Fine-tuning deep learning model parameters for improved super-resolution of dynamic MRI with prior-knowledge

被引:22
作者
Sarasaen, Chompunuch [1 ,2 ,3 ]
Chatterjee, Soumick [1 ,3 ,4 ,5 ]
Breitkopf, Mario [1 ,3 ]
Rose, Georg [2 ,3 ]
Nurnberger, Andreas [4 ,5 ,7 ]
Speck, Oliver [1 ,3 ,6 ,7 ,8 ]
机构
[1] Otto von Guericke Univ, Biomed Magnet Resonance, Magdeburg, Germany
[2] Otto von Guericke Univ, Inst Med Engn, Magdeburg, Germany
[3] Otto von Guericke Univ, Res Campus STIMULATE, Magdeburg, Germany
[4] Otto von Guericke Univ, Fac Comp Sci, Magdeburg, Germany
[5] Otto von Guericke Univ, Data & Knowledge Engn Grp, Magdeburg, Germany
[6] German Ctr Neurodegenerat Dis, Magdeburg, Germany
[7] Ctr Behav Brain Sci, Magdeburg, Germany
[8] Leibniz Inst Neurobiol, Magdeburg, Germany
关键词
Super-resolution; Dynamic MRI; Prior knowledge; Fine-tuning; Patch-based super-resolution; Deep learning; SINGLE-IMAGE SUPERRESOLUTION; RECONSTRUCTION; RESOLUTION; NETWORK;
D O I
10.1016/j.artmed.2021.102196
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Dynamic imaging is a beneficial tool for interventions to assess physiological changes. Nonetheless during dynamic MRI, while achieving a high temporal resolution, the spatial resolution is compromised. To overcome this spatio-temporal trade-off, this research presents a super-resolution (SR) MRI reconstruction with prior knowledge based fine-tuning to maximise spatial information while reducing the required scan-time for dynamic MRIs. A U-Net based network with perceptual loss is trained on a benchmark dataset and fine-tuned using one subjectspecific static high resolution MRI as prior knowledge to obtain high resolution dynamic images during the inference stage. 3D dynamic data for three subjects were acquired with different parameters to test the generalisation capabilities of the network. The method was tested for different levels of in-plane undersampling for dynamic MRI. The reconstructed dynamic SR results after fine-tuning showed higher similarity with the high resolution ground-truth, while quantitatively achieving statistically significant improvement. The average SSIM of the lowest resolution experimented during this research (6.25% of the k-space) before and after fine-tuning were 0.939 +/- 0.008 and 0.957 +/- 0.006 respectively. This could theoretically result in an acceleration factor of 16, which can potentially be acquired in less than half a second. The proposed approach shows that the superresolution MRI reconstruction with prior-information can alleviate the spatio-temporal trade-off in dynamic MRI, even for high acceleration factors.
引用
收藏
页数:9
相关论文
共 71 条
[1]  
[Anonymous], 2015, A neural algorithm of artistic style, DOI DOI 10.1167/16.12.326
[2]  
Bernstein M. A., 2004, Handbook of MRI Pulse Sequences, DOI 10.1016/B978-012092861-3/50021-2
[3]   Super-resolution musculoskeletal MRI using deep learning [J].
Chaudhari, Akshay S. ;
Fang, Zhongnan ;
Kogan, Feliks ;
Wood, Jeff ;
Stevens, Kathryn J. ;
Gibbons, Eric K. ;
Lee, Jin Hyung ;
Gold, Garry E. ;
Hargreaves, Brian A. .
MAGNETIC RESONANCE IN MEDICINE, 2018, 80 (05) :2139-2154
[4]   Efficient and Accurate MRI Super-Resolution Using a Generative Adversarial Network and 3D Multi-level Densely Connected Network [J].
Chen, Yuhua ;
Shi, Feng ;
Christodoulou, Anthony G. ;
Xie, Yibin ;
Zhou, Zhengwei ;
Li, Debiao .
MEDICAL IMAGE COMPUTING AND COMPUTER ASSISTED INTERVENTION - MICCAI 2018, PT I, 2018, 11070 :91-99
[5]  
Choi K, 2017, ARXIV170309179
[6]  
Cicek O, 2016, 3D U-Net: Learning dense volumetric segmentation from sparse annotation, DOI DOI 10.1007/978-3-319-46723-8_49
[7]   Collaborative patch-based super-resolution for diffusion-weighted images [J].
Coupe, Pierrick ;
Manjon, Jose V. ;
Chamberland, Maxime ;
Descoteaux, Maxine ;
Hiba, Bassem .
NEUROIMAGE, 2013, 83 :245-261
[8]  
Dai W., 2007, P 22 NAT C ART INT V, V1, P540
[9]   Accelerating the Super-Resolution Convolutional Neural Network [J].
Dong, Chao ;
Loy, Chen Change ;
Tang, Xiaoou .
COMPUTER VISION - ECCV 2016, PT II, 2016, 9906 :391-407
[10]   Learning a Deep Convolutional Network for Image Super-Resolution [J].
Dong, Chao ;
Loy, Chen Change ;
He, Kaiming ;
Tang, Xiaoou .
COMPUTER VISION - ECCV 2014, PT IV, 2014, 8692 :184-199