GAN-TL: Generative Adversarial Networks with Transfer Learning for MRI Reconstruction
Problem
GAN-based MRI reconstruction from under-sampled data normally requires large training sets of raw k-space patient data, which is difficult to gather in clinical practice.
Why it matters
Reducing the training data needed for MRI reconstruction could make accelerated clinical MRI more practical.
What the AI did
The study combines deep learning-based generative adversarial networks with transfer learning to improve reconstruction generalizability from small sample sets.
Results
Stated to outperform current techniques in PSNR and SSIM for brain and knee MRI reconstruction; no numbers given.
Impact
Published in Applied Sciences with 28 citations; claimed outperformance on brain and knee imaging tasks.
Record
| Venue | Applied Sciences |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Muhammad Yaqub, Jinchao Feng, Shahzad Ahmed, Kaleem Arshid, Muhammad Atif Bilal, Muhammad Pervez Akhter, Muhammad Sultan Zia |
| Citations | 28 (OpenAlex, at import time) |
| References | 55 |
| Model or method | Generative Adversarial Network with transfer learning (GAN-TL) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.3390/app12178841 |
| Open access | gold |
| Keywords | GAN, transfer learning, MRI reconstruction, PSNR, SSIM, k-space |
Importance rating
real but narrower advance
Reducing training-data needs could ease clinical MRI deployment, a real but narrower methods advance; abstract gives no scale of affected patients.