Lying in a 3T MRI scanner induces neglect-like spatial attention bias

The static magnetic field of MRI scanners can induce a magneto-hydrodynamic stimulation of the vestibular organ (MVS). In common fMRI settings, this MVS effect leads to a vestibular ocular reflex (VOR). We asked whether – beyond inducing a VOR – putting a healthy subject in a 3T MRI scanner would also alter goal-directed spatial behavior, as is known from other types of vestibular stimulation. We investigated 17 healthy volunteers, all of which exhibited a rightward VOR inside the MRI-scanner as compared to outside-MRI conditions. More importantly, when probing the distribution of overt spatial attention inside the MRI using a visual search task, subjects scanned a region of space that was significantly shifted toward the right. An additional estimate of subjective straight-ahead orientation likewise exhibited a rightward shift. Hence, putting subjects in a 3T MRI-scanner elicits MVS-induced horizontal biases of spatial orienting and exploration, which closely mimic that of stroke patients with spatial neglect.


84
To address our research question we analyzed oculomotor behavior of 17 healthy subjects inside and outside a 85 standard 3T MRI scanner (Siemens MAGNETOM Prisma). All subjects provided their informed consent according 86 to the guidelines of our institutional ethics board. 87 Parts of our procedures resembled that of typical fMRI experiments (compare supplemental Figure S1 and 88 Methods for further details): We placed our subjects with their back on the scanner table and their head rested 89 inside a 20-channel head-coil. A mirror was mounted on top of the head-coil to provide subjects with an indirect 90 view on our custom-made black "visual search-screen". The screen was placed behind the coil inside the scanner 91 bore at about 110 cm viewing distance. Various glass-fiber-cables penetrated the screen at discrete locations 92 and allowed us presenting small visual stimuli by feeding these cables with LED light signals. Maximal target 93 distance from the screen center amounted to ±12° visual angle in the horizontal direction and to -5° to 6° in the 94 vertical direction. An IR-video-camera unit was mounted next to the mirror to record a video-signal of subjects' 95 right eye. The study was conducted in complete darkness by extinguishing all light sources and covering subjects 96 with a black blanket. This was important, as any visible stimulus that would be present in subjects' visual field, 97 could help them to suppress a MVS-induced VOR as well as to explore the visual scene and to perform the SSA 98 task in an unbiased fashion. 99 To probe for the putative MVS-effects on behavior and attention we designed an experiment comprising of 100 three consecutive phases: (i) an initial "outside 1" phase with our subjects on the scanner-table at its maximal 101 horizontal displacement from the scanner center (head coil at ~ 125 cm horizontal [0 cm vertical] displacement); 102 the strength of the magnetic field in this position is roughly 10 times smaller than at the scanner center 103 (compare supplemental Figure S1); (ii) a subsequent "inside" phase with the subject's head being in the center 104 of the scanner bore where the strength of the magnetic field is 3 Tesla; (iii) as well as a final "outside 2" phase 105 that was identical to outside 1. Accordingly, any MVS effect on behavior or attention that we would observe 106 during these three phases should be stronger for inside as compared to the two outside phases. Each phase 107 started with an initial "calibration task" for eye-tracking (see Methods section for details). We next instructed 108 subjects to fixate at a dim light-point presented in the center of their visual field for 5 seconds. Then the dim 109 light-point was switched off and the subjects had to "look straight ahead" for 60 seconds in complete darkness 110 ("looking straight-ahead task"). Finally, we asked subjects to find and fixate transient light stimuli. During this 111 "visual search task" we presented 6 search targets (each slowly fading-in over a period of 5 seconds). Locations 112 of search targets were pseudorandomized across phases and between participants. The presentation of these 113 stimuli only served to maintain the subjects' motivation to search for possible targets. Our interest was the spatial distribution of subjects' overt spatial attention, as assessed by their exploratory scan path in the absence 115 of any visual targets. Thus, for most of the time (i.e. 140seconds; total duration of visual search task: 172 116 seconds), no visible target was present and subjects were searching in complete darkness. For data analysis we 117 discarded the time periods with targets present (plus an additional grace period of 5 seconds after a light 118 stimulus was turned off; this was done to prevent carry-over effects from prior target-fixation). The search task 119 always ended with a central light stimulus presented during the last 2 seconds (also not considered for data 120 analysis). Please refer to supplemental Figure S2 for an overview over all tasks and phases of our experiment. The three task phases are represented by individual rows. Horizontal eye position data from the looking straight-125 ahead task are shown in the left column. The figure inset additionally expands parts of the inside time