometric landmarks on scanned lateral cephalograms with low contrast and resolution using an attention-based stacked regression network (Ceph-Net). METHODS: The main body of Ceph-Net compromised stacked fully convolutional networks (FCN) which progressively refined the detection of cephalometric landmarks on each FCN. By embedding dual attention and multi-path convolution modules in Ceph-Net, the network learned local and global context and semantic relationships between cephalometric landmarks. Additionally, the intermediate deep supervision in each FCN further boosted the training stability and the detection performance of cephalometric landmarks. RESULTS: Ceph-Net showed a superior detection performance in mean radial error and successful detection rate, including accuracy improvements in cephalometric landmark detection located in low-contrast soft tissues compared with other detection networks. Moreover, Ceph-Net presented superior detection performance on the test dataset split by age from 8 to 16 years old. CONCLUSIONS: Ceph-Net demonstrated an automatic and superior detection of cephalometric landmarks by successfully learning local and global context and semantic relationships between cephalometric landmarks in scanned lateral cephalograms with low contrast and resolutions.

SinusC-Net for automatic classification of surgical plans for maxillary sinus augmentation using a 3D distance-guided network.

The objective of this study was to automatically classify surgical plans for maxillary sinus floor augmentation in implant placement at the maxillary posterior edentulous region using a 3D distance-guided network on CBCT images. We applied a modified ABC classification method consisting of five surgical approaches for the deep learning model. The proposed deep learning model (SinusC-Net) consisted of two stages of detection and classification according to the modified classification method. In detection, five landmarks on CBCT images were automatically detected using a volumetric regression network; in classification, the CBCT images were automatically classified as to the five surgical approaches using a 3D distance-guided network. The mean MRE for landmark detection was 0.87 mm, and SDR for 2 mm or lower, 95.47%. The mean accuracy, sensitivity, specificity, and AUC for classification by the SinusC-Net were 0.97, 0.92, 0.98, and 0.95, respectively. The deep learning model using 3D distance-guidance demonstrated accurate detection of 3D anatomical landmarks, and automatic and accurate classification of surgical approaches for sinus floor augmentation in implant placement at the maxillary posterior edentulous region.

Validation of bone mineral density measurement using quantitative CBCT image based on deep learning.

The bone mineral density (BMD) measurement is a direct method of estimating human bone mass for diagnosing osteoporosis, and performed to objectively evaluate bone quality before implant surgery in dental clinics. The objective of this study was to validate the accuracy and reliability of BMD measurements made using quantitative cone-beam CT (CBCT) image based on deep learning by applying the method to clinical data from actual patients. Datasets containing 7500 pairs of CT and CBCT axial slice images from 30 patients were used to train a previously developed deep-learning model (QCBCT-NET). We selected 36 volumes of interest in the CBCT images for each patient in the bone regions of potential implants sites on the maxilla and mandible. We compared the BMDs shown in the quantitative CBCT (QCBCT) images with those in the conventional CBCT (CAL_CBCT) images at the various bone sites of interest across the entire field of view (FOV) using the performance metrics of the MAE, RMSE, MAPE (mean absolute percentage error), R2 (coefficient of determination), and SEE (standard error of estimation). Compared with the ground truth (QCT) images, the accuracy of the BMD measurements from the QCBCT images showed an RMSE of 83.41 mg/cm3, MAE of 67.94 mg/cm3, and MAPE of 8.32% across all the bone sites of interest, whereas for the CAL_CBCT images, those values were 491.15 mg/cm3, 460.52 mg/cm3, and 54.29%, respectively. The linear regression between the QCBCT and QCT images showed a slope of 1.00 and a R2 of 0.85, whereas for the CAL_CBCT images, those values were 0.32 and 0.24, respectively. The overall SEE between the QCBCT images and QCT images was 81.06 mg/cm3, whereas the SEE for the CAL_CBCT images was 109.32 mg/cm3. The QCBCT images thus showed better accuracy, linearity, and uniformity than the CAL_CBCT images across the entire FOV. The BMD measurements from the quantitative CBCT images showed high accuracy, linearity, and uniformity regardless of the relative geometric positions of the bone in the potential implant site. When applied to actual patient CBCT images, the CBCT-based quantitative BMD measurement based on deep learning demonstrated high accuracy and reliability across the entire FOV.

Structure-preserving quality improvement of cone beam CT images using contrastive learning.

Cone-beam CT (CBCT) is widely used in dental clinics but exhibits limitations in assessing soft tissue pathology because of its lack of contrast resolution and low Hounsfield Units (HU) quantification accuracy. We aimed to increase the image quality and HU accuracy of CBCTs while preserving anatomical structures. We generated CT-like images from CBCT images using a patchwise contrastive learning-based GAN model. Our model was trained on unpaired CT and CBCT datasets with the novel combination of losses and the feature extractor pretrained on our training dataset. We evaluated the quality of the images generated by our model in terms of Fréchet inception distance (FID), peak signal-to-noise ratio (PSNR), mean absolute error (MAE), and root mean square error (RMSE). Additionally, the structure preservation performance was assessed by the structure score. As a result, the generated CT-like images by our model were significantly superior to those generated by various baseline models in terms of FID, PSNR, MAE, RMSE, and structure score. Therefore, we demonstrated that our model provided the complementary benefits of preserving the anatomical structures of the input CBCT images and improving the image quality to be similar to those of CT images.

