Pairing accelerated MRI sequences with deep learning-based reconstruction significantly reduces total scan time while maintaining image quality, research published Wednesday suggests.
Providers rely on MRI to evaluate most shoulder injuries, but many shorter sequences suffer tradeoffs in noise and image resolution. A new, commercially available pipeline, however, may solve that problem.
Radiologists and other researchers from South Korea found a deep learning-driven reconstruction (DLR) tool slashes MRI exam times by 67%, down from more than 9 minutes to just over 3 minutes.
What’s more, independent readers found the new method maintained image quality compared to standard sequences, the authors reported in AJR.
“The findings indicate a role for DLR to facilitate [the] application of accelerated sequences for clinical shoulder MRI, providing substantial time savings with preserved image quality and diagnostic performance,” Seok Hahn, MD, with Haeundae Paik Hospital in Busan, South Korea, and colleagues noted.
For their investigation, Hahn et al. instructed MRI technologists to use the fast sequence—periodically rotated overlapping parallel lines with enhanced reconstruction or PROPELLER—instead of standard approaches when artifacts appeared on patients’ scans.
In total, the retrospective study included 105 participants who underwent 3T shoulder MRIs. Two rads looked over image sets performed via standard, accelerated without DLR and accelerated with DLR sequences.
Rads gave higher scores to MRIs performed with the reconstruction method, using both accelerated and standard sequences. Accuracy, sensitivity and specificity for imaging shoulder tears were not different between the datasets, according to the authors.
To the group’s knowledge, this is the first study to analyze DLR for accelerated shoulder MRI in real-world settings.
Read the entire study here.