DeepResp

• The code is for correction of respidation-induced B0 fluctuation artifacts using deep learning (DeepResp)
• Last update : 2020.10.26
• The source data for training can be shared to academic institutions. Request should be sent to snu.list.software@gmail.com. For each request, individual approval from our institutional review board is required (i.e. takes time)
• For more information, refer to the published paper (https://doi.org/10.1016/j.neuroimage.2020.117432)

References

• DeepResp: Deep learning solution for respiration-induced B0 fluctuation artifacts in multi-slice GRE H. An, H.-G. Shin, S. Ji, W. Jung, S. Oh, D. Shin, J. Park, J. Lee. DeepResp: Deep learning solution for respiration-induced B0 fluctuation artifacts in multi-slice GRE. Neuroimage. 2021 Jan. v 224. https://www.sciencedirect.com/science/article/pii/S1053811920309174

Overview

DeepResp

Requirements

• Python 3.7

• Pytorch 1.5.1

• NVIDIA GPU (CUDA 10.1) (MultiGPUs are avaliable)

Data acquisition

• MR images for simulation were acquired at 3T MRI (SIEMENS), which were from below refereces. The images were either zero-padded or cropped in k-space to match the matrix size to 224 × 224. Each image was masked out noises in the background using an intensity threshold to remove artifacts in the background.

  • QSMnet
    J. Yoon, E. Gong, I. Chatnuntawech, B. Bilgic, J. Lee, W. Jung, J. Ko, H. Jung, K. Setsompop, G. Zaharchuk, E.Y. Kim, J. Pauly, J. Lee. Quantitative susceptibility mapping using deep neural network: QSMnet. Neuroimage. 2018 Oct;179:199-206. https://www.sciencedirect.com/science/article/pii/S1053811918305378
  • QSMnet+
    W. Jung, J. Yoon, S. Ji, J. Choi, J. Kim, Y. Nam, E. Kim, J. Lee. Exploring linearity of deep neural network trained QSM: QSMnet+. Neuroimage. 2020 May; 116619. https://www.sciencedirect.com/science/article/pii/S1053811920301063

• Respiration data for simulation was acquired with a a temperature sensor (Biopac). The data were sampled at 500 Hz and recorded for 7 sessions, each with 390 seconds. A median filter and a bandpass-filter (passband: 0.1 Hz ~ 1 Hz) were applied to reduce noise.

• MR images for in-vivo experments were acquired at 3T MRI (SIEMENS) using a multi-slice GRE sequence with a navigator echo. The scan parameters were as follows: TR = 1200 ms, TE = 6.9 ms, 15.2 ms, 20.5 ms, 25.7 ms, 31.0 ms, 36.3 ms, and 41.5 ms for the images, 55.0 ms for the navigator, flip angle = 70°, bandwidth = 260 Hz/pixel, FOV = 224 × 224 mm2, in-plane resolution = 1 × 1 mm2, slice thickness = 2 mm, distance factor = 20%, and 18 slices for 9 subjects and 16 slices for 1 subject.

Simulation

• The source code for simulation generates the simulated respiration-corrupted images with the MR images and the respiration data.
• MR images : (Height x Width x slices) complex numpy data, Respiration data : (Subjects x data sample) float numpy data,
• Results: the complex-valued numpy images ( read-out x phase-encoding x slice ) are generated.

Training

• The source code for training. The training performed with the saved data from the simulation.

Evaluation

• The source code for evaluation of the trained neural networks.
• The evaluation can be performed with the simulated data and the in-vivo data.
• Results: networks-generated phase errors

Correction

• The source code for correction of corrupted images using phase errors (network-generated or reference)
• Results: DeepResp-corrected images, reference-corrected images