“Our study indicated that motion artifacts were independently associated with higher patient age and motor symptoms. Motion artifacts were associated with lower diagnostic test accuracy and specificity for hemorrhage detection, and they explained up to one-fifth of the incorrect AI hemorrhage predictions,” wrote lead study author Christian Hedeager Krag, M.D., who is affiliated with the Department of Radiology at the University Hospital Copenhagen-Herlev and Gentofte in Copenhagen, Denmark, and colleagues.
Three Key Takeaways
- Advancing age and limb motor symptoms increase motion artifacts.
Patients with suspected stroke who are older or have limb motor symptoms are at significantly higher risk of MRI motion artifacts, which can impair diagnostic accuracy.
- Motion artifacts may reduce hemorrhage detection accuracy.
Motion artifacts were linked to a 21 percent reduction in AI diagnostic accuracy and a 7 percent reduction in radiologist accuracy for detecting hemorrhagic lesions, though they did not significantly affect ischemic or space-occupying lesion detection.
- Protocol adjustments may help minimize artifacts. Patients at higher risk for motion artifacts may benefit from tailored MRI protocols, such as shorter sequence acquisitions or the application of deep learning-based artifact reduction techniques.
The researchers noted longer mean sequence acquisition (92.3 seconds vs. 91.9 seconds) in patients with motion artifacts. They also found that 76 percent of patients with motion artifacts had 3D MRI sequences and noted a 21 percent higher incidence of gradient recalled (GR) susceptibility-weighted imaging (SWI)/T2 sequences among patients with motion artifacts (53 percent vs. 32 percent).
“Patients with a high predicted risk of motion artifacts could be scheduled for a specific protocol with shorter sequences and deep learning artifact reduction to minimize motion artifacts,” posited Krag and colleagues.
(Editor’s note: For related content, see “Can Abbreviated MRI Have an Impact in Neuroimaging?,” “Could Deep Learning Offer Quicker Acute Stroke Detection on Brain MRI Without the Need for T2WI Sequences?” and “Can Deep Learning MRI Have an Impact in Suspected Stroke Cases?”)
In regard to study limitations, the authors acknowledged that a small number of cases involving hemorrhages and intracranial tumors thwarted a full analysis of the impact of MRI motion artifacts upon these lesions. The researchers also acknowledged that a lack of blinding to clinical information may have had an impact on reported diagnostic accuracy.
