Low-dose X-ray computed tomography (CT) simulation from high-dose scan is required in optimizing radiation dose to patients. In this study, we propose a simple low-dose CT simulation strategy in sinogram domain using the raw data from high-dose scan. Specially, a relationship between the incident fluxes of low- and high- dose scans is first determined according to the repeated projection measurements and analysis. Second, the incident flux level of the simulated low-dose scan is generated by properly scaling the incident flux level of high-dose scan via the determined relationship in the first step. Third, the low-dose CT transmission data by energy integrating detection is simulated by adding a statistically independent Poisson noise distribution plus a statistically independent Gaussian noise distribution. Finally, a filtered back-projection (FBP) algorithm is implemented to reconstruct the resultant low-dose CT images. The present low-dose simulation strategy is verified on the simulations and real scans by comparing it with the existing low-dose CT simulation tool. Experimental results demonstrated that the present low-dose CT simulation strategy can generate accurate low-dose CT sinogram data from high-dose scan in terms of qualitative and quantitative measurements.
The Radon transform is widely used in physical and life sciences, and one of its major applications is in medical X-ray computed tomography (CT), which is significantly important in disease screening and diagnosis. In this paper, we propose a novel reconstruction framework for Radon inversion with deep learning (DL) techniques. For simplicity, the proposed framework is denoted as iRadonMAP, i.e., inverse Radon transform approximation. Specifically, we construct an interpretable neural network that contains three dedicated components. The first component is a fully connected filtering (FCF) layer along the rotation angle direction in the sinogram domain, and the second one is a sinusoidal back-projection (SBP) layer, which back-projects the filtered sinogram data into the spatial domain. Next, a common network structure is added to further improve the overall performance. iRadonMAP is first pretrained on a large number of generic images from the ImageNet database and then fine-tuned with clinical patient data. The experimental results demonstrate the feasibility of the proposed iRadonMAP framework for Radon inversion.
To reduce radiation dose in X-ray computed tomography (CT) imaging, one of the common strategies is to lower the milliampere-second (mAs) setting during projection data acquisition. However, this strategy would inevitably increase the projection data noise, and the resulting image by the filtered back-projection (FBP) method may suffer from excessive noise and streak artifacts. The edge-preserving nonlocal means (NLM) filtering can help to reduce the noise-induced artifacts in the FBP reconstructed image, but it sometimes cannot completely eliminate them, especially under very low-dose circumstance when the image is severely degraded. To deal with this situation, we proposed a statistical image reconstruction scheme using a NLM-based regularization, which can suppress the noise and streak artifacts more effectively. However, we noticed that using uniform filtering parameter in the NLM-based regularization was rarely optimal for the entire image. Therefore, in this study, we further developed a novel approach for designing adaptive filtering parameters by considering local characteristics of the image, and the resulting regularization is referred to as adaptive NLM-based regularization. Experimental results with physical phantom and clinical patient data validated the superiority of using the proposed adaptive NLM-regularized statistical image reconstruction method for low-dose X-ray CT, in terms of noise/streak artifacts suppression and edge/detail/contrast/texture preservation.
Ring artifacts often appear in flat-panel detector-based CT images due to the malfunction or mis-calibration of the detector elements that result in stripe artifacts in the line integral projection (sinogram) data. The ring artifacts lower the image quality and affect image-based diagnoses. Here we propose a ring artifacts removal approach based on wavelet filtering in the sinogram domain. The line integral projection (sinogram) dataset were divided into 4 sub-sinogram dataset, and for each of them the wavelet decomposition operation was employed to produce the associated wavelet dataset, followed by filtering operation on the vertical detail band and the low-pass detail band. Wavelet reconstruction operation was then performed, and the weighted moving average filter was used to yield the filtered sinogram, which was processed using filtered back-projection (FBP) for image reconstruction. The results showed that the proposed approach could effectively remove the ring artifacts while preserving the structural information of the image.