Automatic diagnosis for cysts and tumors of both jaws on panoramic radiographs using a deep convolution neural network.

OBJECTIVES: The purpose of this study was to automatically diagnose odontogenic cysts and tumors of both jaws on panoramic radiographs using deep learning. We proposed a novel framework of deep convolution neural network (CNN) with data augmentation for detection and classification of the multiple diseases. METHODS: We developed a deep CNN modified from YOLOv3 for detecting and classifying odontogenic cysts and tumors of both jaws. Our data set of 1282 panoramic radiographs comprised 350 dentigerous cysts (DCs), 302 periapical cysts (PCs), 300 odontogenic keratocysts (OKCs), 230 ameloblastomas (ABs), and 100 normal jaws with no disease. In addition, the number of radiographs was augmented 12-fold by flip, rotation, and intensity changes. We evaluated the classification performance of the developed CNN by calculating sensitivity, specificity, accuracy, and area under the curve (AUC) for diseases of both jaws. RESULTS: The overall classification performance for the diseases improved from 78.2% sensitivity, 93.9% specificity,91.3% accuracy, and 0.86 AUC using the CNN with unaugmented data set to 88.9% sensitivity, 97.2% specificity, 95.6% accuracy, and 0.94 AUC using the CNN with augmented data set. CNN using augmented data set had the following sensitivities, specificities, accuracies, and AUCs: 91.4%, 99.2%, 97.8%, and 0.96 for DCs, 82.8%, 99.2%, 96.2%, and 0.92 for PCs, 98.4%,92.3%,94.0%, and 0.97 for OKCs, 71.7%, 100%, 94.3%, and 0.86 for ABs, and 100.0%, 95.1%, 96.0%, and 0.97 for normal jaws, respectively. CONCLUSION: The CNN method we developed for automatically diagnosing odontogenic cysts and tumors of both jaws on panoramic radiographs using data augmentation showed high sensitivity, specificity, accuracy, and AUC despite the limited number of panoramic images involved.

Quantitative measurement of peri-implant bone defects using optical coherence tomography.

PURPOSE: The purpose of this study was to visualize and identify peri-implant bone defects in optical coherence tomography (OCT) images and to obtain quantitative measurements of the defect depth. METHODS: Dehiscence defects were intentionally formed in porcine mandibles and implants were simultaneously placed without flap elevation. Only the threads of the fixture could be seen at the bone defect site in the OCT images, so the depth of the peri-implant bone defect could be measured through the length of the visible threads. To analyze the reliability of the OCT measurements, the flaps were elevated and the depth of the dehiscence defects was measured with a digital caliper. RESULTS: The average defect depth measured by a digital caliper was 4.88±1.28 mm, and the corresponding OCT measurement was 5.11±1.33 mm. Very thin bone areas that were sufficiently transparent in the coronal portion were penetrated by the optical beam in OCT imaging and regarded as bone loss. The intraclass correlation coefficient between the 2 methods was high, with a 95% confidence interval (CI) close to 1. In the Bland-Altman analysis, most measured values were within the threshold of the 95% CI, suggesting close agreement of the OCT measurements with the caliper measurements. CONCLUSIONS: OCT images can be used to visualize the peri-implant bone level and to identify bone defects. The potential of quantitative non-invasive measurements of the amount of bone loss was also confirmed.

Improved accuracy in periodontal pocket depth measurement using optical coherence tomography.

PURPOSE: The purpose of this study was to examine whether periodontal pocket could be satisfactorily visualized by optical coherence tomography (OCT) and to suggest quantitative methods for measuring periodontal pocket depth. METHODS: We acquired OCT images of periodontal pockets in a porcine model and determined the actual axial resolution for measuring the exact periodontal pocket depth using a calibration method. Quantitative measurements of periodontal pockets were performed by real axial resolution and compared with the results from manual periodontal probing. RESULTS: The average periodontal pocket depth measured by OCT was 3.10±0.15 mm, 4.11±0.17 mm, 5.09±0.17 mm, and 6.05±0.21 mm for each periodontal pocket model, respectively. These values were similar to those obtained by manual periodontal probing. CONCLUSIONS: OCT was able to visualize periodontal pockets and show attachment loss. By calculating the calibration factor to determine the accurate axial resolution, quantitative standards for measuring periodontal pocket depth can be established regardless of the position of periodontal pocket in the OCT image.

Automatic detection of tooth cracks in optical coherence tomography images.

