Effect of Acoustic fMRI-Scanner Noise on the Human Resting State

The gradient coils of MRI scanners emit loud ASN during image acquisition. The peak volume intensity can range from 90 to 130 dB or more in the 3 T scanner types that are used in the clinic or for research (Price et al. 2001; Lee et al. 2017). The obvious discomfort caused by the noise can be mitigated by compulsory provision of earplugs to the person in the scanner, which lower the noise level by approximately 40 dB (Solana et al. 2016). However, fast gradient switching during echo-planar imaging (EPI), which is applied in most studies investigating changes in brain activity related to rest or any kind of perceptive or cognitive task, appears to cause the loudest noise. Thus, the noise remaining despite the use of earplugs still considerably impedes the optimal experimental conditions. Naturally, several attempts have been made to reduce ASN, including noise-cancellation audio systems, installation of better isolation, adjustment of gradient parameters, and introduction of new MRI concepts (Edelstein et al. 2002; Hennel et al. 1999; Cho et al. 1998; Wu et al. 2010; Weiger et al. 2013; Grodzki et al. 2012; Solana et al. 2016). However, this achieved only moderate success, as participants in the latest scanner models are still required to wear hearing protection (protecting only from airborne noise nonetheless).

Regardless, the question of whether mental states recorded in shielded MRI scanner rooms are comparable to mental states in everyday life or even in shielded EEG booths remains unresolved. In addition to ASN, the supine and strongly constricted posture of participants in a study in the scanner might also diminish the transferability of the results to all-day situations or experiments that are performed in a seated posture. As such, the effect of posture on task performance or resting state has been previously investigated. For example, Thibault et al. (2014) and Spironelli and Angrilli (2017) found that supine posture decreases beta (β) and gamma (γ) power of EEG frequencies compared to an upright seated posture. Such posture-dependent surface power alterations might be associated with changes in posterior cerebral spinal fluid volume (Rice et al. 2013). Nonetheless may these posture effects only partially generalize to the particular and heavily confined situation of subjects in an MRI scanner. However, the effects of ASN on brain activity are not well understood. While behavioral data can easily be compared inside and outside the scanner, neurophysiological data comparisons are more challenging. For example, data regarding fMRI resting-state brain activation can only be obtained inside the scanner, and only with ASN. A possibility to measure the influence of ASN on fMRI resting-state networks is to test acquisition schemes with different ASN levels (i.e., “silent” vs. “noisy”). Using this approach, Langers and van Dijk (2011) detected only marginal effects of ASN on resting-state dynamics. Unexpectedly, their “silent” acquisition scheme caused a more disturbed resting state (higher alertness and arousal) than the “noisy” (yet continuous) scheme. Hence, the authors concluded that varying noise levels and characteristics do not eliminate the confounding effect of ASN in resting-state fMRI data. A similar approach to acoustic stimulation was investigated by Yakunina et al. (2015), who observed ASN effects not only in auditory brain regions, but also in networks involving attention. In line with Langers and van Dijk (2011), Yakunina et al. (2015) suggested ASN as a nuisance factor in resting-state fMRI data.

Unlike fMRI, with EEG, it is possible to record brain activity inside and outside the scanner. Conversely, inside the scanner, there is a constant magnetic field regardless of whether an acoustically noisy sequence is running. Moreover, owing to the scanner’s need for permanent cooling, the helium pump emits acoustic noise, which does not allow complete silence as an experimental condition inside the scanner. Taken together, it is impossible to separate the factors of posture, ASN, and magnetic field using a real MRI scanner for this purpose. Another aspect is the definition of the human resting state. Before the discovery of fMRI resting-state networks, a common standard for measuring cerebral resting-state activity was a setting comprising a silent, dark, and shielded room, the typical EEG-lab setting. Considering ASN (amongst other environmental differences), it can be problematic to compare the resting state of an fMRI setting with that of an EEG setting.

In the current study, we used an MRI simulator to investigate the effect of posture and that part of the ASN that is perceived through the ears, without an actual magnetic field complicating the situation. The MRI simulator can be described as a scanner-shell with the necessary equipment to simulate an MRI measurement without actually recording data. This structure comprises a tube the same size as a real MRI, a stretcher with an electric motor to move it in and out, inside light and cooling, and the choice of several MRI-sequence noises from various manufacturers. Most importantly, recording an EEG is also possible without the need to account for magnetic field distortions. On the downside, no magnetic field implies that the well-documented effect of magnetic vestibular stimulation (MVS) on resting-state activity, induced in real MRI scanners, could not be accounted for in this study (Roberts et al. 2011; Boegle et al. 2020; Andreou et al. 2017).

