High-speed MRI recordings of eyeball lifting, retraction and compression during blinks

Blinks occur frequently in normal life and have increasingly been linked to perceptual and cognitive effects. However, the oculomechanics of blink-related eye movements remain mostly uncharted territory. While it has been known for a long time that the eye is being pulled back into its socket during a blink due to co-contraction of extraocular muscles, this elusive eye motion has not been studied in detail due to the technical difficulties that go along with a closed eyelid. Here we use dynamic magnetic resonance imaging (MRI) to obtain videos of this motion and analyse the kinematics with the recently developed MREyeTrack algorithm. We show that the eye is not only retracted but also lifted up during a blink. For some participants we observed eyeball lifting by up to 3 mm, far exceeding the amount of translation believed to occur during natural eye movements. Slow blinks can be accompanied by large tonic rotations of up to 15°. Furthermore, we collected evidence that the co-contraction of extraocular muscles leads to a slight compression of the eyeball. These findings demonstrate the surprising complexity of ocular motility and offer new opportunities to study orbital mechanics in health and disease.

in that their main motion component is translation and not rotation. 48 In this study, we used high-speed dynamic MRI to record the blink-related eye movement from eleven participants at 49 a temporal resolution of 52-54 ms. MRI allows to image an entire cross-section of the eye, which has the advantage 50 of measuring eye movements even when the lid is closed and visualising displacements and deformations of the whole 51 eyeball. Eyeball kinematics, in particular the translational motion, were estimated from single-slice data using the 52 recently developed MREyeTrack algorithm [13]. Participants were instructed to fixate a dot which was presented at During the initial T2 weighed 3D data acquisition, participants were instructed to fixate a black dot of 0.8°diameter 77 on a grey background, so that eyeball motion could be minimized. During the dynamic bSSFP scans, participants 78 had to execute one of the three tasks Short Blink, Slow Blink or Eye Closure, each while fixating the same black dot 79 at center position. For the Short Blink task, participants were instructed to aim for a short, natural blink. Next, 80 we asked participant to make their blinks a bit longer than usual during the Slow Blink task. Finally, participants 81 were instructed to repeatedly close their eyes for a full second during the Eye Closure task. Each participant had 82 to perform each task twice, once while recording the eye motion in the axial plane and once in the sagittal plane. 83 We continuously monitored the imaging slice position between different acquisitions and adjusted the slice position 84 if necessary to ensure that the lens was visible. For stimuli presentation and data analysis we used MATLAB (The  Preprocessing Image intensities for all MR data were rescaled such that the intensities around eyeball center 88 corresponded to a value of one. This was done in order to make the metrics of the subsequently used segmentation 89 algorithm comparable across participants and sequences. Furthermore, head motion in the dynamic bSSFP scans was 90 estimated using an efficient sub-pixel image registration by cross-correlation algorithm [25].

91
Eye Tracking Eye motion in the dynamic bSSFP scans were quantified using the MREyeTrack algorithm described 92 in Kirchner et al. [13], which allows to measure both rotational and translational motion components. The algorithm 93 produces a segmentation of sclera, lens and cornea by finding the optimal projection of a 3D eyeball model for 94 each image. We introduced the following minor modifications to the algorithm. Some of the MR recordings had 95 artefacts in the anterior segment of the eyeball, in particular around the cornea. Therefore, we introduced weights 96 to the individual components of the energy functional to be minimised in MREyeTrack. The original functional 97 E = E sclera + E lens + E cornea consisted of equal contributions from sclera, lens & cornea, which we changed to 98 E = E sclera + 0.5 * E lens + 0.25 * E cornea to make it more robust to artefacts in the anterior segment of the eye. Also, 99 we constricted out-of-plane translation to a maximum of 3 mm and out-of-plane rotation completely. The obtained 100 eye motion time series data was upsampled to a 2 ms time interval via linear interpolation and then smoothed using   Data exclusion criteria MRI is prone to susceptibility artefacts at natural interfaces like that of air and tissue, 118 which can manifest themselves as a local distortion of the magnetic field and subsequent loss of anatomic structure 119 in the image. MR recordings of some participants showed a loss of anatomic structure at the anterior segment of the 120 eye. If these artefacts are small, the MREyeTrack algorithm still produces a reliable segmentation of sclera, lens and 121 cornea. In order to determine whether the segmentation was of sufficient quality, we calculated the average energy 122 functionals of sclera, lens and cornea for all participants (Table 1). Axial recordings were deemed unreliable if the 123 functionals of cornea or lens were below 1.5, while this threshold was set at 0.5 for sagittal recordings. This excluded 124 P1 and P3 from the analysis of axial recordings and P7 and P11 from sagittal recordings. To get an estimate of how precisely eyeball translation can be estimated from our data by the MREyeTrack algorithm, 128 we compared anterior-posterior eyeball motion between both eyes from axial scans of the Slow Blink task. Throughout 129 this study we refer to this motion as retraction, which is the posterior motion of the eyeball center during the fixation of 130 the target dot. Since both eyes likely receive the same neural innervation during blinks, we expected that the two eyes

