Criticality is considered a dynamic signature of healthy brain activity that can be measured on the short-term timescale with neural avalanches and long-term timescale with long-range temporal correlation (LRTC). It is unclear how the brain dynamics change in adult moyamoya disease (MMD). We used BOLD-fMRI for LRTC analysis from 16 hemorrhagic (
Moyamoya disease (MMD) is a chronic cerebrovascular disease characterized by progressive stenosis and occlusion of the terminal portion of the bilateral internal carotid arteries and their main branches. The vascular pathology leads to widespread and continuous cerebral hypoperfusion and gradual formation of compensation from collaterals as a response [
Healthy brain networks in the resting state are generally characterized by well-balanced excitatory and inhibitory synaptic activities [
LRTC is another notable measurement to evaluate the critical dynamics of the neuronal system [
Clinically, rapid and accurate differential diagnosis between acute ischemia and hemorrhage is crucial for early medical and interventional treatment but the optimal time window of treatment is often missed because of its high reliance on a CT or MR scan. The MMD is expected as the promising disease template to develop a more rapid and accurate differential diagnostic tool. According to one published fMRI study of ours, a dynamic measurement of entropy was proposed as an index of the critical dynamics to describe quantitatively the spatiotemporal changes of neural communication in adult MMD [
This study was approved by the Institutional Review Board in our hospital and conducted in accordance with the Helsinki declaration. Informed consent was signed by all the subjects of this study. From March 2017 to August 2018, 50 adult patients with MMD (16
EEG data were acquired at a sampling rate of 1000 Hz in a sound-attenuated room by using a 64-channel actiCHamp Brain Products recording system (Brain Products GmbH Inc., Munich, Germany). The impedance of all channels was below 10 K
All MRI data were collected using a 3.0 Tesla scanner (GE Healthcare, GE Asian Hub, Shanghai, China) with a 32-channel intraoperative head coil. The fMRI data were acquired using gradient echo-planar imaging with the following parameters: 3.2 mm slice thickness, 2000 ms repetition time, 30 ms echo time, 90° flip angle,
The time sequence of each channel was translated from amplitude to mean frequency by the short-term Fourier transform (Figures
Definition and evaluation of events and avalanches in the three groups. (a) An example showing a 60-second span of EEG amplitude data from a subject and the corresponding mean frequency. (b) Segmenting data with four avalanches as examples. A cascade was defined as continuous events as time bins end with no-event time bins. The number of events in the avalanche defined the avalanche size. The number of time bins was defined as the duration. The avalanche sizes in the figure were 3, 8, 9, and 5; the durations were 2, 2, 3, and 1, respectively. (c–e) The probability density distribution (PDF) and cumulative probability density distribution (CDF) of the avalanche size (
For each channel, the threshold for events was defined as the mean
As shown in Figures
The
The value of this parameter at approximately 1 indicates a critical state of the system, while values above or below 1 indicate the super- or subcritical states, respectively.
The branching parameter
Detrended fluctuation analysis was used to calculate the Hurst exponent of LRTC [
The voxel-wise Hurst exponent was adopted to evaluate the LRTC of fMRI data through the classical rescaled range (R/S) analysis [
The
The
Calculating all the subtime series, we obtained the sequence
For EEG data analysis, one-way ANOVA was used and all the pairwise comparisons were assessed using Tukey-Kramer’s multiple comparison method. For fMRI data analysis, one-sample
Table
Demographic features of the 3 groups.
Index | Controls | |||
---|---|---|---|---|
fMRI cohort | ||||
Subjects, | 16 | 34 | 25 | / |
Age (years) | 0.815 (0.309) | |||
Male (%) | 6 (37.5) | 16 (47.1) | 12 (48.0) | 0.509 (0.775) |
Education (years) | 0.729 (0.694) | |||
MMSE | 22.076 (<0.001) | |||
EEG cohort | ||||
Subjects, | 11 | 13 | 21 | / |
Age (years) | 0.326 (0.858) | |||
Male (%) | 4 (36.4) | 5 (38.5) | 12 (57.1) | 2.057 (0.357) |
Education (years) | 0.408 (0.815) | |||
MMSE | 2.305 (0.316) | |||
WM accuracy | 0.670 (0.7165) |
Figure
Differences in critical dynamics among the three groups in both the resting and working memory states. Rows (a–c) show the differences in neuronal avalanches among the three groups in the EC, EO, and WM states, respectively. Row (d) exhibits how critical dynamics transfer from different states in the three groups.
