Adaptive Deep Dictionary Learning for MRI Reconstruction
This work addresses the well known problem of reconstructing magnetic resonance images from their partially samples K-space. Compressed sensing (CS) based techniques have been used rampantly for the said problem. Later studies, instead of employing a fixed basis (like DCT, wavelet etc. as used in CS), learnt the basis adaptively from the image itself. Such studies, loosely dubbed as dictionary learning (DL) showed marked improvement over CS. This work proposes deep dictionary learning based inversion. Instead of learning a single level of basis, we learn multiple levels adaptively from the image, while reconstructing it. The results show marked improvement over all previously known techniques.
KeywordsDictionary learning Deep learning Reconstruction
We are thankful in part to the Infosys Center for Artificial Intelligence @ IIITD for partial support and in part to 5IOA036 FA23861610004 grant by Air Force Office of Scientific Research (AFOSR), AOARD.
- 3.Ravishankar, S., Bresler, Y.: Sparsifying transform learning for compressed sensing MRI. In: IEEE ISBI, pp. 17–20 (2013)Google Scholar
- 6.Kulkarni, K., Lohit, S., Turaga, P., Kerviche, R., Ashok, A.: ReconNet: non-iterative reconstruction of images from compressively sensed measurements. In: IEEE CVPR, pp. 449–458 (2016)Google Scholar
- 8.Combettes, P.L., Pesquet, J.C.: Proximal splitting methods in signal processing. In: Fixed-point Algorithms for Inverse Problems in Science and Engineering, pp. 185–212 (2011)Google Scholar