Search

This study aimed to evaluate the feasibility of a deep learning method for imaging artifact and noise reduction in coronal reformation of contrast-enhanced chest computed tomography (CT).A total of 19,052 coronal reformatted chest CT images of 110 CT image sets (55 pairs of concordant 16- and 320-row CT image sets) were included and used to train a deep learning algorithm for artifact and noise correction. For internal validation, 4093 coronal reformatted CT images of 25 patients from 16-row CT images underwent correction processing. For external validation, chest CT images of 30 patients (1028 coronal reformatted CT images), acquired in other institutions using different scanners, were subjected to correction processing. For both validations, image quality was compared between original ("CT origin ") and deep learning-based corrected ("CT correct ") CT images. Quantitative analysis for stair-step artifact (coefficient of variance of CT density on coronal reformation), image noise, signal-to-noise ratio, and contrast-to-noise ratio were evaluated. Subjective image quality scores were assigned for image contrast, artifact, and conspicuity of major structures.CT correct showed significantly reduced stair-step artifact (mean coefficient of variance: CT origin 7.35 ± 2.0 vs CT correct 5.17 ± 2.4, P0.001) and image noise and improved signal-to-noise ratio and contrast-to-noise ratio in the aorta, pulmonary artery, and liver, compared with those of CT origin ( P0.01). On subjective analysis, CT correct had higher image contrast, lower artifact, and better conspicuity than CT origin . Most results of the external validation were consistent with those obtained from the internal validation, except for those concerning the pulmonary artery.Deep learning-based artifact correction significantly improved the image quality of coronal reformation chest CT by reducing image noise and artifacts.

Dental cone-beam computed tomography (CBCT) has been increasingly used for dental and maxillofacial imaging. However, the presence of metallic inserts, such as implants, crowns, and dental braces, violates the CT model assumption, which leads to severe metal artifacts in the reconstructed CBCT image, resulting in the degradation of diagnostic performance. In this study, we used deep learning to reduce metal artifacts.The metal artifacts, appearing as streaks and shadows, are nonlocal and highly associated with various factors, including the geometry of metallic inserts, energy-dependent attenuation, and energy spectrum of the incident X-ray beam, making it difficult to learn their complicated structures directly. To provide a step-by-step environment in which deep learning can be trained, we propose an iterative learning approach in which the network at each iteration step learns the correction error caused by the previous network, while enforcing the data fidelity in the projection domain. To generate a realistic paired training dataset, metal-free CBCT scans were collected from patients without metallic inserts, and then simulated metal projection data were added to generate the corresponding metal-corrupted projection data.The feasibility of the proposed method was investigated in clinical metal-affected CBCT scans, as well as simulated metal-affected CBCT scans. The results show that the proposed method significantly reduces metal artifacts while preserving the morphological structures near metallic objects and outperforms direct image domain learning.The proposed fidelity-embedded learning can effectively reduce metal artifacts in dental CBCT compared with direct image domain learning.

Objective The purpose of this study was to evaluate the accuracy of an optical tracking system during reference point localization, measurement, and registration of skull models for navigational maxillary orthognathic surgery. Study Design Accuracy was first evaluated on the basis of the position recording discrepancy at a static point and at 2 points of fixed lengths. Ten reference points were measured on a skull model at 7 different locations, and their measurements were compared with predicted positions by using 4 registration methods. Finally, positional tracking of reference points for simulated maxillary surgery was performed and compared with laser scan data. Results The average linear measurement discrepancy was 0.28 mm, and the mean measurement discrepancy with the 5 registered cranial points was 1.53 mm. The average measurement discrepancy after maxillary surgery was 1.91 mm (for impaction) and 1.56 mm (for advancement). The registration discrepancy in jitter and point registration on the y-axis was significantly greater than on the other axes. Conclusions The optical tracking system seems clinically acceptable for precise tracking of the maxillary position during navigational orthognathic surgery, notwithstanding the chance of greater measurement error on the y-axis.