The three groups exhibited significant differences in both
The three groups exhibited a similar transition tendency between different states in all parameters (Figure
The alpha band signal was taken for Hurst exponent analysis. Figure
Hurst exponent mapping of the three groups with reference to the alpha band signal. In order to characterize the EEG amplitude dynamics of alpha oscillation, the data were bandpass filtered from 8 to 13 Hz and the amplitude envelope of the oscillations was extracted using the Hilbert transform. (a) Shows the amplitude envelope of the three groups in the EC state. The controls exhibit a regular oscillation, while the patients tend to present with a more random and rapidly changing amplitude. (b) Shows the fluctuation function of different time bin lengths from 1 to 10 seconds; the slope is defined as the Hurst exponent and represents the long-range temporal correlation. Regarding the slope, the white noise is equal to 0.5, while the controls reached 0.87. (c) The Hurst exponents of the three groups were mapped in the resting and WM states and are presented in rows 1 to 3. (d) The channels marked with red circles are those that show significant differences among the three groups in the EC, EO, and WM states, respectively. (e) The mean Hurst exponent value of the generated channels in Figure
The Hurst exponents of all the recording channels were calculated and mapped in Figure
When switching from the EC to EO state, 18 channels had significant Hurst exponent changes in the
Comparison of long-range temporal correlation between controls and subtypes of MMD separately under state transition. Channels showing significant differences between controls and moyamoya subtypes in Hurst exponent changes during state transition are marked in red in the upper panel and listed in the lower panel.
The Hurst exponent patterns of the three groups are presented in Figures
Long-range temporal correlation patterns of the three groups. The Hurst exponent patterns of the controls (a),
Compared with controls, significant decreases in the Hurst exponent in the
Regional LRTC differences among the three groups. The control group exhibited significantly higher Hurst exponent values than the
Regional LRTC differences between each pair of the three groups.
Brain regions | MNI coordinates (mm) | |||||
---|---|---|---|---|---|---|
BA | Vol (mm3) | Maximum | ||||
Left MOG | 19/39/37 | 783 | −30 | −75 | 24 | 4.627 |
Left SMA | 6 | 567 | −12 | −6 | 78 | 4.533 |
Left PCu | 7/5 | 324 | −15 | −63 | 63 | 4.389 |
Left SPG | 5/7 | 405 | −18 | −63 | 63 | 4.402 |
Left DLPFC | 6 | 918 | −15 | −3 | 78 | 4.311 |
Left DLPFC | 6 | 675 | −18 | −6 | 78 | 5.058 |
Left SMA | 6 | 459 | −12 | 0 | 78 | 4.808 |
Right PoCG | 4/3 | 540 | 15 | −30 | 75 | 4.606 |
Right DLPFC | 6/8 | 972 | 33 | 0 | 63 | 4.458 |
Right ITG | 19/37 | 702 | 48 | −69 | −6 | 4.232 |
Right PreCG | 4/6 | 405 | 15 | −27 | 75 | 4.137 |
Left PreCG | 6 | 540 | −51 | 3 | 27 | 4.156 |
The
Since head micromovements could introduce artefactual interindividual differences in resting-state fMRI metrics [
The EEG channel placement was projected onto the cortical surface and converted to the Talairach Stereotactic System based on a published Brodmann’s area (BA) atlas [
Neuroanatomical visualization showing the LRTC pattern colocalization on EEG and fMRI in MMD as compared to controls. The EEG channel positions with significantly different Hurst exponents between MMD subtypes and controls are projected onto a Brodmann’s area template based on a published atlas. For
The criticality theory provides a novel insight into the neuronal dynamics underlying brain disorders. This study was the first to apply multiscale critical dynamics analysis to examine multimodal dynamical features in two moyamoya subtypes as compared to healthy controls. The neuronal avalanches on both fast and slow timescales were analyzed during rest and task performance, and several critical EEG features were derived. Both hemorrhagic and ischemic MMD exhibited particularly low EEG frequency activity and distinct subcritical dynamics, which could be distinguished easily from those of healthy controls. In addition, the decreased long-term correlations revealed in both high temporal (EEG) and spatial (fMRI) resolution were observed to reflect distinct neurophysiological processes associated with abnormal vascular network patterns in hemorrhagic and ischemic brains. Besides, this study provided clues for further rapid differential diagnosis between acute stroke and hemorrhage at the very early phase by use of EEG instead of CT and MR, which had greater advantages of rapidness, convenience, low cost, and radiation safe. Undoubtedly, time is the brain in treatment of acute ischemic stroke [
