Laplacian Pyramid-based Complex Neural Network Learning for Fast MR - - PowerPoint PPT Presentation

Laplacian Pyramid-based Complex Neural Network Learning for Fast MR Imaging

Haoyun Liang

Paul C. Lauterbur Research Center for Biomedical Imaging, SIAT, CAS

Accelerating MRI

Fast Imaging Sequence Parallel Imaging Compressed Sensing Deep Learning

Motivations

• multi-scale properties are underutilized
• the blurring issue of textures and details of tissues and organs
• normal convolution can not make full use of the information in

complex-valued MR images

Contributions

• pyramid structure decomposition is introduced to leverage multi-

scale properties

• cascaded structure is used for better restore textures and details of

the reconstructed images

• complex convolution is introduced to make full use of the

information in complex-valued MR images

Network Structure

Complex Laplacian pyramid decomposition 4x Shuffle downsample 2x Linear upsample Conv Block - 1 Conv Block - 2 Add Per-pixel Conv 4x Shuffle upsample 2x Shuffle upsample

Network Structure

…

!!" ℎ!" #!" ##$% = #!" ∗ && !#$% = !!"//r ℎ#$% = ℎ!"//r ##$% ℎ#$% !#$%

r X Shuffle downsample

!#$% ℎ#$% ##$%

…

#!" ℎ!" !!" ##$% = #!"//&& !#$% = !!" ∗ r ℎ#$% = ℎ!" ∗ r

r X Shuffle upsample

…

Residual Block

Complex conv ReLU

Add

Complex conv ReLU

Result

Reference Under-sampling U-Net KIKI-Net Cascade-Net kspace learning CLP-Net

Result

Reference Under-sampling U-Net KIKI-Net Cascade-Net kspace learning CLP-Net