Abstract:<h3> Purpose</h3><p>Deep‐neural‐network‐based image reconstruction has demonstrated promising performance in medical imaging for under‐sampled and low‐dose scenarios. However, it requires large amount of memory and extensive time for the training. It is especially challenging to train the reconstruction networks for three‐dimensional computed tomography (CT) because of the high resolution of CT images. The purpose of this work is to reduce the memory and time consumption of the training of the reconstruction networks for CT to make it practical for current hardware, while maintaining the quality of the reconstructed images.</p><h3> Methods</h3><p>We unrolled the proximal gradient descent algorithm for iterative image reconstruction to finite iterations and replaced the terms related to the penalty function with trainable convolutional neural networks (CNN). The network was trained greedily iteration by iteration in the image‐domain on patches, which requires reasonable amount of memory and time on mainstream graphics processing unit (GPU). To overcome the local‐minimum problem caused by greedy learning, we used deep UNet as the CNN and incorporated separable quadratic surrogate with ordered subsets for data fidelity, so that the solution could escape from easy local minimums and achieve better image quality.</p><h3> Results</h3><p>The proposed method achieved comparable image quality with state‐of‐the‐art neural network for CT image reconstruction on 2D sparse‐view and limited‐angle problems on the low‐dose CT challenge dataset. The difference in root‐mean‐square‐error (RMSE) and structural similarity index (SSIM) was within [−0.23,0.47]HU and [0,0.001] respectively with 95% confidence level. For 3D image reconstruction with ordinary‐size CT volume, the proposed method only needed 2 GB graphics processing unit (GPU) memory and 0.45 seconds per training iteration as minimum requirement, whereas existing methods may require 417 GB and 31 minutes. The proposed method achieved improved performance compared to total‐variation‐ and dictionary‐learning‐based iterative reconstruction for both 2D and 3D problems.</p><h3> Conclusions</h3><p>We proposed a training‐time computationally efficient neural network for CT image reconstruction. The proposed method achieved comparable image quality with state‐of‐the‐art neural network for CT reconstruction, with significantly reduced memory and time requirement during training. The proposed method is applicable to 3D image reconstruction problems such as cone‐beam CT and tomosynthesis on mainstream GPUs.</p><p>This article is protected by copyright. All rights reserved.</p>

Combination Strategy of Deep Learning and Direct Back Projection for High-efficiency Computed Tomography Reconstruction

Deep Fusion Network Based Sparse View CT Reconstructions for Clinical Diagnostic Scanners

Deconvolution-Based Backproject-Filter (BPF) Computed Tomography Image Reconstruction Method Using Deep Learning Technique

Deep Filtered Back Projection for CT Reconstruction

Deep Learning Single View Computed Tomography Guided by FBP Algorithm

Computationally efficient deep neural network for computed tomography image reconstruction

Wide Activation for Accurate and Fast Computed Tomography Reconstruction in Limited-Angle Problem

A Deep Learning Reconstruction Method for Sparsely Spaced Multidetector Helical CT Using Cascaded Dual-Domain CNN

A Deep RNN for CT Image Reconstruction

Derivative-Hilbert-Backprojection Based Image Reconstruction from Truncated Projections in Helical Cone-Beam Ct

Combining Deep Learning and Adaptive Sparse Modeling for Low-dose CT Reconstruction

A Dual-Domain CNN-Based Network for CT Reconstruction

Research on High Sparse Sampling CT Image Reconstruction Algorithm Based on Convolutional Neural Networks

Direct PET Image Reconstruction Incorporating Deep Image Prior and a Forward Projection Model

A Deep Convolutional Gated Recurrent Unit for CT Image Reconstruction.

Dual-domain sparse-view CT reconstruction with Transformers

A Deep Learning Reconstruction Framework for Differential Phase-Contrast Computed Tomography With Incomplete Data

Image Prediction for Limited-angle Tomography via Deep Learning with Convolutional Neural Network

Low-dose CT image reconstruction based on a deep neural network

An Effective Cuda Parallelization Of Projection In Iterative Tomography Reconstruction

Residual Back Projection With Untrained Neural Networks