PURPOSE: The aims of the present study were to compare the image quality and visibility of tooth cracks between conventional methods and swept-source optical coherence tomography (SS-OCT) and to develop an automatic detection technique for tooth cracks by SS-OCT imaging. METHODS: We evaluated SS-OCT with a near-infrared wavelength centered at 1,310 nm over a spectral bandwidth of 100 nm at a rate of 50 kHz as a new diagnostic tool for the detection of tooth cracks. The reliability of the SS-OCT images was verified by comparing the crack lines with those detected using conventional methods. After performing preprocessing of the obtained SS-OCT images to emphasize cracks, an algorithm was developed and verified to detect tooth cracks automatically. RESULTS: The detection capability of SS-OCT was superior or comparable to that of trans-illumination, which did not discriminate among the cracks according to depth. Other conventional methods for the detection of tooth cracks did not sense initial cracks with a width of less than 100 µm. However, SS-OCT detected cracks of all sizes, ranging from craze lines to split teeth, and the crack lines were automatically detected in images using the Hough transform. CONCLUSIONS: We were able to distinguish structural cracks, craze lines, and split lines in tooth cracks using SS-OCT images, and to automatically detect the position of various cracks in the OCT images. Therefore, the detection capability of SS-OCT images provides a useful diagnostic tool for cracked tooth syndrome.

The relationship between dental implant stability and trabecular bone structure using cone-beam computed tomography.

PURPOSE: The objective of this study was to investigate the relationships between primary implant stability as measured by impact response frequency and the structural parameters of trabecular bone using cone-beam computed tomography(CBCT), excluding the effect of cortical bone thickness. METHODS: We measured the impact response of a dental implant placed into swine bone specimens composed of only trabecular bone without the cortical bone layer using an inductive sensor. The peak frequency of the impact response spectrum was determined as an implant stability criterion (SPF). The 3D microstructural parameters were calculated from CT images of the bone specimens obtained using both micro-CT and CBCT. RESULTS: SPF had significant positive correlations with trabecular bone structural parameters (BV/TV, BV, BS, BSD, Tb.Th, Tb.N, FD, and BS/BV) (P<0.01) while SPF demonstrated significant negative correlations with other microstructural parameters (Tb.Sp, Tb.Pf, and SMI) using micro-CT and CBCT (P<0.01). CONCLUSIONS: There was an increase in implant stability prediction by combining BV/TV and SMI in the stepwise forward regression analysis. Bone with high volume density and low surface density shows high implant stability. Well-connected thick bone with small marrow spaces also shows high implant stability. The combination of bone density and architectural parameters measured using CBCT can predict the implant stability more accurately than the density alone in clinical diagnoses.

Relationship between physical factors and subjective image quality of cone-beam computed tomography images according to diagnostic task.

OBJECTIVE: This study was designed to investigate the relationship between physical factors and the subjective quality of cone beam computed tomography (CBCT) images used for different diagnostic tasks. STUDY DESIGN: CBCT images of a real skull phantom and a SedentexCT IQ phantom were acquired under different exposure conditions (one Dinnova3 CBCT scanner, 60-110 kV and 4-10 mA). Radiologists evaluated subjective image quality of real skull phantom images for each diagnostic task. On the basis of the evaluation results, the images were classified into two groups: acceptable and unacceptable. The modulation transfer function (MTF), contrast-to-noise ratio (CNR), and image uniformity were measured using the SedentexCT IQ phantom images. The differences in physical factors were evaluated. RESULTS: MTF and CNR values showed statistical differences in image quality in two groups with regard to all diagnostic tasks. In the maxilla, MTF and CNR values showed no significant differences between periapical diagnosis and implant planning in the acceptable groups. Higher MTF and CNR values were required in the periapical diagnosis compared with the implant planning of the mandible. CONCLUSIONS: This study proved that MTF and CNR values have a significant association with subjective image quality. The diagnostic task should be considered in evaluation of CBCT image quality.

Material classification of multi-energy CT images using multiple discriminant analysis.

Energy resolved photon-counting detectors could achieve more than one spectral measurement. The goal of this study is to investigate, with experiment, the ability to decompose five materials using energy discriminating detectors and multiple discriminant analysis (MDA). A small field-of-view multi-energy CT system was built. Linear attenuation coefficient was considered as features of multiple energy CT. MDA was used to decompose five materials with six measurements of the energy dependent linear attenuation coefficients. The results of the experimental study showed that a CT system based on CdTe detectors with MDA can be used to decompose five materials.

Radiation dose reduction in CBCT imaging using K-edge filtering and energy weighting.

This paper presents K-edge filtering and energy weighting methods which enhance the contrast with less radiation does. Usually, energy weighting methods are used with photon-counting detector based CT for each energy bin data obtained to enhance the quality of image. However, we used these methods combine with K-edge filtering in energy-integrating detector. Using K-edge filtering, different energy bin data for energy weighting methods were obtained, and then energy weighting factors were calculated to enhance the contrast of image. We report an evaluation of the contrast-to-noise ratio (CNR) of reconstructed image with and without these two methods. This evaluation was proceeded with two phantoms; one is the phantom created personally, and the other is Sendentexct IQ dental CBCT (SENDENTEXCT, EU). As for the phantom created personally, the CNR of images reconstructed with these methods were increased than CNR of standard images. It was seen that 31% to 81% in each energy weighting method for optimizing each material (cortical bone, inner bone, soft tissue, iodine (18.5 g/l), iodine (37 g/l)). In conclusion, we can enhance the contrast of CT images with less radiation dose using K-edge filtering and energy weighting method.