Ptychography is a powerful technique that combines scanning transmission X-ray microscopy (STXM) and coherent diffraction imaging (CDI). Thanks to the fluctuation of light and the development of phase retrieval algorithms, ptychography has achieved ultra-high imaging resolution. However, iterative solutions are time consuming. A cascaded residual network is used to recover the amplitude and phase information of the sample directly from the received diffraction patterns. With the powerful coding and decoding capabilities of the residual blocks, the correspondence between diffraction patterns and sample distribution information can be learned by the network. Experimental results show that the use of cascaded residual blocks enhances the reconstructive capability of the network.
X-ray tomographic imaging has become an important analytical tool with a wide range of applications. It is inevitable that noise is introduced in CT images, and noise reduction is necessary. To solve this problem, we considered to use the nonlocal property of similar block search and proposed a deep learning network based on similar block learning for noise reduction of micro CT short exposure time scanned images to improve the scanning efficiency while ensuring high quality imaging. The method uses the output of the nonlocal method as a data preprocessing algorithm by combining a nonlocal block matching algorithm with a convolutional neural network, and uses a residual channel attention mechanism to learn the features after feature extraction, which reduces noise while preserving image details. Experimental results show that the method can remove noise from CT images quickly and effectively, and compared with the classical CPCE noise reduction method, the method improves the PSNR index by 1.52 dB, which is consistent with the theoretical assumption.
Computed Tomography (CT) is one of the essential techniques for non-destructive testing. The acquisition of accurate reconstructed images is the basis for the subsequent analytical processing tasks. This paper proposes a convolutional neural network-based CT reconstruction algorithm to generate reconstructed CT images directly from sinogram by the feature coding and decoding capability. The reconstruction of abdominal scanning data is carried out by this method, and the results show that we can quickly obtain corresponding reconstruction results. During the network training, we designed different data pre-processing methods. We analysed the role of each module in the network by visualizing the output features of each module. Finally, the role of different modules in feature extraction and image generation is further analysed. We found that the conversion from projection to image can be effectively achieved using only convolution operations. It is essential for reconstructing CT images using deep learning techniques.
Low dose computed tomography (LDCT) has attracted considerable interest in medical imaging fields. Reducing tube current intensity and decrease the exposure time are the two main ways in clinic applications. Nevertheless, the resulting statistical noise will seriously degrade CT image quality for diagnosis. To make full use of the original projection data as well as further improving the small dataset processing ability of U-net, this study aimed to investigate a low dose X-ray CT image denoising method via U-net in projection domain (PDI U-net). Meanwhile, in view of avoiding the excessive smoothing of the small structures, the inception module is introduced in the encoding stage of the network. And different convolution kernel operations of 1×1, 3×3, and 5×5 are used in parallel to obtain multi-scale image features, increasing the depth and width of the network while reducing the parameters. Furthermore, the shortcut connection is utilized to transfer the low-level area local detail information to the high-level area. By merging it with the global information of the high-level area, the proposed network can maintain the image details while de-noising. The experimental results show that the method proposed can significantly improve the image quality with clear feature edges and close visual appearance to the reference high dose CT images. Compared with LDCT and Residual Encoder-Decoder Convolutional Neural Network (RED-CNN), the peak signal to noise radio (PSNR) is improved by 9.02dB and 2.74dB, and the structural similarity (SSIM) is improved by 0.43 and 0.07, respectively.
Due to the limitation of scanning conditions and the emerging clinical application of local imaging, local tomography has become a research hotspot. Traditionally, completing the projection data through interpolation or spatial transformation iteration is popular to overcome the truncation artifact, such as iterative reconstruction algorithms. Recently, deep learning networks have been incorporated in dealing with such problems. Instead of global filtering on truncated data, the proposed work focuses on the full feature extraction using U-net as well as the usage of the redundancy between projection sinogram, by performing extrapolation of truncated projection data through data learning and then filter back projection local reconstruction with high efficiency. During the learning process, 3439 projections are selected as complete projection data , and the corresponding truncated data is simulated according to the actual truncation situation. Then, 150 of the truncated data are randomly selected as the test samples, while the rest 3289 of those as the training samples in the U-net. The output sinogram data is compared with the original complete data by calculating the L2 loss function of both. And the Adam optimizer is used to continuously optimize the parameters of the network. RMSE and NMAD are used to quantitatively evaluate the reconstruction effect. Experimental results show that the proposed method based on truncated data extrapolation network can obviously suppress the ring artifacts and compared with images directly reconstructed using truncated projection data, the RMSE is reduced by an average of 43.185%, and the NMAD is reduced by 44.24%.
Computed tomography (CT) has been extensively used in nondestructive testing, medical diagnosis, etc. In the field of modern medicine, metal implants are widely used in people's daily life, and the serious artifacts in CT reconstruction images caused by metal implants cannot be ignored. Sinogram contains the most realistic projection information of patients. Processing in the sinogram domain directly can make the effective information maximum extent preserved. In this paper, we propose a novel method based on full convolutional network (FCN) for metal artifact reduction in the sinogram domain. The networks we introduced use the complete sinogram data to learn a mapping function to correct the metal-corrupted sinogram data. The network takes the metal-corrupted sinogram as the input and takes the artifact-free sinogram as the target. Compared with the existing deep learning-based CT artifact reduction methods, our work just uses the sinogram information to correct the metal artifacts. The proposed network can process images of different sizes. Our initial results on a simulated dataset to demonstrate the potential effectiveness of this new approach to suppressing artifacts.
Metal objects inside the field of view would introduce severe artifacts in x-ray CT images, which would severely degrade the quality of CT data and bring huge difficulties for subsequent image processing and analysis. Correction of metal artifacts has become a hot and difficult issue in X-ray CT. In recent years, deep learning has rapidly gained attention for employment on image processing. In this study, we introduce a Fully Convolutional Networks (FCNs) into the MAR in image domain. The network reduces metal artifacts by learning an end-to-end mapping of images from metal-corrupted CT images to their corresponding artifact-free ground truth. The network takes the metal-corrupted CT images as the input and takes the artifact-free images as the target. The convolution layers extract features from the input images and map them to the target images, and the deconvolution layers use these features to build the predicted outputs. Experimental results demonstrate that the proposed method can well reduce metal artifacts of CT images, and take a shorter time to process the images than traditional method.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.