The application of general anesthesia during rs-fMRI studies unavoidably interferes with resting-state BOLD signal. Hereupon, understanding confounding effects of anesthetics is of the essence in MRI acquisition, design, and data analysis in the experiments need to get involved with anesthetics. This study provides the comprehensive evaluation of voxel-wise test-retest reliability of rs-fMRI metrics and its relationship with scan length in two states of awake and under volatile gaseous anesthesia. Voxel-wise ALFF, fALFF, ReHo, and FC were chosen to estimate the reliability of resting-state BOLD response since they have been demonstrated as robust measurements and stable to noise interference 25. Reliable test-retest reliability of single subject metric is crucial to obtain imaging biomarkers that assist in detection and evaluation of developmental changes of neurological diseases 26. Nevertheless, test-retest reliability has been shown to be regionally variable across the entire brain. Indeed, higher order cortical networks including default mode network, sensory/motor network, visual network, as well as dorsal attention, ventral attention, and frontoparietal control are known to have highest reproducibility 24. Hence, this experiment was constrained within gray matter area through the entire data processing.
Two main observations were noted. First, our findings showed improvement of reliability of rs-fMRI metrics measurements under anesthesia compared to the awake state that was more prominent for the scans less than 12 minutes. Second, following non-linear logarithmic regression modeled to the rs-fMRI metrics, we showed that optimal scan length is shorter under anesthesia than in awake state. Although previous studies examined the effect of anesthesia and scan length as separate experiments on variability of resting-state BOLD signal to some extent 12,22,23,27,28, none has addressed this question by comparing test-retest reliability of rs-fMRI metrics obtained in awake and under anesthesia states and combining this factor with the impact of scan length.
To that end, we compared ICC maps obtained from rs-fMRI metrics including voxel-wise ALFF, fALFF, ReHo, and FC in awake and anesthetized states. In terms of the effect of anesthesia, our finding indicated that test-retest reliability enhances under anesthesia across the rs-fMRI metrics was more significant for FC; meanwhile, ALFF exhibited the least ICC difference between awake and under anesthesia states. Additionally, among good-to-excellent test-retest reliability of the rs-fMRI measurements, fALFF showed moderate values in both awake and anesthetized states that might be linked to the high sensitivity of this metric to detecting physiological signals compared to ALFF 24,29,30,41 . This finding is consistent with the prior literature reporting that ALFF have higher test-retest reliability than fALFF. As such, it is recommended to use fALFF, or at least combine it with ALFF, in rs-fMRI studies because of the high sensitivity of this measurement to reduce physiological noise 8.
-Effects of Anesthesia on Intracranial Physiology
Our results suggest that neurophysiology characteristic of the anesthetic agents may play a main role in reduction of BOLD signal variability under anesthesia which leads to enhancement of the test-retest reliability of rs-fMRI metrics compare to wakefulness state. It has been shown that both intravenous and inhaled anesthetic agents such as propofol and sevoflurane modulate γ-aminobutyric acid type A (GABAA) receptors, which is the fastest inhibitory neurotransmitter receptor in the central neural system. Specially, propofol one of the potent modulators of GABAA receptors, enhances the gating of the receptors and thereby reduces neural excitability. The volatile anesthetics including sevoflurane enhances GABAA receptor function which leads to increasing channel opening and inhibition enhancement at both synaptic and extrasynaptic receptors. by reducing neural excitability 13,32. Such reduction in the receptive field size under anesthesia has been revealed for somatosensory cortical and sub-cortical neurons which suppresses consciousness through actions that control sleep-wake states 11,16,33.
Our results were qualitatively comparable to the results from previous literature finding that anesthesia reduces temporal variability across midline cortical regions during loss of consciousness that was measured by standard deviation (SD) of BOLD response 14,18,19,34. A recent animal study comparing BOLD signal variability in awake and anesthetized rats showed reduction of variability across much of the brain under anesthesia. They proposed that variability can be used as a robust signature of consciousness that distinguishes anesthesia-induced unconsciousness from the awake state 17.
