Deep learning cuts cardiac MRI scan time 13x with real-time imaging
Achieving 13x faster cardiac scans with deep learning is impressive, but the real breakthrough is the shift from chasing acceleration factors to embedding artifact suppression directly into the acquisition pipeline.
A team of researchers (Åkesson et al.) developed a rapid online deep artifact suppression technique for real-time spiral bSSFP cardiac magnetic resonance (CMR) with blipped-CAIPI simultaneous multi-slice (SMS) imaging at 1.5 T. Traditional real-time CMR requires 10-16 slices for functional assessment, leading to prolonged scan times. SMS imaging with non-Cartesian trajectories typically relies on iterative compressed sensing (CS) reconstructions that are too slow for online use. The new method combines slice separation in k-space with a 3D U-Net for deep artifact suppression in image space, enabling real-time reconstruction.
In a study with 10 healthy volunteers, the RT-SMS acquisition was ~13x faster than breath-hold imaging (15 s vs 3 min 15 s), and reconstruction using deep artifact suppression was ~50x faster than CS (30 s vs 24 min 55 s). Deep artifact suppression consistently outperformed CS in both quantitative and qualitative image quality (p<0.001). Functional measurements (LVEDV, LVESV, RVEDV, RVESV, LVM) showed good agreement with reference-standard breath-hold imaging. This work paves the way for free-breathing, high-resolution cardiac MRI with dramatically reduced scan and reconstruction times, potentially improving patient comfort and clinical throughput.
- The 13x scan acceleration for cardiac cine imaging is a best-in-class result, but it has only been demonstrated at 1.5T on a single-site cohort—generalizability to 3T and diverse patients is unproven.
- Commercial AI MRI solutions (Siemens Deep Resolve, GE AIR Recon DL) offer 2-4x speedups with broader applicability; the new method's advantage is sequence-specific optimization, which may complicate vendor adoption.
- The shift from post-processing artifact removal to inline deep learning artifact suppression in the acquisition pipeline could become the dominant paradigm for accelerated cardiac MRI, but clinical validation at multi-center scale is the critical next step.
Why It Matters
This deep learning method could transform cardiac MRI workflow efficiency, but clinical adoption hinges on validation at 3T and across diverse populations.