Restriction by hardware caused the very low projection number at a single phase for 4-dimensional cone beam (4D-CBCT) CT imaging, and reconstruction using conventional reconstruction algorithms is thus constrained by serious streak artifacts and noises. To address this problem, we propose an approach to reconstructing 4D-CBCT images with multi-phase projections based on the assumption that the image at one phase can be viewed as the motion-compensated image at another phase. Specifically, we formulated a cost function using multi-phase projections to construct the fidelity term and the TV regularization method. For fidelity term construction, the projection data of the current phase and those at other phases were jointly used by reformulating the imaging model. The Gradient-Projection-Barzilai-Line search (GPBL) method was used to optimize the complex cost function. Physical phantom and patient data results showed that the proposed approach could effectively reduce the noise and artifacts, and the introduction of additional temporal correlation did not introduce new artifacts or motion blur.
Tremendous research efforts have been devoted to minimizing the radiation exposure to patients by acquiring the X-ray computed tomography (CT) transmission data at as low radiation exposure as reasonably practical (ALARP) and developing the corresponding image reconstruction methods. To address the ALARP radiation, this study aims to develop texture-enhancing image reconstruction algorithms and texture-based image quality evaluation strategies because image textures play an essential role for many clinical tasks. The image reconstruction is based on the maximum a posteriori probability given the acquired data, where the a priori knowledge is learnt tissue textures from the existing diagnostic full-dose CT image, and the transmission data fidelity is modeled by a shift Poisson statistic considering both the X-ray quanta fluctuation and the system electronic background noise. The image evaluation is based on the regional gray-scale co-occurrence texture measures. Evaluation of the developed methodologies was performed on patient data acquired with 120kVp and 100 mAs settings, followed on simulated data at 20, 10, 5 and 1mAs. The image texture measures showed a monotonic drop as the dose level decreased from 20 to 1 mAs. The most striking observation is a critical turning point on the plot of the relative change of texture measure vs. the mAs levels. This critical turning point indicates the minimum dose level that a CT scanner hardware configuration and image reconstruction software can achieve with a reasonable image quality. The effect of the background noise is also evaluated through the simulated data in this study.
In the clinics, unenhanced CT (unCT) are first used for acute ischemic stroke detection, but sometimes it is difficult to delineate acute infarction in the unCT in the first few hours. Currently, deep learning methods can be used for stroke segmentation, but they are based on a large amount of annotation data, especially in unCT imaging. Meanwhile, the Diffusion-weighted magnetic resonance imaging (DWI) can clearly distinguish infarcted brain tissue from healthy region. In this work, we investigate whether external information, i.e., DWI images, can be used to help more clearly delineates acute infraction in the unCT images by a transfer learning network. We detail the network architectures in the transfer learning framework, and describe the training process used to transfer segmentation task in DWI images to unCT images. In particular, intermediate layers in the network trained with DWI images are frozen and then directly are transferred to the network trained with unCT task by parameter fine-tuning. The ISLES2018 dataset is used to validate and evaluate this task-oriented transfer network. Experimental results demonstrate that this task-oriented transfer network can obtain the dice coefficient with 0.793, which is much better than the network that only uses unCT input (dice coefficient: 0.570).
Restricted by the hardware, the projection number at a single phase for 4D-CBCT imaging is very low or even less than 10, thus the associated reconstruction by using conventional reconstruction algorithms will be constrained by serious streak artifacts and noises. To address this problem, in this paper, we are aiming to develop an approach to reconstruct the 4D-CBCT image with multi-phase projections, which means that when the images at one phase were estimated not only from the projection data of the current phase but also the projections at the other phases. The proposed approach is based on the assumption that the image at one phase can be viewed as the motion-compensated image of another phase. Specifically, in this work, we formulate a cost function using multi-phase projections to construct the fidelity term and the TV regularization method was adopted. The Gradient-Projection-Barzilai-Linesearch (GPBL) method was used to optimize the complex cost function. Physical phantom and real patient data were used to evaluate the proposed algorithm. Results show that the proposed approach can effectively reduce the noise and artifacts, which suggest that the introduction of additional temporal correlation (along the phase direction) can improve the 4D-CBCT image quality.
'/%3e%3cuse xlink:href='%23a' transform='rotate(23.036 .093 25.536)'/%3e%3cuse xlink:href='%23a' transform='rotate(45.87 1.273 16.18)'/%3e%3cuse xlink:href='%23a' transform='rotate(69.945 .996 12.078)'/%3e%3cuse xlink:href='%23a' transform='rotate(20.66 -19.689 31.932)'/%3e%3c/svg%3e)