139
Translational eyeball motion during blinks 140 We analysed eyeball translation along all three axes, i.e. translation from anterior to posterior which we called 141 retraction, translation from inferior to superior which we called lifting and translation from medial to lateral. We 142 observed only very little translational motion along the medial to temporal axis, which was typically in the range of 143 0.1 to 0.2 mm. No participant in any of the tasks showed medial or lateral translational motion of more than 0.5 mm. 144 We focused the remainder of our analysis of translation on the measures of retraction and lifting in the sagittal plane.  overall translation that we measured throughout the experiment was 3.33 mm during a slow blink of P3.

153
Holding state for prolonged lid closure 154 While the translational trajectories associated with blinks were fairly stereotypical when performing the same task, 155 we often observed that amplitudes for the Short Blink task were smaller compared to the Slow Blink task. For 156 even longer periods of lid closure during the Eye Closure task, there was no further increase in amplitude but 157 instead the eyes held out in a retracted and lifted state while the eyes remained closed (Fig. 3a, times following a slight decay from full amplitude. Blinks performed during the Short Blink task often did not last long 160 enough to reach this holding state, but nevertheless followed the same translational trajectory (Fig. 3c). In particular

167
An interesting observation we made in our data from sagittal scans, was that some participants exhibited very large 168 rotations for longer blinks. Most noticeably, P1 rotated upwards by up to 17°during the Slow Blink task (Fig. 4a &  2°at maximum. These excessive rotations during slower blinks also lagged the onset of eyeball retraction by 100 ms and became larger with longer blink durations. Only some participants made these excessively large rotations which 172 were also highly variable within participants and could be either up-or downwards (Fig. 4b). defined to run perpendicular to the longitudinal diameter and passing through eyeball center (Fig. 5a). Only sagittal 179 recordings were used for this analysis, because they contain less out-of-plane motion than axial recordings. Then, we 180 tracked changes in diameter during the Short Blink task. We observed a sharp, consistent decrease in longitudinal 181 diameter closely time-locked to blink onset for every single participant in the range of 0.30 to 0.85 mm (Fig. 5b). On 182 average, the longitudinal diameter decreased by 0.59 mm (SD = 0.21 mm). In contrast to that, the transversal diameter 183 remained constant for most participants and decreased only slightly for others (Fig. 5c). Even though the decrease in closure. This finding strengthens the hypothesis that the blink-related eye movement is primarily a translational 207 motion due to co-contraction of extraocular muscles and that the accompanying rotational motion is an incidental 208 derivative of the global translation [11,14,15]. In agreement with previous studies, we observed large tonic rotations 209 during longer blinks but these were not only upwards but could be downwards as well [13,14]. The observation of eyeball lifting also sheds new light on the neural innervation of extraocular muscles during blinks.

218
It had been suggested that the co-contraction of inferior and superior rectus muscles would be sufficient to explain the 219 blink-related eye movement [15], but the novel finding of eyeball lifting shows that other muscles must be involved as 220 well. Our results suggest an activation of the superior oblique in order to explain eyeball lifting. It might be possible 221 that there is an imbalance between superior oblique and its antagonist the inferior oblique, but this might not even be except the superior oblique contract during blinks [11]. Rabbits, like many mammals, have a retractor bulbi, an eye 231 muscle which is specifically responsible for eyeball retraction. Perhaps the superior oblique has taken over this task in 232 humans.

233
Our results further suggest that simultaneous contraction of the rectus muscles lead to a slight compression of the 234 eyeball. Based on the video recordings, we considered whether the decrease in longitudinal diameter really reflects 235 eyeball deformation or instead could be explained by lens accommodation. However, we are not aware of any reports 236 on accommodation during blinks and an earlier MRI study showed that the lens diameter increased by only 0.33 mm 237 on average during accommodation [29]. This is too small to explain the 0.59 mm decrease in longitudinal diameter 238 that we observed. Additionally, it seems plausible to us that the enlarged force applied by the eye muscles during 239 blinks is powerful enough to result in a slight deformation. In this context, it would be interesting to investigate  Dynamic MRI proved to be a valuable tool to study ocular motility during blinks. We obtained detailed trajectories of 245 this motion using the MREyeTrack algorithm. We found that the blink-related eye movement could be accompanied  Therefore, blink-related eye movements could be used to test the validity of different models. Future studies could 250 focus on the effect of blinking and the affiliated co-contraction on extraocular muscles and the pulley system.

251
Data availability All data are available from the corresponding author upon reasonable request.