Retinal swelling due to the accumulation of fluid is associated with the most vision-threatening retinal diseases. Optical coherence tomography (OCT) is the current standard of care in assessing the presence and quantity of retinal fluid and image-guided treatment management. Deep learning methods have made their impact across medical imaging, and many retinal OCT analysis methods have been proposed. However, it is currently not clear how successful they are in interpreting the retinal fluid on OCT, which is due to the lack of standardized benchmarks. To address this, we organized a challenge RETOUCH in conjunction with MICCAI 2017, with eight teams participating. The challenge consisted of two tasks: fluid detection and fluid segmentation. It featured for the first time: all three retinal fluid types, with annotated images provided by two clinical centers, which were acquired with the three most common OCT device vendors from patients with two different retinal diseases. The analysis revealed that in the detection task, the performance on the automated fluid detection was within the inter-grader variability. However, in the segmentation task, fusing the automated methods produced segmentations that were superior to all individual methods, indicating the need for further improvements in the segmentation performance.

Background: We evaluated whether the brain regional features could enhance the prediction of the occurrence of the post-stroke delirium along with consideration of clinical features in acute ischemic stroke patients. Method: We analyzed magnetic resonance imaging and clinical datasets prospectively collected from 1,024 acute ischemic stroke patients (620 men; mean age±SD, 70.2±11.8 years), who were admitted in stroke unit. Delirium occurrence in the patients was initially screened with Confusion Assessment Method and finally diagnosed using the criteria of the Diagnostic and Statistical Manual of Mental Disorders (fifth edition). The prediction performance of the clinical features was evaluated using the logistic regression analysis, and then, compared with the performance obtained after adding of the brain lesional features. Results: Post-stroke delirium occurred in 100 (9.8%) of the total patients. Independent features of the clinical features were old age (odds ratio [OR] for the delirium prediction: 1.07, 95% confidence interval [CI] 1.038-1.103), preexisting dementia (OR 0.423, CI 0.209-0.853) and atrial fibrillation OR 1.599, CI 0.938-2.726), neglect (OR 2.395, CI 1.152-4.976), visual defect (OR 2.209, CI 0.994-4.907), diplopia (OR 2.42, CI 1.021-5.738), dysuria (OR 2.404, 95% CI 1.155-5.000), Foley catheter insertion (OR 0.46, CI 0.24-0.881) and pneumonia (OR 1.954, CI 1.004-3.803). Right-sided hemispheric and mixed acute lesions (p=0.003), lesion locations in occipital lobe (OR 1.814, CI 0.923-3.564), cerebellum (OR 2.875, CI 1.32-6.258) and brainstem (OR 4.057, CI 1.714-9.601), and the presence of the periventricular white matter change (OR 1.321, CI 0.964-1.811) increased the risk for delirium. Sensitivity to predict the delirium of the regression model was enhanced from 10% obtained with the clinical features to 18% by adding the brain lesional features. Area under the curve of the receiver operating characteristics analysis was also enhanced from 0.827 with the clinical features to 0.847 after adding the brain lesions to the clinical features. Conclusion: The present study shows brain lesional characteristics that could enhance the prediction of the post-stroke delirium in acute ischemic stroke patients.

Our purpose in this study is to evaluate the clinical feasibility of deep-learning techniques for F-18 florbetaben (FBB) positron emission tomography (PET) image reconstruction using data acquired in a short time. We reconstructed raw FBB PET data of 294 patients acquired for 20 and 2 min into standard-time scanning PET (PET20m) and short-time scanning PET (PET2m) images. We generated a standard-time scanning PET-like image (sPET20m) from a PET2m image using a deep-learning network. We did qualitative and quantitative analyses to assess whether the sPET20m images were available for clinical applications. In our internal validation, sPET20m images showed substantial improvement on all quality metrics compared with the PET2m images. There was a small mean difference between the standardized uptake value ratios of sPET20m and PET20m images. A Turing test showed that the physician could not distinguish well between generated PET images and real PET images. Three nuclear medicine physicians could interpret the generated PET image and showed high accuracy and agreement. We obtained similar quantitative results by means of temporal and external validations. We can generate interpretable PET images from low-quality PET images because of the short scanning time using deep-learning techniques. Although more clinical validation is needed, we confirmed the possibility that short-scanning protocols with a deep-learning technique can be used for clinical applications.

