The recent observation of central veins in MS lesions, also referred to as the central-vein-sign (CVS), has been recognized as a promising new research direction in the MRI-based study of Multiple Sclerosis (MS)¹. To this end, a fusion of T2-FLAIR² and 3D-T2* weighted (T2*w) images³, also referred to as FLAIR*, is commonly used to assess the relative location of veins (T2*w) and lesions (FLAIR). However, due to the way k-space is sampled, 3D-sequences are intrinsically sensitive to movements with motion artifacts that are often not immediately apparent on the images. In particular, the visibility of small veins on T2*w images is likely affected by motion⁴, potentially resulting in false negative or even false positive findings of the CVS. This is problematic because patients with neurodegenerative diseases are known to have more difficulties holding still during an MRI exam than normal controls, potentially introducing an entirely motion-related bias into CVS-based studies. The purpose of this work is to study the detriment of small vein visibility on T2*w imaging caused by subtle motion. To be able to study different amounts of motion in the same dataset, we performed a post hoc numerical simulation of head motion. Translations in the spatial and the Fourier domains are connected through a phase-gradient in the respective other domain. This means that translation in the spatial domain (movement) can be simulated by adding a linear phase gradient to the k-space.

Data acquisition: We acquired a high-resolution dataset from a healthy subject at 7 Tesla using a triple-echo 3D-T2*w sequence (Siemens MAGNETOM, 32-channel coil, matrix 512×512×208, 0.32x0.32x1.2 mm³, TE1/TR=8ms/26 ms, TA=10:17). The multi-channel data was combined using COMPOSER⁵.

Simulation: Our simulation assumed that translational motion occurred in a discrete fashion between successive phase encoding steps (TR); motion during the read-out (few ms) was neglected. We simulated four different pseudo-random motion trajectories (M1-4;Table 1) in the transversal plane (assuming negligible axial motion) defined by three parameters: motion amplitudes in x and y directions (standard-deviation of Gaussian distribution), respectively, and frequency of the movements. Considering a forward sampling of k-space (phase-1 first, then phase-2) we partitioned the k-space of the first echo image into segments without motion and simulated different head positions within the respective segments (movements) by adding phase gradients to the segments.

Analysis: Two blinded raters assessed the visibility of veins in 20 selected regions-of-interest (ROI; 80-120 mm² ) that showed between 0 and 3 veins on the original magnitude image. The ROIs covered in total 26 veins. The ROIs were padded with zeros (to 512x512 pixels) and presented to the raters in a randomized fashion, who were asked to count the number of veins visible in every ROI. We assessed the ratings in a self-calibrated fashion by using the presented motion-free counts as the ground-truth of the respective rater.

Figure 1 illustrates exemplary slices of the original (M0) and motion corrupted (M1-4) images along with some representative ROIs used for the blinded assessment. Inter-rater agreement was excellent for non-motion images (M0; 3 veins difference between raters) and declined with added motion: M1: 5; M2: 3; M3: 10; M4: 4. Figures 2 and 3 show false-negative (veins missed in M1-4 compared to M0) and false-positive (more veins detected in M1-4 compared to M0) detections. Both errors increased with added motion and obtained their maximum in the most severe motion condition (M3).Our results indicate that even subtle sporadic motion (below 1 mm movement every 26 seconds for a 3:50 scan, as suggested in reference 6) can result in artifacts on T2*w images that substantially degrade the visibility of small veins. The blurring effect of the motion may be severe enough to mask the central vein in MS lesions and thus result in an erroneous classification of a lesion as non-CVS (false-negative). However, most importantly, the overall visible image degradation may not be severe enough to justify rejection of the scan in QA procedure (fig.1). Our results also indicate that motion artifacts can be misinterpreted as small veins, potentially leading to false-positive classifications. We recommend that studies on the CVS should account for the effect of motion on the visibility of veins. This involves using effective means to avoid movement artifacts during the scan. In particular, research studies need to ensure that QA of T2*w images is performed and a low threshold is applied for rejection of a scan. Due to the difficulty in identifying subtle motion on 3D T2*w images, ideally only groups with potentially similar motion patterns should be compared to avoid bias.