Purpose Convolutional neural network (CNN)‐based image denoising techniques have shown promising results in low‐dose CT denoising. However, CNN often introduces blurring in denoised images when trained with a widely used pixel‐level loss function. Perceptual loss and adversarial loss have been proposed recently to further improve the image denoising performance. In this paper, we investigate the effect of different loss functions on image denoising performance using task‐based image quality assessment methods for various signals and dose levels. Methods We used a modified version of U‐net that was effective at reducing the correlated noise in CT images. The loss functions used for comparison were two pixel‐level losses (i.e., the mean‐squared error and the mean absolute error), Visual Geometry Group network‐based perceptual loss (VGG loss), adversarial loss used to train the Wasserstein generative adversarial network with gradient penalty (WGAN‐GP), and their weighted summation. Each image denoising method was applied to reconstructed images and sinogram images independently and validated using the extended cardiac‐torso (XCAT) simulation and Mayo Clinic datasets. In the XCAT simulation, we generated fan‐beam CT datasets with four different dose levels (25%, 50%, 75%, and 100% of a normal‐dose level) using 10 XCAT phantoms and inserted signals in a test set. The signals had two different shapes (spherical and spiculated), sizes (4 and 12 mm), and contrast levels (60 and 160 HU). To evaluate signal detectability, we used a detection task SNR (tSNR) calculated from a non‐prewhitening model observer with an eye filter. We also measured the noise power spectrum (NPS) and modulation transfer function (MTF) to compare the noise and signal transfer properties. Results Compared to CNNs without VGG loss, VGG‐loss‐based CNNs achieved a more similar tSNR to that of the normal‐dose CT for all signals at different dose levels except for a small signal at the 25% dose level. For a low‐contrast signal at 25% or 50% dose, adding other losses to the VGG loss showed more improved performance than only using VGG loss. The NPS shapes from VGG‐loss‐based CNN closely matched that of normal‐dose CT images while CNN without VGG loss overly reduced the mid‐high‐frequency noise power at all dose levels. MTF also showed VGG‐loss‐based CNN with better‐preserved high resolution for all dose and contrast levels. It is also observed that additional WGAN‐GP loss helps improve the noise and signal transfer properties of VGG‐loss‐based CNN. Conclusions The evaluation results using tSNR, NPS, and MTF indicate that VGG‐loss‐based CNNs are more effective than those without VGG loss for natural denoising of low‐dose images and WGAN‐GP loss improves the denoising performance of VGG‐loss‐based CNNs, which corresponds with the qualitative evaluation.
The noise power spectrum ͑NPS͒ is a useful metric for understanding the noise content in images. To examine some unique properties of the NPS of fan beam CT, the authors derived an analytical expression for the NPS of fan beam CT and validated it with computer simulations. The nonstationary noise behavior of fan beam CT was examined by analyzing local regions and the entire field-of-view ͑FOV͒. This was performed for cases with uniform as well as nonuniform noise across the detector cells and across views. The simulated NPS from the entire FOV and local regions showed good agreement with the analytically derived NPS. The analysis shows that whereas the NPS of a large FOV in parallel beam CT ͑using a ramp filter͒ is proportional to frequency, the NPS with direct fan beam FBP reconstruction shows a high frequency roll off. Even in small regions, the fan beam NPS can show a sharp transition ͑discontinuity͒ at high frequencies. These effects are due to the variable magnification and therefore are more pronounced as the fan angle increases. For cases with nonuniform noise, the NPS can show the directional dependence and additional effects.
