Unlike medical computed tomography (CT), dental CT often suffers from severe metal artifacts stemming from high-density materials employed for dental prostheses. Despite the many metal artifact reduction (MAR) methods available for medical CT, those methods do not sufficiently reduce metal artifacts in dental CT images because MAR performance is often compromised by the enamel layer of teeth, whose X-ray attenuation coefficient is not so different from that of prosthetic materials. We propose a deep learning-based metal segmentation method on the projection domain to improve MAR performance in dental CT. We adopted a simplified U-net for metal segmentation on the projection domain without using any information from the metal-artifacts-corrupted CT images. After training the network with the projection data of five patients, we segmented the metal objects on the projection data of other patients using the trained network parameters. With the segmentation results, we corrected the projection data by applying region filling inside the segmented region. We fused two CT images, one from the corrected projection data and the other from the original raw projection data, and then we forward-projected the fused CT image to get the fused projection data. To get the final corrected projection data, we replaced the metal regions in the original projection data with the ones in the fused projection data. To evaluate the efficacy of the proposed segmentation method on MAR, we compared the MAR performance of the proposed segmentation method with a conventional MAR method based on metal segmentation on the CT image domain. For the MAR performance evaluation, we considered the three primary MAR performance metrics: the relative error (REL), the sum of square difference (SSD), and the normalized absolute difference (NAD). The proposed segmentation method improved MAR performances by around 5.7% for REL, 6.8% for SSD, and 8.2% for NAD. The proposed metal segmentation method on the projection domain showed better MAR performance than the conventional segmentation on the CT image domain. We expect that the proposed segmentation method can improve the performance of the existing MAR methods that are based on metal segmentation on the CT image domain.

BackgroundMetal artifacts appearing as streaks and shadows often compromise readability of computed tomography (CT) images. Particularly in a dental CT in which high resolution imaging is crucial for precise preparation of dental implants or orthodontic devices, reduction of metal artifacts is very important. However, metal artifact reduction algorithms developed for a general medical CT may not work well in a dental CT since teeth themselves also have high attenuation coefficients.MethodsTo reduce metal artifacts in dental CT images, we made prior images by weighted summation of two images: one, a streak-reduced image reconstructed from the metal-region-modified projection data, and the other a metal-free image reconstructed from the original projection data followed by metal region deletion. To make the streak-reduced image, we precisely segmented the metal region based on adaptive local thresholding, and then, we modified the metal region on the projection data using linear interpolation. We made forward projection of the prior image to make the prior projection data. We replaced the pixel values at the metal region in the original projection data with the ones taken from the prior projection data, and then, we finally reconstructed images from the replaced projection data. To validate the proposed method, we made computational simulations and also we made experiments on teeth phantoms using a micro-CT. We compared the results with the ones obtained by the fusion prior-based metal artifact reduction (FP-MAR) method.ResultsIn the simulation studies using a bilateral prostheses phantom and a dental phantom, the proposed method showed a performance similar to the FP-MAR method in terms of the edge profile and the structural similarity index when an optimal global threshold was chosen for the FP-MAR method. In the imaging studies of teeth phantoms, the proposed method showed a better performance than the FP-MAR method in reducing the streak artifacts without introducing any contrast anomaly.ConclusionsThe simulation and experimental imaging studies suggest that the proposed method can be used for reducing metal artifacts in dental CT images.

Abstract:We introduce an efficient ring artifact correction method for a cone-beam computed tomography (CT). In the first step, we correct the defective pixels whose values are close to zero or saturated in the projection domain. In the second step, we compute the mean value at each detector element along the view angle in the sinogram to obtain the one-dimensional (1D) mean vector, and we then compute the 1D correction vector by taking inverse of the mean vector. We multiply the correction vector with the sinogram row by row over all view angles. In the third step, we apply a Gaussian filter on the difference image between the original CT image and the corrected CT image obtained in the previous step. The filtered difference image is added to the corrected CT image to compensate the possible contrast anomaly that may appear due to the contrast change in the sinogram after removing stripe artifacts. We applied the proposed method to the projection data acquired by two flat-panel detectors (FPDs) and a silicon-based photon-counting X-ray detector (PCXD). Micro-CT imaging experiments of phantoms and a small animal have shown that the proposed method can greatly reduce ring artifacts regardless of detector types. Despite the great reduction of ring artifacts, the proposed method does not compromise the original spatial resolution and contrast.

The successful development of the image denoising techniques for low-dose computed tomography (LDCT) was largely owing to the public-domain availability of spatially-aligned high- and low-dose CT image pairs. Even though low-dose CT scans are also highly desired in dental imaging, public-domain databases of dental CT image pairs have not been established yet. In this paper, we propose a dental CT image denoising method based on the transfer learning of a generative adversarial network (GAN) from the public-domain CT images. We trained a generative adversarial network with the Wasserstein loss function (WGAN) using 5,100 high- and low-dose medical CT image pairs of human chest and abdomen. For the generative network of GAN, we used the U-net structure of five stages to exploit its high computational efficiency. After training the proposed network, named U-WGAN, we fine-tuned the network with 3,006 dental CT image pairs of two different human skull phantoms. For the high- and low-dose scans of the phantoms, we set the tube current of the dental CT to 10 mA and 4 mA, respectively, with the tube voltage set to 90 kVp in both scans. We applied the trained network to denoising of low-dose dental CT images of dental phantoms and adult humans. The U-net processed images showed over-smoothing effects even though U-net had a good performance in the quantitative metrics. U-WGAN showed similar denoising performance to WGAN, but it reduced the computation time of WGAN by a factor of 10. The fine-tuning procedure in the transfer learning scheme enhanced the network performance in terms of the quantitative metrics, and it also improved visual appearance of the processed images. Even though the number of fine-tuning images was very limited in this study, we think the transfer learning scheme can be a good option for developing deep learning networks for dental CT image denoising.

A small head motion of the patient can compromise the image quality in a dental CT, in which a slow cone-beam scan is adopted. We introduce a retrospective head motion estimation method by which we can estimate the motion waveform from the projection images without employing any external motion monitoring devices. We compute the cross-correlation between every two successive projection images, which results in a sinusoid-like displacement curve over the projection view when there is no patient motion. However, the displacement curve deviates from the sinusoid-like form when patient motion occurs. We develop a method to estimate the motion waveform with a single parameter derived from the displacement curve with aid of image entropy minimization. To verify the motion estimation method, we use a lab-built micro-CT that can emulate major head motions during dental CT scans, such as tilting and nodding, in a controlled way. We find that the estimated motion waveform conforms well to the actual motion waveform. To further verify the motion estimation method, we correct the motion artifacts with the estimated motion waveform. After motion artifact correction, the corrected images look almost identical to the reference images, with structural similarity index values greater than 0.81 in the phantom and rat imaging studies.