Magnetic resonance electrical impedance tomography (MREIT) is a high-resolution conductivity imaging method utilizing measured magnetic flux density data induced by externally injected currents. Most MREIT image reconstruction methods including the harmonic algorithm adopt iterative schemes to handle the non-linear relation between conductivity and magnetic flux density. Iterative methods, however, may not guarantee a reliable conductivity reconstruction when the measured magnetic flux density data are contaminated with a significant amount of noise. In this paper, we propose a new image reconstruction method which alleviates the technical difficulties of the iterative harmonic algorithm. It effectively reduces the number of iterations by two at most. To improve the image quality, it incorporates the influence of non-transversal current densities. Providing theoretical observations and details of the proposed algorithm, we present results of numerical simulations and phantom experiments for its va...

Background and Objective: The present study was performed to develop an automated algorithm to measure the carotid stenosis by considering both deep learning and mathematical model using carotid duplex ultrasonography (CDU) images. Methods: We first obtained cine images of CDU from right and left carotid arteries of 18 ischemic stroke patients by continuous moving from supraclavicular to submandibular area. Then, we collected raw axial-CDU images from the cine-CDU images, and then, labelled segmentation of the stenosis caused by atherosclerotic plaque from the vessel wall in the individual axial image by two experts. To develop segmentation algorithm from the axial-CDU images, we first applied a deep learning algorithm to segment vessel lumen from vessel wall. Next, a mathematical algorithm using Gaussian mixture was used to segment atherosclerotic stenosis from the vessel lumen. Dice coefficient was obtained to evaluate whether the segmentation algorithms could accurately segment lumen of carotid artery and predict the stenosis severity measured by the experts using python packages including TensorFlow and scikit-learn on a workstation (Intel i9-7900X CPU, Nvidia Titan Xp GPU and 128G 2400GHz memory). Results: We finally collected total 13,586 raw axial-CDU images from the cine-CDU images of the 18 patients. After application of two steps of segmentation algorithms to the axial-CDU images, accuracy of the algorithm to segment lumen from the carotid vessel wall was mean 0.92 (±standard deviation 0.46) of dice coefficient. And, accuracy to estimate the stenotic area was mean 0.201 (±standard deviation 0.137) of dice coefficient. Conclusions: We proposed an algorithm to automatically quantify the carotid stenosis using two steps of approach. First, a deep learning based-algorithm to segment lumen of carotid artery; second, a mathematical model based-algorithm using Gaussian mixture to segment carotid stenosis from the lumen. Even though we need more studies to increase the accuracy to predict the stenosis, the present prediction algorithms provide a possible tool to automatically measure the severity and regional characteristics of carotid stenosis using cine-CDU images.

Objective To evaluate the diagnostic performance of a deep learning algorithm for the automated detection of developmental dysplasia of the hip (DDH) on anteroposterior (AP) radiographs. Materials and Methods Of 2601 hip AP radiographs, 5076 cropped unilateral hip joint images were used to construct a dataset that was further divided into training (80%), validation (10%), or test sets (10%). Three radiologists were asked to label the hip images as normal or DDH. To investigate the diagnostic performance of the deep learning algorithm, we calculated the receiver operating characteristics (ROC), precision-recall curve (PRC) plots, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) and compared them with the performance of radiologists with different levels of experience. Results The area under the ROC plot generated by the deep learning algorithm and radiologists was 0.988 and 0.988–0.919, respectively. The area under the PRC plot generated by the deep learning algorithm and radiologists was 0.973 and 0.618–0.958, respectively. The sensitivity, specificity, PPV, and NPV of the proposed deep learning algorithm were 98.0, 98.1, 84.5, and 99.8%, respectively. There was no significant difference in the diagnosis of DDH by the algorithm and the radiologist with experience in pediatric radiology (p = 0.180). However, the proposed model showed higher sensitivity, specificity, and PPV, compared to the radiologist without experience in pediatric radiology (p < 0.001). Conclusion The proposed deep learning algorithm provided an accurate diagnosis of DDH on hip radiographs, which was comparable to the diagnosis by an experienced radiologist.

