Framing U-Net via Deep Convolutional Framelets: Application to Sparse-View CT

By Yoseob Han, Jong Chul Ye. IEEE Transactions on Medical Imaging, 2017

580 citations recorded by the source corpus.

X-ray computed tomography (CT) using sparse projection views is a recent approach to reduce the radiation dose. However, due to the insufficient projection views, an analytic reconstruction approach using the filtered back projection (FBP) produces severe streaking artifacts. Recently, deep learning approaches using large receptive field neural networks such as U-Net have demonstrated impressive performance for sparse-view CT reconstruction. However, theoretical justification is still lacking. Inspired by the recent theory of deep convolutional framelets, the main goal of this paper is, therefore, to reveal the limitation of U-Net and propose new multi-resolution deep learning schemes. In particular, we show that the alternative U-Net variants such as dual frame and tight frame U-Nets satisfy the so-called frame condition which makes them better for effective recovery of high frequency edges in sparse-view CT. Using extensive experiments with real patient data set, we demonstrate that the new network architectures provide better reconstruction performance.

Subfields: Biomedical Engineering, Radiology, Nuclear Medicine and Imaging

View Paper: https://www.semanticscholar.org/paper/efa7f1ecba74c4702e1e99c4e67426bcd4584a20

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Biomedical Engineering

In Radiology, Nuclear Medicine and Imaging

Citations in the corpus

fastMRI: An Open Dataset and Benchmarks for Accelerated MRI (2018, methodology): “U-Net models and their variants have successfully been used for many image-to-image prediction tasks including MRI reconstruction [17, 12] and image segmentation [29].”
Transformers in Medical Imaging: A Survey (2022, background): “Sparse-view CT [312] can effectively reduce the effective radiation dose by acquiring fewer projections.”
Deep Learning on Image Denoising: An overview (2019, unknown): “To overcome the blurry and false image artifacts, a dual U-Net with skip connection was proposed for computed tomography (CT) image reconstruction [73].”
Evolutionary Compression of Deep Neural Networks for Biomedical Image Segmentation (2020, background): “features to be captured may vary drastically from different tasks [11], [12].”
Deep Convolutional Framelets: A General Deep Learning Framework for Inverse Problems (2017, methodology): “For example, in image denoising [40, 76] or streaking artifact problems [28, 35], a residual network works better, since the artifacts are random and have complicated distribution.”
Modified U-Net (mU-Net) with Incorporation of Object-Dependent High Level Features for Improved Liver and Liver-Tumor Segmentation in CT Images (2019, methodology): “[12] and Han and Ye [20], but it is too much of a burden to the network in terms of memory (more than 100 layers).”
U-Net and Its Variants for Medical Image Segmentation: A Review of Theory and Applications (2020, methodology): “that implemented U-net to remove artifacts from images by reconstructing the images [79], [349]–[352], as well as a paper that used U-net for de-aliasing [80].”
FISTA-Net: Learning a Fast Iterative Shrinkage Thresholding Network for Inverse Problems in Imaging (2020, methodology): “The reference images were reconstructed by iradon operator in MATLAB using all 720 views [20], [51].”
Solving inverse problems using data-driven models (2019, background): “Hansen (1992), ‘Analysis of discrete ill-posed problems by means of the Lcurve’, SIAM Review 34, 561–580.”
A Gentle Introduction to Deep Learning in Medical Image Processing (2018, methodology): “Using wavelets for the multi-scale transform has been successfully applied in many applications ranging from denoising [109] to sparse view computed tomography [110].”
Multi-level Wavelet-CNN for Image Restoration (2018, background): “Note that our MWCNN is quite different with DCF [54]: DCF incorporates CNN with DWT in the view of decomposition, where different CNNs are deployed to each subband.”
Low-dose CT image and projection dataset (2020, unknown): “A multiresolution deep learning U-net for sparse-view CT.(30)”