2.1 Participants
We enrolled 24 migraine patients without aura and 24 controls in this study, and all patients were recruited from the outpatient clinic of Nanjing Brain Hospital. The study was approved by the medical ethics committee of Nanjing Brain Hospital, and each subject provided written informed consent. The inclusion criteria for migraine patients were based on the International Classification of Headache Disorders, 3rd edition (ICHD-Ⅲbeta) of 2013 (Headache Classification Subcommittee of the International Headache Society, 2013). All subjects were right-handed, and all systemic disorders and other neurological diseases were excluded based on both clinical interviews and structural MRI. Besides, migraineurs should stop using the drug within 1 week before MEG recording. The features of migraine clinical assessment included the onset age, headache frequency, duration of last headache attacks, accompanied symptoms, and pain intensity (VAS) were collected. The Hamilton Anxiety Scale (HAM-A) and Hamilton Depression Scale (HAM-D) were used to evaluate the subjects’ anxiety and depression symptoms.
2.5 Data Processing
The averaged data without no noise and other artifacts were marked “clean data”, and were preprocessed by removing the direct current offset, then two main neuromagnetic components were obtained. We selected 90-180ms after triggering from MEG data as a time window to analysis. All data were filtered with band-pass filters at pre-defined bandwidths of delta (1–4 Hz), theta (4–8 Hz), alpha (8–12 Hz), beta (12–30 Hz), gamma (30–90 Hz), and we use a 50Hz notch filter to eliminate power-line noise. All data were proceeded additional analyses in these bands separately. A sample data is shown in Fig. 2.
Based on previous reports, the neural network was investigated at the source level[19]. Granger causality (GC) and covariance analysis were used to estimate EC. To analyze the EC network, we localized the significant neuromagnetic activity through real-time source imaging, which was defined as the volumetric source activity (or virtual sensor waveform) over each time point, and was specifically developed and optimized to analyze activities in multiple frequency bands[23, 24]. Source activity was computed by a two-step beamformed method. The detailed method was described in our earlier articles[17, 23]. The whole brain was scanned at 6mm resolution (around 17,160 voxels/source) in this study. We analyzed the correlation of all virtual sensor signals in time-windows corresponding to the event-related magnetic fields to estimate global connectivity[17, 25]. Next, we statistically analyzed the correlation of two virtual sensor signals between two source pairs by calculating the correlation coefficient based on the following mathematical formula:
$$R({X}_{a },{X}_{b})=\frac{C({X}_{a},{X}_{b})}{{S}_{{X}_{a}}{S}_{{X}_{b}}}$$
In this formula, R (Xa, Xb) represents the correlation between two source pairs in two positions (“a” and “b”). Xa and Xb indicate the MEG signals of two paired sources, used to calculate connections. C (Xa, Xb) and SxaSxb represent the mean and the standard deviation of the signals in the two sources, respectively. Moreover, to reduce bias, we also calculated every possible connection for each dual-source pair at the source level. Notably, any two voxels less than 10mm was recognized as one voxel.
Similar to recent reports, we use multivariate GC to analyze the directivity of connections[26, 27]. It can be interpreted that if one source activity could predict another source activity in a few milliseconds, we defined the two sources connected. Otherwise, the two sources were not connected. Then, neuromagnetic networks at the source level were overlapped onto structural MRIs of individual subjects. We visualized effective connectivity based on magnetic source imaging in three views (axial, coronal, and sagittal) to analyze the excitatory and inhibitory connections, which were shown in red and blue. An excitatory connection represents a positive connection where the amplitude of the signals in a source pair is positively correlated. An inhibitory connection represents a negative connection where the amplitude of the signals in two connected sources is negatively correlated[26].The yellow point indicates the node drive to other nodes while the pink point indicates the node were driven by other nodes.
Furthermore, we also perform graph theory analysis to quantify the characteristic of neural networks at the source level. In graph theory, a graph is a mathematical representation of a network and consists of a set of nodes and edges. To exactly define and compare the global EC networks properties, we calculated degree (D), strength (S), characteristic path length (L), and clustering coefficient (C) for each source pairs. The D of a node,often used to quantify centrality, indicates the number of connections between the node and the others. The L represents the number of edges in it[28]. The S parameter is defined as the measure of all possible links to all sources. The C refers to the probability that the neighbors of this node are also connected to each other and is considered to be a measure of the local connectivity of a graph. We set a threshold as a checkpoint to guarantee the data quality. All networks with EC values above the threshold could be shown. To define the threshold, we compute the t values for all source pairs.
$${T}_{P}=R\sqrt{\frac{K-2}{1-{R}^{2}}}$$
In the equation above, Tp indicates the t value of a correlation; R indicates the correlation of a source pair; K represents the number of data points for connection. The Tp value used had a corresponding p-value < 0.01 as the thresholding for obtaining the EC network. All algorithms used above were performed using the MEG Processor software (Cincinnati, OH, USA).