Previous investigations have suggested that the healthy brain in the resting state is usually characterized by well-balanced excitatory and inhibitory synaptic activities. These balanced levels of excitation and inhibition drive irregular spontaneous firing activities that exhibit scale-free avalanche distributions in the brain. Such a scale-free state can be effectively described by criticality [
When switched to the EO state, both healthy controls and MMD presented with more neural activity and the MMD group remained less active than the controls. However, the ischemic moyamoya brains demonstrate stronger neural activities than hemorrhagic ones in the EO state. When switched to the WM state, all three groups exhibited more neural activities than that in the EO state, as was expected. Interestingly, the ischemic subtype surpassed the controls, while the hemorrhagic subtype still remained the least active. All parameters of neuronal avalanches exhibited similar results and were mutually verified. For healthy subjects, this phenomenon is reasonable because working memory is a behavioral state and requires effective neuronal activity to accomplish tasks [
For healthy brains in the EC state, spatially distributed neuronal activity may oscillate in phase with each other and result in high LRTC values [
To further locate the regions with significant LRTC abnormities in MMD, we also examined BOLD fluctuations on fMRI due to the high spatial resolution of this modality. The results indicate that in the EC state, the patterns of the LRTC decreases in the two moyamoya subtypes are mutually independent but overlap in the left DLPFC of the executive control network and the left SMA of the salience network. Nevertheless, all regions of these patterns are key nodes involved in planning or direct control of movement, language, and visual information [
This study has several limitations that must be addressed. First, the EEG and fMRI data were not acquired at the same time. In order to generate a more stable and reliable result, the simultaneous EEG-fMRI technology should be used in future studies. Second, the study is based on a small sample size because completing EEG tasks is difficult for moyamoya patients with executive dysfunction. More patients are in need not only to increase the statistical power but to involve patients with Suzuki grading I–II and V–VI so as to obtain more knowledge of disease progression. Nevertheless, this study is the first of its kind to characterize the variability of brain dynamics in MMD on both short-term and long-term timescales and to show different neurophysiological features of its hemorrhagic and ischemic subtypes.
The data used to support the findings of this study are available from the corresponding author upon request.
The authors declare no competing financial interests.
Y Mao, YX Gu, and YG Yu provided the idea of the article. Y Lei, YZ Li, and LC Yu wrote the article. LZ Xu, GX Zheng, and XY Qi analyzed the data. X Zhang, L Chen, and W Zhang collected the data. Yu Lei, Yuzhu Li, and Lianchun Yu contributed equally to this work.
This work was supported by the National Natural Science Foundation of China (grant numbers 81801155, 81771237, and 81761128011), the Shanghai Science and Technology Committee (grant numbers 18511102800 and 16JC1420100), the Shanghai Health and Family Planning Commission (grant number 2017BR022), the Shanghai Municipal Science and Technology Commission 795 Major Project (grant number 2018SHZDZX01), the Program for Professor of Special Appointment (Eastern Scholar) (grant number SHH1140004), and the Fundamental Research Funds for the Central Universities (grant number lzujbky-2015-119).
Figure S1: the schematic instruction of the process of EEG data. EEG data was preprocessed and then transferred to mean frequency data to extract the cascade through event detection. The cascade size and duration were qualified to provide critical information used for clinical diagnosis. The artifact-free data was also bandpass filtered to measure the long-range temporal correlation altered in the patients. Figure S2: the EEG recording procedure. The EEG recording was started with a 5-minute eyes-closed (EC), a 5-minute eyes-open (EO) resting state, and then 30 trials of a delayed-response working memory (WM) task for around 20 minutes and ended with a procedure composed of a 5-minute EC and a 5-minute EO resting state to examine the consistency of the data recording quality. Figure S3: abnormal functional network connectivity mode of the three groups (alpha band). WM: working memory. Figure S4: Difference of dynamic (left) and static (right) mean frequencies with nonoverlapping 10 seconds’ epoch of EEG data among the three groups of hemorrhagic MMD, ischemic MMD, and healthy control in the resting and working memory states. Figure S5: differences of head motion among the three groups. The mean framewise displacement (FD) derived from Jenkinson’s relative root mean square (RMS) algorithm [