Fast multi-contrast magnetic resonance imaging of the brain : a diagnostic performance study

Abstract

Background: Fast MRI is beneficial in time-critical situations like ischemic stroke and motion-prone patients and to meet the increasing clinical demand. Two fast and motion-robust, single-scan multi-contrast MRI techniques, EPIMix and NeuroMix, have been developed by the MR-physics group at the Department of Neuroradiology, Karolinska University Hospital. EPIMix visualizes the whole brain with echo-planar-based two-dimensional axial images in just above one minute. The improved method NeuroMix renders, in addition to echo-planar images, even fast spin echo and three-dimensional images in 1.5-2.5 minutes, providing higher image resolution and fewer artifacts.

Purpose: Evaluation of the diagnostic performance of EPIMix and NeuroMix against a reference standard.

Population: Patients with suspected brain pathology (Paper I) or suspected acute ischemic stroke (Paper II-IV).

Methods: In all studies, a consecutive paired cross-over multi-reader multicase study design was used to measure diagnostic accuracy. MRI of the brain with both cMRI (5-20 min) and EPIMix (1.5 min) or NeuroMix (2.5 min) were acquired and analyzed by blinded readers. The primary outcome, including sensitivity and specificity against the reference standard, was evaluated by the area under the receiver operating characteristic curve (AUC) and compared by DeLong’s test. Interrater and intrarater agreements were evaluated using kappa statistics. EPIMix sensitivity was further tested for non-inferiority compared to the reference standard in Paper III.

Main results:

Paper I: In a prospective cohort of 82 patients, the diagnostic performance to categorize brain MRI as abnormal or normal was high for EPIMix with 93-95% sensitivity, 100% specificity and an AUC of .99. Disease categorization was congruent between EPIMix and clinical routine MRI in 90% (Reader 2) and 93% (Reader 1). Image quality was rated lower for EPIMix (P < .001). The interrater agreement was almost perfect; κ =.89. Paper II: In a retrospective cohort of 172 patients, 80 (47%) had acute infarctions, 60 (35%) were normal and 32 (19%) had other pathology. EPIMix had a sensitivity of 88–91% and a specificity of 98–100% for detecting acute infarctions. AUC was .94–.95 for EPIMix (DeLong P >.02 compared to cMRI). The interrater agreement was almost perfect; κ =.90 for EPIMix. Paper III: In a prospective cohort of 118 included patients, 30 (25%) had acute infarctions. EPIMix was non-inferior to cMRI with 90–100% sensitivity. Specificity was 98–100% for EPIMix. AUC was .94-1.00 (DeLong P > .15 compared to cMRI). Paper IV: In a retrospective cohort of 44 patients, including patients in the reperfusion treatment window of stroke, 34 (77%) had acute infarction. The sensitivity and specificity of infarct detection were 94–100% and 90–100%, and AUC > .95 for Neuromix (DeLong P > .15 compared to cMRI). Relative signal intensity in the infarcted area for NeuroMix strongly correlated with cMRI (R = .73 for rDWI, R =.83 for rT2-FLAIR). Inter- and intrarater agreement were almost perfect, κ > .88 for NeuroMix.

Conclusions: The fast MRI techniques EPIMix and NeuroMix have high diagnostic performance for the detection of brain pathology and acute ischemic stroke.