For the purpose of this study, we compared the resting-state EEG of healthy participants under the experimental conditions of three postures: seated outside the MRI simulator, supine inside the MRI simulator, and supine inside the MRI simulator with a standard Siemens EPI-sequence noise overlapping with a helium pump noise (for better clarity and brevity, the conditions are hereafter termed as seated, supine, and supnoise). Note that this study does not allow disentangling posture and time (i.e. sequence) effects, since the sequence of the postures was fixed. Instead, effects observed in this study merely reflect resting state alterations of participants who are initially seated, then supine, and finally supine with the addition of ASN. We compared the EEG frequency spectrum and vigilance over time between these conditions.

Due to the lack of comparable studies, the current analysis of EEG data as a function of posture and ASN is exploratory. However, in line with Thibault et al. (2014), we expected decreased β and γ power in the supine posture compared to the seated posture. For the vigilance analysis, we anticipated lower vigilance stages to be more present than in the supine posture than in the seated posture. Of note, no hypotheses were postulated for the supnoise condition because of the available data being inconclusive.

The findings of the current study have a series of relevant implications for the interpretation of resting-state brain activity observed in an active MRI scanner. They indicate that the scanner environment along with the progress of time systematically affected organization of resting-state EEG activity and associated vigilance stages in our group of healthy young subjects. This organization and regulation of spontaneous brain-activity is on one side known to be altered by various important factors such as age (Koenig et al. 2002) and gender (Tomescu et al. 2018), and neuropsychiatric conditions such as the presence of ADHD (Strauß et al. 2018), depression (Hegerl et al. 2012), mania (Wittekind et al. 2016) or dementia (Smailovic et al. 2019). On the other side, nearly all brain information processing is affected by the brain’s background functional state (McCormick et al. 2015; Huang et al. 2017). In an MRI environment, changes in information processing that are associated with such factors as age, gender, or neuropsychiatric conditions must thus be considered as the product of an interaction between the task-demand itself, and the way this task-demand interacts with how particular groups of subjects adapt to the MRI environment.

In our analyses, we assessed the EEG power of common frequency bands. In addition, we explored topographical modulations within the frequency bands stratified by the three posture conditions. Next, we assessed differences in vigilance stages between the seated, supine, and supnoise postures, which provide a reliable measure of dynamic brain arousal changes (Huang et al. 2015). Hence, we will first discuss the power findings, followed by implications raised by the vigilance results.

The frequency-band power analysis confirmed the previously circumscribed and anticipated decrease in β and γ power in the supine compared to the seated posture (Thibault et al. 2014; Spironelli and Angrilli 2017). In addition, our data showed decreased global power in the θ band (Fig. 2F). While this runs against findings of increases in δ and θ power in the supine posture found e.g. by Spironelli et al. (2016), this difference may be explained by the fact that our subjects were not resting in a bed but in the rather unusual environment of an MRI simulator. Moreover, other than in most comparable studies, the supine posture always followed and never preceded the seated posture. Sequence effects might therefore explain some of the findings. Thus considering our particular study design, the finding of decreased θ power in supine posture, indicating a higher vigilance or lower drowsiness, appears unexpected. However, given the clear limitation of our design, such a conclusion would need an independent replication with a design with randomized sequences of posture conditions. Comparing power topographies (Fig. 2A–E), we also found occipital β and γ reduction in the supine posture compared to the seated posture, as reported by Thibault and colleagues (2014). However, we found an increase in frontal electrodes instead of a decrease in the two frequency bands, which contradicts the findings of Thibault et al. (2014). These local increases can be identified on the maps in Fig. 2D, E labelled “seated-supine,” showing blue (i.e., negative) t-values over frontal electrodes. Since the frontal increase in power in supine compared to seated can be seen in all the frequency bands we analyzed (δ, θ, and α were not analyzed in Thibault et al. 2014), this might be a finding that could be related to the particular recording setting in the MR simulator in the combination with elapsed time. Since this is a highly speculative interpretation, such a finding needs replication.

Our exploratory power analysis indicated that the introduction of ASN to the supine posture, as well as sequence effects of time that might have generated a general confound, revealed complex results. Conversely, global power was lowest in supnoise in all frequency bands except γ (Fig. 2F). The global power difference in supnoise was most likely explained by decreased power, mainly over central parietal electrodes, as illustrated by t-maps in Fig. 2A–D. Conversely, the same t-maps indicate occipital increases in power in δ and θ, to mention the most distinct. This finding might be related to the higher ratio of drowsiness in supnoise, as a consequence of the fixed recording setting order with supnoise in the last position. Furthermore, as outlined in the introduction of the EEG vigilance stages, δ and θ are more pronounced in a state of drowsiness, B2/3, despite the finding that total δ and θ power were lowest in supnoise.