This paper presents a new approach to automatic three-dimensional (3D) cephalometric annotation for diagnosis, surgical planning, and treatment evaluation. There has long been considerable demand for automated cephalometric landmarking, since manual landmarking requires considerable time and experience as well as objectivity and scrupulous error avoidance. Due to the inherent limitation of two-dimensional (2D) cephalometry and the 3D nature of surgical simulation, there is a trend away from current 2D to 3D cephalometry. Deep learning approaches to cephalometric landmarking seem highly promising, but there exist serious difficulties in handling high dimensional 3D CT data, dimension referring to the number of voxels. To address this issue of dimensionality, this paper proposes a shadowed 2D image-based machine learning method which uses multiple shadowed 2D images with various lighting and view directions to capture 3D geometric cues. The proposed method using VGG-net was trained and tested using 2700 shadowed 2D images and corresponding manual landmarkings. Test data evaluation shows that our method achieved an average point-to-point error of 1.5 mm for the seven major landmarks.

Automatic annotation for three-dimensional (3D) cephalometric analysis has been limited by computational complexity and computing performance. The purpose of this study was to evaluate the accuracy of our newly-developed automatic 3D cephalometric annotation system using a deep learning algorithm. Our model system mainly consisted of a 3D convolutional neural network and image data resampling. Discrepancies between the referenced and predicted coordinate values in three axes and in 3D distance were calculated to yield prediction errors of 3.26, 3.18, and 4.81 mm (for three axes) and 7.61 mm (for 3D). Moreover, there was no difference (p > 0.05) among the landmarks of three groups (midsagittal plane, horizontal plane and mandible). Although our 3D convolutional neural network-based annotation system could not achieve the level of accuracy demanded by immediate clinical applications, it can nevertheless serve as an initial approximate guide to landmarks, thus reducing the time needed for annotation.

Dual-energy CT can be represented as the dual-energy equations by decomposing the linear attenuation coefficient of the X-ray scanned object into two material basis functions of photoelectric absorption and Compton scatter. To solve the dual-energy equations, in this paper, we apply the mean-value theorem for integrals and propose a new projection-based iterative algorithm. We discuss the convergence of the proposed algorithm and carry out various numerical simulations for demonstrating its feasibility.

Kirchhoff migration was applied to the fast identification of the shape and location of extended dielectric targets that were completely hidden in an inhomogeneous medium with smooth boundary. This is based on the factorization of the collected Multi-Static Response (MSR) matrix. Numerical simulations demonstrated that the use of Kirchhoff migration made it possible to detect almost all inhomogeneities.

If there are metals located in the X-ray scanned object, a point outside the metals gas its range of projection angle at which projections passing through the point are disturbed by the metals. Roughly speaking, this implies that attenuation information at the point is missing in the blocked projection range. So conventional projection completion MAR algorithms to use the undisturbed projection data on the boundary of the metaltrace is less efficient in reconstructing the attenuation coefficient in detailed parts, in particular, near the metal region. In order to overcome this problem, we propose the algebraic correction technique (ACT) to utilize a prereconstructed interim image of the attenuation coefficient outside the metal region which is obtained by solving a linear system designed to reduce computational costs. The reconstructed interim image of the attenuation coefficient is used as prior information for MAR. Numerical simulations support that the proposed correction technique shows better performance than conventional inpainting techniques such as the total variation and the harmonic inpainting.

In this contribution, we apply MUltiple SIgnal Classification (MUSIC) algorithm to the imaging of extended dielectric inhomogeneities that are completely embedded in an inhomogeneous medium with a smooth boundary. This application is based on the so-called physical factorization of the collected Multi-Static Response (MSR) matrix. Various numerical experiments are presented to demonstrate the pros and cons of the considered technique.

