Auditory stimulation
The steady-state auditory stimulus was created by 1Â kHz carrier sinusoidal wave and
modulated at 37Â Hz with 100% modulation depth 31]. Monaural auditory stimulus was presented to right and left ears for each participant
in separate sessions. The steady-state auditory stimulus was 180-s duration for each
session, and triggers were given at every second. In addition to steady-state auditory
stimulus, 100 trials of short pure-tone bursts (1Â kHz carrier frequency with 300Â ms
duration) were applied binaurally to induce auditory evoked fields (AEF) for the generation
of spatial template (see below). The sound pressure of auditory stimulus was set at
75Â dB (SPL) which was generated by digital-to-analog conversion card (D/A) conversion
card (NI USB-6259, National Instrument, Austin, Texas, USA) controlled by LabView
software (National Instruments, USA).
Subjects and tasks
Five normal subjects, numbered as N1–N5 (four males, one female, all right handed;
mean age 52.2 ± 6.9 years, ranged from 45 to 62 years) and five tinnitus patients,
numbered as P1 to P5 (four males, one female, all right handed; mean age 57.8 ± 8.8 years,
ranged from 48 to 69Â years), were recruited to participate in this experiments. Among
the tinnitus patients, three patients were right-ear tinnitus and the other two were
left-ear tinnitus patients. Subjects were asked to sit in a comfortable armchair in
a dimly illuminated electro-magnetic shielded room. All participants were requested
to participate in three auditory stimulation sessions, including one binaural pure-tone
stimulation and two monaural steady-state auditory stimulations (one for right ear
and one for left ear). Three-minute empty room measurement was also recorded for each
participant to monitor MEG background noise. All participants gave informed consent,
and the study was approved by the Ethics Committee of Institutional Review Board (IRB),
Taipei Veterans General Hospital, Taiwan. The demographic data of tinnitus patients
is provided in Table 1.
Table 1. Demographic data of tinnitus patient
MEG recordings
Cortical magnetic signals were recorded with a 306-channel (102 sensor units) whole-head
neuromagnetometer (band-pass filtered within 0.05–250 Hz; digitized at 1 kHz; Vectorview;
Neuromag Ltd., Helsinki, Finland) with subjects in sitting position. Each sensor unit
was composed of a pair of planar gradiometers and a magnetometer. The magnetometer
measured magnetic flux (), normal to the sensor unit, while the gradiometers measured two tangential derivatives
of ( and , mutually orthogonal) along the longitudinal and latitudinal directions, respectively.
Bipolar horizontal and vertical electro-oculograms (EOG) were recorded using electrodes
placed below and above the left eye and at the bilateral outer canthi to monitor eye
movement and blinks. The exact position of the head with respect to the sensor array
was determined by measuring magnetic signals from four head position indicator (HPI)
coils placed on the scalp. Coil positions were identified with a three-dimensional
digitizer with respect to three predetermined landmarks (naison and bilateral preauricular
points) on the scalp, and this data was used to superimpose MEG source signals on
individual MRI images obtained by a 3.0 T Bruker MedSpec S300 system (Bruker, Kalsrube,
Germany). The anatomical image was acquired using a high-resolution T1-weighted, 3D
gradient-echo pulse sequence (MDEFT: modified driven equilibrium Fourier transform;
TR/TE/TI = 88.1 ms/4.12 ms/650 ms, 128 × 128 × 128 matrix, FOV = 250 mm).
Complementary empirical mode decomposition (CEEMD) and creation of spatial maps for
intrinsic mode functions
The whole-head MEG signals recorded in monaural steady-state auditory stimulations
were stored in hard disk for subsequent off-line CEEMD processing. Since the planar
gradiometers have better sensitivity and localized power 9], 32], only MEG gradiometers were used for data analysis in this study 33], 34]. The two gradiometer channels, one at right hemisphere and one at left hemisphere
in the vicinity of auditory areas, presenting largest AEFs, were designated as right
and left channel-of-interest (COI) channels. The signals (180Â s in each session) recorded
from the two COIs located in both hemispheres were separately processed by CEEMD to
extract noise-suppressed SSAEF.
For each session, the MEG data set contained K sensor units (K = 102) with 2 K gradiometer channels and N (N = 1,80,000) time points. The paired gradiometers ( and , along the longitudinal and latitudinal directions), were arranged into two K × N submatrices, Bx
and By
. The data matrix B was arranged as .
The CEEMD adopts the idea of noise-assisted data analysis (NADA) by adding positive
and negative white noise in pairs to generate complementary IMFs. Each IMF of CEEMD
is the ensemble average of complementary IMFs in the same scale. For one MEG gradiometer
recording , the signal was decomposed by the following CEEMD steps 25]:
1. Set an ensemble number EN (ENÂ =Â 1,000) for the CEEMD process;
2. Set (i = 1 at the beginning of the CEEMD process), where is an MEG epoch of vector with N sampled points.
3. Generate white noise, ;
4. Add white noise to [signal-to-noise ratio (SNR)Â =Â 0.01] to obtain a pair of noise-added data and , in which and , respectively;
5. Apply EMD to and , separately, to generate two series of IMFs and , where and are two matrix, containing J IMFs obtained from and , respectively;
6. Repeat step (2) to step (5) until i reaches the preset ensemble number EN;
7. Calculate the ensemble means of IMFs , where is the ith
IMF of CEEMD.
