Tianming Du1, Yuemeng Li1, Honggang Zhang2, Stephen Pickup1, Rong Zhou1, Hee Kwon Song1, and Yong Fan1
1Radiology, University of Pennsylvania, Philadelphia, PA, United States, 2Beijing University of Posts and Telecommunications, Beijing, China
1Radiology, University of Pennsylvania, Philadelphia, PA, United States, 2Beijing University of Posts and Telecommunications, Beijing, China
A
deep learning model with adaptive convolutional neural networks is developed for
k-space data interpolation. Ablation and experimental results show that our method
achieves better image reconstruction than existing state-of-the-art techniques.
Figure-1. A
residual Encoder-Decoder network of CNNs (top and middle), enhanced by
frequency-attention and channel-attention layers (bottom), for image
reconstruction from undersampled k-space
data.
Figure-2. Visualization
of representative cases of the Stanford knee dataset (top row) and the fastMRI
brain dataset (bottom row), including images reconstructed from the fully
sampled data and from under-sampled data without CNN processing. The difference
images were amplified 5 times for Stanford cases and 10 times for fastMRI
dataset. Yellow and red boxes indicate the zoomed-in and difference images,
respectively.