Conductivity imaging based on the current-injection MRI technique has been developed in magnetic resonance electrical impedance tomography. Current injected through a pair of surface electrodes induces a magnetic flux density distribution inside an imaging object, which results in additional magnetic field inhomogeneity. We can extract phase changes related to the current injection and obtain an image of the induced magnetic flux density. Without rotating the object inside the bore, we can measure only one component B(z) of the magnetic flux density B = (B(x), B(y), B(z)). Based on a relation between the internal conductivity distribution and B(z) data subject to multiple current injections, one may reconstruct cross-sectional conductivity images. As the image reconstruction algorithm, we have been using the harmonic B(z) algorithm in numerous experimental studies. Performing conductivity imaging of intact animal and human subjects, we found technical difficulties that originated from the MR signal void phenomena in the local regions of bones, lungs and gas-filled tubular organs. Measured B(z) data inside such a problematic region contain an excessive amount of noise that deteriorates the conductivity image quality. In order to alleviate this technical problem, we applied hybrid methods incorporating ramp-preserving denoising, harmonic inpainting with isotropic diffusion and ROI imaging using the local harmonic B(z) algorithm. These methods allow us to produce conductivity images of intact animals with best achievable quality. We suggest guidelines to choose a hybrid method depending on the overall noise level and existence of distinct problematic regions of MR signal void.

Magnetic resonance electrical impedance tomography (MREIT) is a new bio-imaging modality providing cross-sectional conductivity images from measurements of internal magnetic flux densities produced by externally injected currents. Recent experimental results of postmortem and in vivo imaging of the canine brain demonstrated its feasibility by showing conductivity images with meaningful contrast among different brain tissues. MREIT image reconstructions involve a series of data processing steps such as k-space data handling, phase unwrapping, image segmentation, meshing, modelling, finite element computation, denoising and so on. To facilitate experimental studies, we need a software tool that automates these data processing steps. In this paper, we summarize such an MREIT software package called CoReHA (conductivity reconstructor using harmonic algorithms). Performing imaging experiments of the postmortem canine abdomen, we demonstrate how CoReHA can be utilized in MREIT. The abdomen with a relatively large field of view and various organs imposes new technical challenges when it is chosen as an imaging domain. Summarizing a few improvements in the experimental MREIT technique, we report our first conductivity images of the postmortem canine abdomen. Illustrating reconstructed conductivity images, we discuss how they discern different organs including the kidney, spleen, stomach and small intestine. We elaborate, as an example, that conductivity images of the kidney show clear contrast among cortex, internal medulla, renal pelvis and urethra. We end this paper with a brief discussion on future work using different animal models.

In this paper, we provide a mathematical approach to reconstruct the Compton scatter and photo-electronic coefficients using the dual-energy CT system. The proposed imaging method is based on the mean value theorem to handle the non-linear integration coming from the polychromatic energy based CT scan system. We show a numerical simulation result for the validation of the proposed algorithm

The proposed structure of lasonolide A was synthesized employing radical cyclization reactions of beta-alkoxyacrylates for preparation of the tetrahydropyranyl units A and B, but the spectroscopic data did not match those of the natural product. Both enantiomers of a revised structure featuring 17E,25Z double bonds were synthesized, and the (-)-isomer was found to be the biologically active enantiomer.

In magnetic resonance electrical impedance tomography, among several conductivity image reconstruction algorithms, the harmonic B(z) algorithm has been successfully applied to B(z) data from phantoms and animals. The algorithm is, however, sensitive to measurement noise in B(z) data. Especially, in in vivo animal and human experiments where injection current amplitudes are limited within a few milliampere at most, measured B(z) data tend to have a low SNR. In addition, magnetic resonance (MR) signal void in outer layers of bones and gas-filled organs, for example, produces salt-pepper noise in the MR phase and, consequently, B(z) images. The B(z) images typically present areas of sloped transitions, which can be assimilated to ramps. Conductivity contrasts change ramp slopes in B(z) images and it is critical to preserve positions of those ramps to correctly recover edges in conductivity images. In this paper, we propose a ramp-preserving denoising method utilizing a structure tensor. Using an eigenvalue analysis, we identified local regions of salt-pepper noise. Outside the identified local regions, we applied an anisotropic smoothing to reduce noise while preserving their ramp structures. Inside the local regions of salt-pepper noise, we used an isotropic smoothing. After validating the proposed denoising method through numerical simulations, we applied it to in vivo animal imaging experiments. Both numerical simulation and experimental results show significant improvements in the quality of reconstructed conductivity images.