In this study, we identified SSAEF-related IMFs based on a template-based matching
approach. A spatial map was created for each IMF in order to facilitate the IMF selection
process. The correlation coefficients between each IMF and the measured signals in
all MEG gradiometers were computed. All the correlation coefficient values were then
used to create the spatial map by multiplying the data matrix B with C T
:
(1)
where represents the correlation values between the IMFs and the data of longitudinal gradiometers,
and represents the correlation values between the IMFs and the data of latitudinal gradiometers,
respectively.
Spatial map for jth
IMF can then be created by
(2)
where contains the vector sums of the correlation values in the jth
column vector of M, which presents the topographic distribution of jth
IMF over all MEG sensors.
Selection of pertinent IMFs using K-means for reconstruction of noise-suppressed SSAEF
Since different brain areas usually have their own specialized functions, the spatial
distribution, rather than temporal waveform, was utilized for selecting SSAEF-related
IMFs. Accordingly, the correlation coefficient between the spatial map of each IMF
and the spatial template (see below) was calculated. The correlation coefficients
obtained from all IMFs were further categorized into highly-, middlely-, and lowly-correlated
groups using K-means classifier 35]. Only those IMFs belonged to highly-correlated group are chosen as SSAEF-related
IMFs and subjected to the subsequent reconstruction of noise-suppressed SSAEF. The
reconstruction process was achieved by summating the chosen IMF portions in all MEG
channels as:
(3)
where is a group contains the index number of the IMFs belonged to highly-correlated group
and B recon
is a 2K × N matrix, in which the first K rows contain the reconstructed data of longitudinal gradiometers and the other K rows contain the reconstructed data of latitudinal gradiometers. The reconstructed
magnetic fields B recon
were further filtered within 1–100 Hz to remove high-frequency spiky noise.
The auditory-induced source activities were estimated by means of minimum norm estimation
(MNE) (BrainStorm software, University of South California; http://neuroimage.usc.edu/brainstorm), with realistic head model generated from individual magnetic resonance image (MRI)
using brainVISA software (http://brainvisa.info/). The estimated neural sources were overlaid on anatomical MRI and only those cortical
surface nodes with source amplitudes survived statistical significance (p  0.05)
among total surface nodes were rendered on MRI.
Creation of right- and left-hemisphere spatial templates based on amplitude of N100m
peak in auditory evoked fields (AEF)
The present CEEMD-based approach utilized a template matching process. Spatial maps
of IMFs were correlated with the pre-defined templates to identify auditory-related
IMFs. The N100m peak in AEF has been treated analogous to N100 (or N1) peak in EEG
auditory evoked potential (AEP) 36]. Both are generated from the neural populations in the primary and association auditory
cortices, located in the superior temporal gyrus 37].
Since SSAEF is also the neuromagnetic response originated from auditory cortex, the
AEF obtained from binaural pure-tone stimulation in each subject was used to create
his/her own right- and left-hemisphere spatial templates, in order to facilitate the
selection of SSAEF-related IMFs. Figure 1a shows the channel plot of AEF in binaural pure-tone stimulation in subject I. It
can be observed that the MEG1333 (marked by red circle) and MEG242 (marked by blue
circle) which showed largest N100m peaks in the right and left hemispheres, respectively,
were designated as COIs. The magnetic fields at the latency of N100m are shown in
Figure 1b and its absolute value is shown in Figure 1c. The Figure 1c was then divided into right- (unshaded part) and left-hemisphere (shaded part) spatial
templates to facilitate the IMF selections in right and left hemispheres, respectively.
Figure 1. Spatial template creation using N100m. a The channel plot of induced AEF in subject I. b The magnetic fields of N100m in all MEG channels. c The normalized map of absolute value of b. c is separated into right (unshaded) and left (shaded) hemispheric parts for selecting SSAEF-related IMFs.
Figure 2 shows the schematic diagram for the signal processing of the proposed CEEMD-based
approach. The MEG signals recorded from right and left COIs (MEG 1333 and MEG 242)
were decomposed separately into different sets of IMFs. The right/left parts of spatial
maps generated from IMFs of right/left COI were correlated with right-/left-hemisphere
spatial template to find SSAEF-related IMFs. The signals reconstructed from SSAEF-related
IMFs of right and left COIs were summated to obtain noise-suppressed SSAEFs.
Figure 2. The schematic diagram for the signal processing of the proposed CEEMD-based approach.
Calculation of laterality index (LI)
Because the laterality index (LI) has been introduced as an effective indicator to
quantitatively describe the ASSR hemispheric asymmetry 16], the right hemispheric laterality of auditory-induced source activity was calculated
in this study. The LI was calculated as the difference of source activities in right
and left auditory areas, normalized to the current strength of their summation. The
calculation of LI is represented as follows,
(4)
in which is the operator of expected value, and I right
and I left
are the estimated auditory-induced source activities in right and left auditory areas,
respectively.