Deep ADMM-Net for Compressive Sensing MRI

Authors: yan yang, Jian Sun, Huibin Li, Zongben Xu

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on MRI image reconstruction under different sampling ratios in k-space demonstrate that it significantly improves the baseline ADMM algorithm and achieves high reconstruction accuracies with fast computational speed.
Researcher Affiliation Academia Yan Yang Xi an Jiaotong University yangyan92@stu.xjtu.edu.cn Jian Sun Xi an Jiaotong University jiansun@mail.xjtu.edu.cn Huibin Li Xi an Jiaotong University huibinli@mail.xjtu.edu.cn Zongben Xu Xi an Jiaotong University zbxu@mail.xjtu.edu.cn
Pseudocode No No structured pseudocode or algorithm blocks were found in the paper.
Open Source Code No The paper does not provide any explicit statement or link for open-sourcing the code for the methodology described.
Open Datasets Yes We train and test ADMM-Net on brain and chest MR images2. For each dataset, we randomly take 100 images for training and 50 images for testing. ADMM-Net is separately learned for each sampling ratio. The reconstruction accuracies are reported as the average NMSE and Peak Signalto-Noise Ratio (PSNR) over the test images. The sampling pattern in k-space is the commonly used pseudo radial sampling. All experiments are performed on a desktop with Intel core i7-4790k CPU. 2CAF Project: https://masi.vuse.vanderbilt.edu/workshop2013/index.php/Segmentation_Challenge_Details
Dataset Splits No The paper states, "For each dataset, we randomly take 100 images for training and 50 images for testing." but does not mention a validation set split or provide details on cross-validation.
Hardware Specification Yes All experiments are performed on a desktop with Intel core i7-4790k CPU.
Software Dependencies No The paper mentions "L-BFGS1" with a URL and "Rice Wavelet Toolbox: http://dsp.rice.edu/software/rice-wavelet-toolbox" but does not provide specific version numbers for these or other general software dependencies.
Experiment Setup Yes We choose normalized mean square error (NMSE) as the loss function in network training. We learn the parameters by minimizing the loss w.r.t. them using L-BFGS. We initialize the network parameters Θ according to the ADMM solver of the following baseline CS-MRI model. In this model, we set Dl as a DCT basis and impose l1-norm regularization in the DCT transform space. In the nonlinear transform layer, we uniformly choose 101 positions located within [-1,1]. The sampling pattern in k-space is the commonly used pseudo radial sampling.