Four-Minute Stroke MRI: Deep-Learning Reconstruction Turns “Time is Brain” into “Time Saved”

What are the key takeaways of this article?
 

In the emergency suite, every minute before thrombolytics costs 1.9 million neurons. Investigators prospectively scanned 211 suspected-stroke patients with a 1.5 T MRI protocol accelerated four-fold by a proprietary deep-learning (DL) reconstruction engine shrinking table time from 14 min 18 s to 3 min 4 s.
 

Blinded neuroradiologists graded DL images “interchangeable” with the standard set for detecting acute infarcts, hemorrhage, vessel-territory involvement, and mass effect (κ ≥ 0.84). Motion artifacts actually lessened, pushing diagnostic confidence slightly higher in the ultrafast arm.
 

Operational math is equally compelling: faster scans translate to quicker door-to-needle metrics, fewer sedations, and ~30 % more daily scanner slots. Limitations include single-vendor software and need for multi-site validation, but the message is clear that DL may let MRI sprint at CT-like speed without sacrificing nuance.
 

Publication Date: February 2024

Reference: Altmann S, Grauhan N, Brockstedt L, et al. Radiology. 2024;310(2):e231938. doi:10.1148/radiol.231938

Summary By: Tauqeer