Several factors may impact on reliability of resting-state BOLD response, and as such, test-retest reliability can vary fundamentally between datasets ranging from poor-to-excellent ICC 35. We note that, since head motion during scanning can affect test-retest reliability that was intrinsically restricted under anesthesia, higher-order regression models including Friston 24-parameter model were employed at individual level to minimize the head motion artifacts in both states of awake and under anesthesia 5,8,36–38. In addition, other variations related to non-neuronal physiology, such as white matter (WM) and cerebrospinal fluid (CSF), that represent nuisance signals were excluded as covariate factors in both states 6. Therefore, it is unlikely that our results of ICC differences reflected motion artifacts and other physiological noises. Rather, in interpreting our finding, we affirm that improvement of test-retest reliability under anesthesia is associated with suppression of neural factor contributed to resting-state BOLD response variability. Although test-retest reliability of resting-state BOLD response improves under anesthesia, due to high reliability of rs-fMRI metrics including ALFF, ReHo, and FC, our results revealed that these rs-fMRI metrics as robust biomarkers can effectively be employed in rs-fMRI studies in either awake or anesthetized state.
-Effect of Scan length in Awake and Anesthetized States
Numerous studies have proven that scan length is one the key parameters in design of MRI acquisition that plays an important role in generating robust and stable results, particularly in the experiments candidates are anesthetized or have difficulty staying during scanning. They have reported the appropriate scan duration to achieve stable brain’s function connectivity in wakeful state 7,21–23,35,39. For instance, R.M. Birn et al. 2013 showed that test-retest reliability of functional connectivity can be significantly improved by increasing the length of imaging to 9-13 minutes or longer 22. In line with this, a previous study using machine-learning classifier suggested that minimum of 15-25 minutes rs-fMRI imaging in a single-subject is required to obtain moderate reproducibility of quantitative functional connectivity 40. Consistent with these, another study in the context of seed based functional connectivity within resting-state networks (RSN), examined the effect of scan duration using 9 distinct time points (3 to 27 minutes) on test-retest reliability and found improvement of ICC over time, until the time point where the plateau reached around 12-16 min and 8-12 min for intra-session and inter-session respectively 22. Additionally, a recent study on Human Connectome Project (HCP) computed ICC for rs-fMRI graph metrics. Depending on the sample size and the number of time points (duration of scan), they found that for large sample size (for instance, 100 subjects) most of the global and regional graph metrics were reliable for minimum scan duration of 7 minutes. For small sample size of 40 subjects, they found most of the global graph metrics were reliable in long scan duration of 14 min. However, at regional level graph metrics were reliable in the areas located at the default mode network, visual and motor areas 20.
While previous studies estimated the optimal scan duration to obtain reliable resting-state function connectivity measures, our study is the first standardizing scan length of the rs-fMRI studies in anesthetized and wakefulness states using test-retest reliability of rs-fMRI metrics. In this experiment, rs-fMRI metrics were computed for 15 time points (with 1 min interval between) of 15-min scan length individually in both states of awake and under anesthesia. For each metric, ICC measurements over 15 time points were fitted to the logarithmic function using non-linear regression (y = a + b ln (x), where x defines as scan duration (min)). The optimized scan duration was determined at the time where the standardized function reached the plateau ((|dy/dx|) < 0.01). Our results agree well with previous literature finding that test-retest reliability improves as scan duration increases 21. Furthermore, we propose that resting-state scans under anesthesia require shorter scan length to achieve stable and reliable response compared to scans in awake state. According to the optimized scan length calculated for distinct rs-fMRI metrics, we suggest that optimal scan length range of 14.7-20.3 min and 8.6-17.2 min for rs-fMRI scans in awake and under anesthesia respectively are required to ensure stable and reproducible rs-fMRI measurements. Our findings agree well with the previous study suggesting that reliability of functional connectivity can be improved by increasing the imaging duration to 13 min and longer in awake state 22. Moreover, we showed that general anesthesia affects significantly on reliability of the resting-state metrics which results in shorter optimal scan length compared to awake state. Additionally, among the quantitative maps, fALFF reaches to the optimal ICC in shorter scan duration (8.6 min under anesthesia and 14.7 min in awake) and FC in longer scan length (17.2 min under anesthesia and 20.3 min in awake) that can be useful in determination of the appropriate imaging biomarker in rs-fMRI studies with different scan lengths.
