Compressed Sensing 3D-FLAIR Shortens Multiple Sclerosis MRI Without Losing Manual Lesion Detection
In multiple sclerosis, every minute spent in the scanner carries practical weight. Lesion conspicuity on 3D-FLAIR remains central to diagnosis and follow-up, yet long acquisition times can stretch examination slots, increase patient discomfort, and invite motion that blunts image quality. This single-center prospective non-inferiority study tested whether compressed sensing-accelerated 3D-FLAIR at 1.5 T could preserve lesion detection while trimming scan time against an optimized GRAPPA-based sequence.
It did.
Investigators enrolled 68 patients with known or suspected MS, with a mean age of 40.1 ± 13.3 years and 61.8% women. Each MRI examination incorporated both optimized accelerated sequences, allowing a direct comparison within the same clinical setting. The primary endpoint was manual lesion count assessed by two blinded readers, with non-inferiority defined by a 10% margin. In this framework, the compressed sensing sequence met the bar for both readers, with p values of < .0001 and .005.
Scan-Time Reduction Came With Preserved Reader Confidence
The most immediate operational difference was speed. Compressed sensing 3D-FLAIR reduced acquisition time by 31.8% compared with GRAPPA, finishing in 4:00 minutes rather than 5:52 minutes. In neuroradiology, that difference is not cosmetic. Shorter acquisitions reduce the window for patient repositioning, swallowing, and subtle motion drift, all of which can degrade lesion assessment in a disease where small periventricular or juxtacortical abnormalities may alter diagnostic confidence.
Reader performance stayed strong across both techniques. Interrater agreement was excellent for the compressed sensing sequence, with an intraclass correlation coefficient of 0.921 and 95% confidence interval of 0.837 to 0.957. GRAPPA also performed well, with an ICC of 0.851 and 95% confidence interval of 0.769 to 0.906. Subjective image quality remained high in both arms, suggesting that the scan-time savings did not come at the expense of routine interpretability.
A shorter sequence that still supports confident manual lesion counting can change how MS MRI is scheduled.
Objective Signal Metrics Favored GRAPPA, While Automation Counted More Lesions on Compressed Sensing
The image-quality story becomes more complex when objective metrics enter the frame. GRAPPA produced higher signal-to-noise ratio and contrast-to-noise ratio, which is consistent with the conventional tradeoff seen when aggressive acceleration is introduced. Compressed sensing, however, generated higher automated lesion counts and volumetry. This divergence raises a familiar imaging question: algorithmic segmentation may respond differently than human readers when texture, sharpness, or noise structure shifts between reconstruction methods.
That separation matters because the study’s manual endpoint and automated secondary endpoint do not point in the same direction. Manual reading remained non-inferior, while automated quantification favored compressed sensing. In practice, this suggests that lesion visibility to the radiologist and lesion enumeration by software are not interchangeable readouts, especially in a disease where post-processing pipelines can be sensitive to reconstruction artifacts, local contrast gradients, and thresholding behavior.
Why the Tradeoff Makes Physiologic and Technical Sense
FLAIR imaging suppresses cerebrospinal fluid signal to reveal periventricular and cortical-subcortical abnormalities against a darker background. In MS, where lesion burden often determines whether the disease has been recognized, followed, or newly active, a 3D acquisition is valuable because it offers near-isotropic coverage and multiplanar review. Acceleration methods try to preserve that diagnostic geometry while compressing k-space sampling. GRAPPA reconstructs missing data from parallel coil information, whereas compressed sensing relies on sparsity assumptions and iterative reconstruction. The result can be a scan that is faster, but also one with a different noise texture and lesion edge appearance.
Those physics are not abstract. They influence what the eye sees and what software measures.
The present dataset suggests that human readers adapted well to both sequences, even as the quantitative metrics differed. That combination may be especially relevant in busy MS pathways, where rapid imaging is desirable but manual lesion detection remains the clinical anchor for diagnosis and follow-up decisions. A shorter 3D-FLAIR also fits into broader MRI workflow constraints at 1.5 T, where scanner availability and throughput often matter as much as raw image quality metrics.
What This Means for Routine MS Work-Up at 1.5 T
This study supports compressed sensing-accelerated 3D-FLAIR as a credible alternative to optimized GRAPPA in the diagnostic work-up of MS at 1.5 T. The key strength is not only non-inferior lesion detection, but the practical gain of a substantially shorter acquisition. For radiology departments, that can mean more flexible scheduling and less dependence on patient cooperation during a long sequence. For patients, it can mean a shorter exam and potentially fewer motion-related repeats.
Still, the findings stop at a preprint stage and come from a single center. Generalizability across scanners, reconstruction settings, and reader expertise remains unproven here. The automated lesion-count discrepancy also deserves caution, because software outputs may not transfer cleanly between reconstruction types even when manual interpretation stays stable. Future work will need to determine whether the faster sequence performs equally well across broader MS populations, different vendor environments, and longitudinal follow-up studies where small changes in lesion burden drive clinical interpretation.
The important point is narrower than a generic claim of superiority. In this cohort, compressed sensing delivered a faster 3D-FLAIR without sacrificing manual lesion detection, but the technical advantages and the quantitative imaging differences do not collapse into a single result. Radiology teams adopting faster MS protocols will need to decide whether scan-time savings, reader performance, and automated analysis all need to move in the same direction before a sequence can be considered fully interchangeable.