Deep Cascade Wavelet Network for Compressed Sensing-MRI
| Autor: | Zhao Li, Chaoyang Liu, Qinjia Bao |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
Computer science
business.industry Deep learning ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION Wavelet transform 030218 nuclear medicine & medical imaging Image (mathematics) 03 medical and health sciences 0302 clinical medicine Wavelet Compressed sensing Sampling (signal processing) Cascade Feature (computer vision) Artificial intelligence business Algorithm 030217 neurology & neurosurgery |
| Zdroj: | Neural Information Processing ISBN: 9783030638290 ICONIP (1) |
| Popis: | Compressed sensing (CS) theory can accelerate magnetic resonance imaging (MRI) by sampling partial k-space measurements. Recently, deep learning models have been introduced to solve CS-MRI problem. It is noticed that the wavelet transform can obtain the coarse and detail information of the image, so we designed a deep cascade wavelet network (DCWN) to solve the CS-MRI problem. Our network consists of several sub-networks and each sub-network is delivered to the next one by dense connection. The input of each sub-network comprises 4 sub-bands of the former predictions in wavelet coefficients and outputs are residuals of 4 sub-bands of reconstructed MR images in wavelet coefficients. Wavelet transform enhances the sparsity of feature maps, which may greatly reduce the training burden for reconstructs high-frequency information, and provide more structural information. The experimental results show that DCWN can achieve better performance than previous methods, with fewer parameters and shorter running time. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|