Our study for the first proposed a systematic approach standardizing rs-fMRI scan length for either wakefulness or anesthetized state that can predict the trend of reliability over any timepoint. However, prior analytical studies employed different strategies to determine the optimal scan duration that were elaborative and less effective 22,41,42. For instance, a recent study investigated ICC of dynamic function connectivity (dFC) using different TRs and following sliding window approach and trade-off between large and small window sizes. They defined the appropriate time point at the window size where the highest reliability was arisen. They proposed that dFC ICC follows an inverted U curve which gives the optimal window size in the middle range of time 43.
There are some potential limitations involved in our study. First, this is important to keep in mind that resting-state BOLD response may be modulated by number of participants included in the study. It has been shown that poor reproducibility is associated with small sample size due to the low statistical power of ICC measurements. Indeed, sample size defines the number of degrees of freedom, which is a key element in determining the statistical power of ICC measurement at a group level. Prior literature reported the minimum number of 20 subjects that permits reliable rs-fMRI measurements 20,44,45. As such, a trade-off between number of subjects and scan length is needed to achieve reliable outcomes. Hence, we suggest that smaller sample size (below 40 subjects) requires longer scan duration to obtain reproducible results as we reported in our experiment. Future research is needed to take this into account by investigating reliability with a larger sample size at different time points. Taking together, number of factors including sample size, scan duration, and effect of general anesthesia play the key roles in reliability of rs-fMRI metrics that can be optimized through an appropriate trade-off between these parameters.
Secondly, we note that our finding of improvement of reliability under anesthesia is consistent for the scan duration of 15 minutes that we acquired in our experiment. Based on non-linear logarithmic regression modeled to the ICC measurements, we showed that for all the rs-fMRI metrics at different time points, ICC is higher under anesthesia than in awake state. However, the trend of ICC for the measurements can change beyond 15 minutes. As such, ICC may drop under anesthesia than in awake at specific time point. This conclusion is convincing commensurate with the pathophysiology of anesthetics in which by increasing time the effect of the anesthetics may be diminished, or physiological noise may generate fluctuations and intervene the estimated fMRI BOLD signal.
Finally, we note that this study was focused on the scans acquired for a distinct repetition time (TR) or number of volumes per specific time. However, scan duration and number of acquired volumes are mutually dependent. On the other hand, different TR may impact BOLD signal due to different factors involved, such as time of recovery and in-flow effect. Thus, shorter TR (rapid imaging pulse sequences) would improve not only the efficiency of detecting connections, but also enhance the test-retest reliability of estimated rs-fMRI BOLD signal that can be the investigated by future research.
With increasing interest in reproducible findings using rs-fMRI, there has been growing number of studies evaluating reliability of rs-fMRI typically measured by the ICC. Recent findings from prior literature have pointed towards the low reliability of rs-fMRI BOLD response, along with the number of factors including study design and analysis decisions that play the main roles to boost the reproducibility of the measurements. Our results contribute to the body of research standardizing the reliability of rs-fMRI metrics through combining factors including application of anesthetic agents and scan duration. In the present study we comprehensively examined the reliability of rs-fMRI metrics within spontaneous low-frequency fluctuations (0.01-0.08 Hz) of brain activity from scans up to 15 minutes in duration. Our findings demonstrated improvement of reliability under anesthesia compared to awake state across the time points. Additionally, we revealed that shorter acquisitions are required to achieve comparable reliability under anesthesia. This systematic exploration of reliability of rs-fMRI measures is helpful in implication of anesthetic agents in clinical practice and longitudinal studies of patients need to get scanned under anesthesia.