Magnetic resonance electrical impedance tomography (MREIT) attempts to provide conductivity images of an electrically conducting object with a high spatial resolution. When we inject current into the object, it produces internal distributions of current density J and magnetic flux density B=(Bx,By,Bz). By using a magnetic resonance imaging (MRI) scanner, we can measure Bz data where z is the direction of the main magnetic field of the scanner. Conductivity images are reconstructed based on the relation between the injection current and Bz data. The harmonic Bz algorithm was the first constructive MREIT imaging method and it has been quite successful in previous numerical and experimental studies. Its performance is, however, degraded when the imaging object contains low-conductivity regions such as bones and lungs. To overcome this difficulty, we carefully analyzed the structure of a current density distribution near such problematic regions and proposed a new technique, called the local harmonic Bz algorithm. We first reconstruct conductivity values in local regions with a low conductivity contrast, separated from those problematic regions. Then, the method of characteristics is employed to find conductivity values in the problematic regions. One of the most interesting observations of the new algorithm is that it can provide a scaled conductivity image in a local region without knowing conductivity values outside the region. We present the performance of the new algorithm by using computer simulation methods.

(+)-Lasonolide A was synthesized following the established procedure. (-)-Lasonolide A was found to be the biologically active enantiomer.

The topological derivative-based non-iterative imaging algorithm has demonstrated its applicability in limited-aperture inverse scattering problems. However, this has been confirmed through many experimental simulation results, and the reason behind this applicability has not been satisfactorily explained. In this paper, we identify the mathematical structure and certain properties of topological derivatives for the imaging of two-dimensional crack-like thin penetrable electromagnetic inhomogeneities that are completely embedded in a homogeneous material. To this end, we establish a relationship with an infinite series of Bessel functions of integer order of the first kind. Based on the derived structure, we discover a necessary condition for applying topological derivatives in limited-aperture inverse scattering problems, and thus confirm why topological derivatives can be applied. Furthermore, we analyze the structure of multi-frequency topological derivative, and identify why this improves the single-frequency topological derivative in limited-aperture inverse scattering problems. Various numerical simulations are conducted with noisy data, and the results support the derived structure and exhibit certain properties of single- and multi-frequency topological derivatives.

The goal of magnetic resonance electrical impedance tomography (MREIT) is to produce a tomographic image of a conductivity distribution inside an electrically conducting object. Injecting current into the object, we measure a z-component Bz of an induced magnetic flux density using an MRI scanner. Based on the relation between the conductivity and measured Bz data, we may reconstruct cross-sectional images of the conductivity distribution. In a two-dimensional imaging slice, we can see that conductivity changes along equipotential lines are determined by the measured Bz data. Since the equipotential lines themselves depend nonlinearly on the unknown conductivity distribution, it is not possible to recover the conductivity distribution directly from measured Bz data. Conductivity image reconstruction algorithms such as the harmonic Bz algorithm utilize an iterative procedure to update conductivity values and in each iteration we need to solve an elliptic boundary value problem with a presumed conductivity distribution. This iteration is often troublesome in practice due to excessive amounts of noise in some local regions where weak MR signals are produced. This motivated us to develop a non-iterative reconstruction algorithm which does not depend on a global structure of the conductivity distribution. In this paper, we propose a new MREIT conductivity image reconstruction algorithm, called the non-iterative harmonic Bz algorithm, which provides a conductivity image directly from measured Bz data. From numerical simulations, we found that the new method produces absolute images of a conductivity distribution with a high accuracy by maximizing the use of reliable Bz data. It effectively reduces adverse effects of excessive noise in some local regions of weak MR signals by restricting the influences within them.