course to 126 better illustrate the alternation between the slow phase VOR towards the right and the fast resetting saccades 127 directed in the opposite direction. Saccade endpoints are depicted by the cyan circles and the blue lines reflect 128 the respective SSA estimates. Corresponding horizontal eye velocity traces of the looking straight ahead-task are 129 shown in the middle column. Note that the cyan peaks in these time-courses refer to individual saccades, which 130 have been removed from the eye velocity records to allow estimation of slow-phase velocity in isolation (the 131 blue lines indicate the respective horizontal VOR-estimates). Finally, 2D eye-position eye data from the visual 132 search task are depicted in the rightmost column (time periods with search targets plus 5 seconds are excluded). 133 The horizontal center of visual search is depicted by the blue vertical lines, reflecting the median of horizontal 134 saccade endpoints (cyan circles). Positive values indicate the rightward/upward direction. 135 Figure 1 illustrates the resulting eye-data of an exemplary subject throughout all tasks and phases. It shows that 136 during the inside phase the typical saw-tooth-like eye movement pattern, which is characteristic for a vestibular 137 nystagmus, was observed. It consisted of a slow rightward VOR that was accompanied by fast resetting saccades 138 in the opposite direction. This prototypical nystagmus-pattern was largely reduced for the outside 1 phase and it 139 was practically absent for the outside 2 phase (also compare our supplemental video). To express the strength of 140 the VOR quantitatively, we calculated the median de-saccaded horizontal eye-velocity (starting 5 seconds after 141 the offset of the central light stimulus). This estimate is reflected by the blue lines in the respective horizontal 142 eye-velocity time-courses shown in the middle column of Figure 1. 143 The same qualitative effects were present in all of our 17 subjects, as is shown in Figure 2. The average slow eye 144 velocity for the outside 1 phase was 0.30°/s on average (±0.49°/s standard deviation [SD]) and increased to 145 1.42°/s (±1.49°/s SD) for the inside phase. After removing the subjects from the scanner bore to outside 2, 146 average slow eye velocity decreased to -0.01°/s (±0.40°/s SD). The within-subject differences in velocity between 147 outside 1 and inside and between inside and outside 2 were significant (one-tailed paired t-test:  Across our 17 subjects, statistical analysis of subjects' exploratory spatial behavior (Fig. 3)   (rightward) horizontal VOR when lying inside a 3T MRI scanner. As compared to the outside 1&2 phases, this 216 effect was present in every single subject. The velocity of the VOR inside the MRT amounted to 1.42°/s during 217 our looking straight-ahead task. Our task instruction to fixate an "imaginary target", namely into the direction of a previously illuminated target LED in the medial plane, probably reduced VOR-velocity by about 25% as 219 compared to a situation without such imaginary fixation (cf. Schmäl et al 2000). Given this latter consideration, 220 our estimate well corresponds to the roughly 2°/s VOR reported in an earlier MVS study, which used a 221 comparable experimental setting at 3T but without imaginary fixation (Boegle et al 2016) (for further discussion 222 of our VOR results, please refer to our supplemental discussion). 223 Our experiment further revealed that lying inside the 3T MRI scanner also led to a spatially biased neglect-like 224 behavior when subjects were trying to look straight ahead and when they were performing visual search. MVS induces neglect-like alterations of spatial behavior in healthy subjects and these alterations are supposedly 264 due to its interference with the cortico-vestibular system, the same system that is typically affected in patients 265 suffering from spatial neglect. 266 Given the aforesaid, the question arises whether MVS might be a viable option for treatment of spatial neglect. 267 In this context, the observations that healthy subjects exposed to the static magnetic field of a MR scanner 268 developed a persistent horizontal nystagmus that slowly diminished but did not extinguish ( (one-tailed one-sample/paired t-test). To account for putative subject dropouts, we decided to recruit four 294 additional subjects, amounting to a total sample size of 17 subjects. 295

MRI Setting 296
We used a 3T Siemens MAGNETOM Prisma MRI Scanner to apply MVS. No radio frequency (RF) or gradient coil 297 fields were applied. Given standard subject positioning, the magnetic field vector of our MRI system pointed 298 from subjects' toes to their head. Note that this is the exact opposite direction as compared to the situation in 299 the study of Roberts and colleagues (2011). Accordingly, our MVS-induced VOR effects were in the opposite 300 horizontal direction as compared to theirs. A map of the magnetic fieldlines of our system is provided in 301 Supplemental Figure S1. This figure also details the location of our subjects' head for the outside 1&2 phases as 302 well as for the inside phase. target was always at 0°/0°. The same central target was shown during the initial 5 seconds of the looking straight 334 ahead task. Finally, the following 5 targets served for eye-calibration: 0°,0°; -6°/5°; 6°/5°; -6°/-5°; 6°/-5°. 335

