TY - GEN
T1 - RCA-U-Net
T2 - 22nd International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2019
AU - Fang, Zhenghan
AU - Chen, Yong
AU - Nie, Dong
AU - Lin, Weili
AU - Shen, Dinggang
N1 - Funding Information:
This work was supported in part by NIH grant EB006733.
PY - 2019
Y1 - 2019
N2 - Magnetic resonance fingerprinting (MRF) is a relatively new imaging framework which allows rapid and simultaneous quantification of multiple tissue properties, such as T1 and T2 relaxation times, in one acquisition. To accelerate the data sampling in MRF, a variety of methods have been proposed to extract tissue properties from highly accelerated MRF signals. While these methods have demonstrated promising results, further improvement in the accuracy, especially for T2 quantification, is needed. In this paper, we present a novel deep learning approach, namely residual channel attention U-Net (RCA-U-Net), to perform the tissue quantification task in MRF. The RCA-U-Net combines the U-Net structure with residual channel attention blocks, to make the network focus on more informative features and produce better quantification results. In addition, we improved the preprocessing of MRF data by masking out the noisy signals in the background for improved quantification at tissue boundaries. Our experimental results on two in vivo brain datasets with different spatial resolutions demonstrate that the proposed method improves the accuracy of T2 quantification with MRF under high acceleration rates (i.e., 8 and 16) as compared to the state-of-the-art methods.
AB - Magnetic resonance fingerprinting (MRF) is a relatively new imaging framework which allows rapid and simultaneous quantification of multiple tissue properties, such as T1 and T2 relaxation times, in one acquisition. To accelerate the data sampling in MRF, a variety of methods have been proposed to extract tissue properties from highly accelerated MRF signals. While these methods have demonstrated promising results, further improvement in the accuracy, especially for T2 quantification, is needed. In this paper, we present a novel deep learning approach, namely residual channel attention U-Net (RCA-U-Net), to perform the tissue quantification task in MRF. The RCA-U-Net combines the U-Net structure with residual channel attention blocks, to make the network focus on more informative features and produce better quantification results. In addition, we improved the preprocessing of MRF data by masking out the noisy signals in the background for improved quantification at tissue boundaries. Our experimental results on two in vivo brain datasets with different spatial resolutions demonstrate that the proposed method improves the accuracy of T2 quantification with MRF under high acceleration rates (i.e., 8 and 16) as compared to the state-of-the-art methods.
UR - http://www.scopus.com/inward/record.url?scp=85075641283&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85075641283&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-32248-9_12
DO - 10.1007/978-3-030-32248-9_12
M3 - Conference contribution
AN - SCOPUS:85075641283
SN - 9783030322472
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 101
EP - 109
BT - Medical Image Computing and Computer Assisted Intervention – MICCAI 2019 - 22nd International Conference, Proceedings
A2 - Shen, Dinggang
A2 - Yap, Pew-Thian
A2 - Liu, Tianming
A2 - Peters, Terry M.
A2 - Khan, Ali
A2 - Staib, Lawrence H.
A2 - Essert, Caroline
A2 - Zhou, Sean
PB - Springer
Y2 - 13 October 2019 through 17 October 2019
ER -