In-situ asphalt mixture density is critically important to the performance of flexible airport pavements: density that is too high, or too low, may cause early pavement distresses.Traditionally, two methods have been commonly used for in-situ asphalt mixture density measurement: laboratory testing on field-extracted cores and in-situ nuclear gauge testing.However, both these methods have limitations. The coring method damages pavement, causes traffic interruption, and provides only limited data at discrete locations. The nuclear gauge method also provides limited data measurement. Moreover, it requires a license for the operators because it uses radioactive material. To overcome the limitations of these traditional methods, this study proposes to develop a nondestructive method of using ground penetrating radar (GPR) to measure in-situ asphalt mixture density accurately, continuously, and rapidly.The prediction of asphalt mixture density using GPR is based on the fact that the dielectric constant of an asphalt mixture, which can be measured by GPR, is dependent on the dielectric and volumetric properties of its components. According to electromagnetic (EM) mixing theory, two candidate specific gravity models, namely the modified complex refractive index model (CRIM) and the modified Bottcher model, were developed to predict the bulk specific gravity of asphalt mixture from its dielectric constant.To evaluate the performance of these two models, a full-scale six-lane test site with four sections in each lane was carefully designed and constructed. Forty cores were extracted from the test site, and their densities were measured in the laboratory and compared to the GPRpredicted values using the two models. Both models were found effective in predicting asphalt mixture density, although the modified Bottcher model performed better. To account for the effect of the non-spherical inclusions in asphalt mixture and further improve the density prediction accuracy, a shape factor was introduced into the modified Bottcher model. Nonlinear least square curve fitting of the field core data indicated that a shape factor of -0.3 provided the best-performance model, which is referred to as the Al-Qadi Lahouar Leng (ALL) model.The performance of the ALL model was validated using data collected from an active pavement construction site in Chicago area. It was found that when the ALL model was employed, the prediction accuracy of the GPR was comparable to, or better than, that of the traditional nuclear gauge. For the asphalt mixtures without slags, the average density prediction errors of GPR were between 0.5% and 1.1% with two calibration cores, while those of the nuclear gauge were between 1.2% and 3.1%. iv Due to the importance of accurate input of the dielectric constant of asphalt mixture to the prediction accuracy of the specific gravity model, this study also looked into alternative methods for asphalt mixture dielectric constant estimation. The extended common mid-point (XCMP) method using two air-coupled antenna systems ...
Purpose:The authors examine the nonstationary noise behavior of a cone-beam CT system with FDK reconstruction. Methods: To investigate the nonstationary noise behavior, an analytical expression for the NPS of local volumes and an entire volume was derived and quantitatively compared to the NPS estimated from experimental air and water images. Results: The NPS of local volumes at different locations along the z-axis showed radial symmetry in the f x -f y plane and different missing cone regions in the f z direction depending on the tilt angle of rays through the local volumes. For local volumes away from the z-axis, the NPS of air and water images showed sharp transitions in the f x -f y and f y -f z planes and lack of radial symmetry in the f x -f y plane. These effects are mainly caused by varying magnification and different noise levels from view to view. In the NPS of the entire volume, the f x -f y plane showed radial symmetry because the nonstationary noise behaviors of local volumes were averaged out. The nonstationary sharp transitions were manifested as a high-frequency roll-off. Conclusions:The results from noise power analysis for local volumes and an entire volume demonstrate the spatially varying noise behavior in the reconstructed cone-beam CT images.
Purpose: In this paper, we propose a convolutional neural network (CNN)-based efficient model observer for breast computed tomography (CT) images. Methods: We first showed that the CNN-based model observer provided similar detection performance to the ideal observer (IO) for signal-known-exactly and background-known-exactly detection tasks with an uncorrelated Gaussian background noise image. We then demonstrated that a singlelayer CNN without a nonlinear activation function provided similar detection performance in breast CT images to the Hotelling observer (HO). To train the CNN-based model observer, we generated simulated breast CT images to produce a training dataset in which different background noise structures were generated using filtered back projection with a ramp, or a Hanning weighted ramp, filter. Circular, elliptical, and spiculated signals were used for the detection tasks. The optimal depth and the number of channels for the CNN-based model observer were determined for each task. The detection performances of the HO and a channelized Hotelling observer (CHO) with Laguerre-Gauss (LG) and partial least squares (PLS) channels were also estimated for comparison. Results: The results showed that the CNN-based model observer provided higher detection performance than the HO, LG-CHO, and PLS-CHO for all tasks. In addition, it was shown that the proposed CNN-based model observer provided higher detection performance than the HO using a smaller training dataset. Conclusions: In the presence of nonlinearity in the CNN, the proposed CNN-based model observer showed better performance than other linear observers.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.