Manual Responses 336
During the search task, subjects were asked to find and fixate any target that they would spot. In addition, we 337 asked them to press a button on a MRI-compatible response pad (5-Button Diamond Response Pad; Current motivation high throughout the search period. As mentioned above, we were not interested in target hits but in 340 subjects' exploratory scan paths when no target was present. Thus, we did not systematically analyze subjects' 341 responses. Still we provide an estimate of response performance for the inside and outside 2 phases, in which 342 we collected data in the vast majority of subjects (n=14 and n=15 [manual responses were not reliably recorded 343 in all subjects for technical reasons]): hit-rates were 91.7% ± 17.0% SD and 93.3%±13.8% SD, and average 344 reaction times were 3236ms±698ms SD and 2995ms±697ms SD, respectively. 345

Eye Tracking 346
The position of subjects' right eye was monitored at 50 Hz sampling rate with an MR-compatible camera with 347 integrated infrared LED illumination (MRC Systems; Model: 12M-i IR-LED). We used the ViewPoint Monocular 348 Integrator System and the View Point Software (Arrington Research, software version 2.8.3.437) to digitize the 349 eye-camera video and to obtain uncalibrated eye-position data (by means of dark pupil tracking). Eye-tracking 350 was realized on a different WIN PC that was remote-controlled through the ViewPoint Ethernet-Client running 351 on our laptop PC. Eye movement analyses were performed off-line using custom routines written in MATLAB 352 R2017b (MathWorks). In brief, eye position samples were filtered using a second-order 10 Hz digital low-pass 353 filter. A 5-point calibration was performed based on the data obtained in our calibration task. In addition, we 354 compensated for any tonic eye position offset in all other tasks as well. Compensation was performed separately 355 for each task and phase, namely by removing the (average) difference in position between the visible 356 fixation/search target(s) and eye position during target fixation(s) from the eye position record. Eye velocity was 357 calculated based on 2-point differentiation of our eye position data. Saccades were detected using an absolute 358 eye velocity threshold of >15°/s. Saccade-onset was defined as the sample prior threshold-crossing. Saccade-359 offset was defined accordingly, namely as the first sample after eye velocity dropped below the threshold 360 (compare corresponding saccade endpoints depicted as cyan circles in Figure 1). Time periods with blink artifacts 361 were excluded from analyses. To obtain our velocity estimates for MVS-induced VOR we used de-saccaded 362 horizontal/vertical eye-velocity traces (time-periods from saccade-onset to saccade-offset were treated as 363 missing values; compare middle column of Figure 1). 364

Statistical Analyses 365
Apart from the de-saccaded eye velocity estimates obtained at the inside phase, all other data of interest (and 366 their respective paired differences between phases) were normally distributed (Shapiro-Wilk-Test; p0.01). 367 Accordingly, we applied paired t-tests (alpha=5%) to these latter data, namely between inside and outside 1, between inside and outside 2, and between outside 1 & 2 phases, respectively. For statistical comparison of the 369 de-saccaded eye velocity estimates during the inside phase with both outside 1&2 phases, we log-transformed 370 the respective paired differences to ensure normal distribution (Shapiro-Wilk-Test; p0.01). These log-371 transformed data were further analyzed by means of one sample t-tests (alpha=5%). Due to clear directional 372 hypotheses concerning MVS-effects on behavior between inside and outside 1 & 2 phases we applied one-tailed 373 tests. Two-tailed tests were applied when comparing outside 1 & 2 phases and when analyzing vertical eye 374 velocity, which should be largely unaffected by MVS. Note, that all reported p-values also survived Bonferroni-375 correction for multiple testing within each measure of interest (adjusted alpha=1.7%). Effect size estimates 376 (Hedges g1±CI95%) were calculated using the Matlab Toolbox 'Measures of Effect Size' (Version 1.6; by H. 377 Hentschke and M.C. Stüttgen). 378 Data sharing and code availability 379 All data and custom code from this study will be made available by the authors upon reasonable request. instructions) amounted to 505s±58s SD in the outside 1 phase. Due to the additional search training in this 499 phase, this duration was larger than for the inside phase and for the outside 2 phase, which lasted 349s±15s SD 500 and 347s±12s SD, respectively. The average inter-phase-intervals amounted to 233s±87s SD between outside 1 501 and inside phases as well as 241s±67s SD between the inside phase and outside 2 phase. The time of subject's 502 head being in the center of the MRI-scanner was about 10 minutes on average.