Thus, not only the supine posture but also supnoise seem to interact with the power of the resting brain activity, in one or the other direction depending on the brain region. To make the image even more intertwined, frequency bands appear to be distinctly affected by the influence of posture, noise, and elapsed time. For example, δ and α power were not affected by posture and time, yet adding ASN with more elapsed time elicited power changes. In contrast, γ power did not change with the introduction of ASN, but was altered by the supine posture. In θ and β, both posture and noise possibly with the interaction of the confounded sequence effect influenced the power. To the same extent as addressed above, frequency-specific topographical changes might as well have been generated by time alone, a possibility that this study cannot rule out.

Our hypothesis that vigilance would be lower in the supine than in the seated posture was confirmed by the fact that drowsiness (B1 and B2/3) was more dominant in the supine postures. This effect would be also expected from elapsed time alone. In our view, this is one of the main implications of this study that follows from the observation that our study participants spent, on average, more than a quarter of the time in states classified as either B1 or B2/3 (Fig. 1), which are biological indices of drowsiness and transition to sleep onset (Sander et al. 2015). Behaviorally, these states are characterized by lowered attention and alertness, and correlate with a loss of control over logic of thought (Yang et al. 2010). The structure, the “mode of operation” and the mental correlates of self-organizing functional resting-state brain networks recorded in these circumstances may thus be systematically different from the states that determine our typical everyday behavior. The sometimes sought connection between variance and abnormalities in fMRI based resting-state connectivity and variance and abnormalities in individual behavior outside of the scanner (e.g., Menon 2011; Woodward and Cascio 2015; Bolton et al. 2020) may thus be mediated by brain mechanisms that regulate vigilance, and individual variance or abnormalities in these mechanisms. The fact that this particular finding might have been related to time alone raises the need for replication in future studies with counterbalanced posture and ASN settings. However, neither ASN nor time had any influence on the B1 or B2/3 ratio (Fig. 1).

In light of a possible sequence effect confound in our data, we should not ignore that some findings did not concur with what the literature would predict (Hegerl et al. 2012; Strauß et al. 2018) from a vigilance time course alone, whereas others did. Despite the fact that our design makes this unexpected observation inconclusive, it is still informative to compare our time-course of EEG changes to what the literature would predict, which might yield hypotheses for future studies about the direction of the effects. We found that stage A1 was more dominant in the supine or supnoise postures compared to seated. This indicates that the conditions in the MRI simulator were accompanied by a high alertness ratio, opposing a mere time-based hypothesis of a gradual decay of alertness with time (Olbrich 2012b). In contrast, stage A2 showed the highest ratio in seated with decreasing ratios in supine and supnoise. Yet, the lower ratio was only significant in supnoise compared with seated. In the normally expected gradual decrease of alertness in a normal environment, one would expect that A2 stages first increase at the “cost” of A1 stages, and may later be replaced by more stages of type B1 or B2/3. However, in our data, the relatively large decrease of A2 in later conditions might be explained only to a small part by an increase of stage B1, whereas the increase of stage A1 seems to account for most of this decrease. Hence, an increased A1 at the cost of A2 ratio paired with a higher B1 ratio in supine and supnoise compared to seated might reflect both a more alert rest and more pronounced drowsiness. However, which effects were driven by the interaction of time and/or posture could not be disentangled with the available data of this study. The implication that still can be taken is since stages A1 and A2 can be described as most closely related to “an awake state,” those would be the relevant stages to be included in analyses focusing on awake resting states.

With these vigilance findings, the above outlined implications gain further relevance by the observation that in our data, vigilance regulation was systematically altered depending on the experimental conditions and time. In other words, our data confirms, rather unsurprisingly, that we cannot simply observe resting-state brain activity per se, but must instead measure the way an individual’s spontaneous brain activity adapts to a given situation, which is further subject to changes with progress of time. Importantly, however, the environment that is constituted by an MRI scanner seems to present a situation in which the participant adapts in a way that is systematically different from situations that are ecologically more valid for most of the typically studied conditions. This issue should be taken into account when interpreting resting-state data and might be especially relevant in populations with systematically affected vigilance regulation (e.g., depression, attention-deficit/hyperactivity syndrome). In conclusion, we encourage the performance of future research to reinforce the much underrepresented discussion of drowsiness in resting-state fMRI studies (https://​www.​webofscience.​com/​wos/​woscc/​summary/​367095b6-7a66-4aff-bfe9-e6eb9d266cfd-02547779/​times-cited-descending/​1).