An Adaptive Intelligence Algorithm for Undersampled Knee MRI Reconstruction
Problem
MRI acquisition is slow, so images must be reconstructed from undersampled k-space data to accelerate scanning.
Why it matters
Accelerating MRI acquisition could cut scan times and expand access to diagnostic imaging.
What the AI did
A compressed-sensing-inspired iterative learning-based reconstruction scheme with a novel deep neural network refined and corrected prior reconstruction assumptions from undersampled knee MRI data.
Results
Ranked #1 (8x accelerated multi-coil), shared #1 (4x multi-coil), and #3 on a third fastMRI challenge track.
Impact
Top-ranked performance on an independently evaluated fastMRI challenge dataset supports deep learning for accelerated knee MRI.
Record
| Venue | IEEE Access |
|---|---|
| Published | 2020 |
| Work type | article |
| Authors | Nicola Pezzotti, Sahar Yousefi, Mohamed S. Elmahdy, Jeroen Hendrikus Fransiscus Van Gemert, Christophe Schülke, Mariya Doneva, Tim Tolker‐Nielsen, Sergey Kastryulin et al. |
| Citations | 136 (OpenAlex, at import time) |
| References | 72 |
| Model or method | deep neural network (unnamed) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1109/access.2020.3034287 |
| Open access | gold |
| Keywords | MRI reconstruction, deep learning, fastMRI, compressed sensing, knee imaging |
Importance rating
significant — thousands to millions affected
Accelerating MRI reconstruction could cut scan times and widen imaging access, a clear significant benefit, but the abstract gives no scale evidence beyond top rankings on the fastMRI knee challenge.