Clinical Neurophysiology 127 (2016) 790–802
Contents lists available at ScienceDirect
Clinical Neurophysiology journal homepage: www.elsevier.com/locate/clinph
Frequency characteristics of neuromagnetic auditory steady-state responses to sinusoidally amplitude-modulated sweep tones Asuka Otsuka a,b, Masato Yumoto b, Shinya Kuriki c, Takuya Hotehama a, Seiji Nakagawa a,⇑ a
Biomedical Research Institute, The National Institute of Advanced Industrial Science and Technology (AIST), Osaka, Japan Graduate School of Medicine, The University of Tokyo, Tokyo, Japan c Research Center for Science and Technology, Tokyo Denki University, Chiba, Japan b
a r t i c l e
i n f o
Article history: Accepted 2 May 2015 Available online 11 May 2015 Keywords: Magnetoencephalography (MEG) Auditory steady-state response (ASSR) Loudness modeling Sound intensity calibration Frequency characteristics Sinusoidally amplitude-modulated (SAM) sweep tone
h i g h l i g h t s Frequency characteristics of neuromagnetic auditory steady-state response (ASSR) were captured for
0.1–12.5 kHz. Strength of the ASSR obtained at constant SPL was maximum at 0.5 kHz. Corresponding loudness model was plateaued between 0.5 and 4 kHz.
a b s t r a c t Objective: This study aimed to capture the neuronal frequency characteristics, as indexed by the auditory steady-state response (ASSR), relative to physical characteristics of constant sound pressure levels (SPLs). Relationship with perceptual characteristics (loudness model) was also examined. Methods: Neuromagnetic 40-Hz ASSR was recorded in response to sinusoidally amplitude-modulated sweep tones with carrier frequency covering the frequency range of 0.1–12.5 kHz. Sound intensity was equalized at 50-, 60-, and 70-dB SPL with an accuracy of ±0.5-dB SPL at the phasic peak of the modulation frequency. Corresponding loudness characteristics were modeled by substituting the detected individual hearing thresholds into a standard formula (ISO226:2003(E)). Results: The strength of the ASSR component was maximum at 0.5 kHz, and it decreased linearly on logarithmic scale toward lower and higher frequencies. Loudness model was plateaued between 0.5 and 4 kHz. Conclusions: Frequency characteristics of the ASSR were not equivalent to those of SPL and loudness model. Factors other than physical and perceptual frequency characteristics may contribute to characterizing the ASSR. Significance: The results contribute to the discussion of the most efficient signal summation for the generation of the ASSR at 0.5 kHz and efficient neuronal processing at higher frequencies, which require less energy to retain equal perception. Ó 2015 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
1. Introduction Auditory steady-state response (ASSR) is an electrical component of neuronal activity encoding information of periodic modulations of a sound (Langner, 1992; Picton et al., 2003; Joris et al.,
⇑ Corresponding author at: Biomedical Research Institute, The National Institute of Advanced Industrial Science and Technology (AIST), 1-8-31 Midorigaoka, Ikeda-shi, Osaka 563-8577, Japan. Tel.: +81 (0)72 751 8785; fax: +81 (0)72 751 9517. E-mail address: [email protected] (S. Nakagawa).
2004; Stapells et al., 2004). The ASSR component is specifically characterized by its sensitivity to carrier frequencies (fc) (Galambos et al., 1981; Pantev et al., 1996; Ross et al., 2000, 2003; Wienbruch et al., 2006), providing a useful measure with respect to evaluating the perceptual hearing characteristics per fc, particularly at threshold levels (Aoyagi et al., 1994, 1996, 1999; Lins et al., 1996; Perez-Abalo et al., 2001; Dimitrijevic et al., 2002; Herdman and Stapells, 2003; Picton et al., 2005; Vander Werff and Brown, 2005; Scherf et al., 2006; Ahn et al., 2007). However, the relationship between the frequency characteristics of the ASSR component and hearing characteristics at the
http://dx.doi.org/10.1016/j.clinph.2015.05.002 1388-2457/Ó 2015 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
A. Otsuka et al. / Clinical Neurophysiology 127 (2016) 790–802
supra-threshold levels remains unclear (Vander Werff and Brown, 2005; Ménard et al., 2008; Zenker Castro et al., 2008), despite its possible application, for instance, to adjust the hearing aids objectively and automatically. The key technical issues concerned with the investigation of the frequency characteristics of the neuromagnetic ASSR component at the supra-threshold level involve the frequency width and frequency resolution to be tested and the accuracy in the calibration of the sound intensity at the output. In the neuromagnetic studies so far, the neuronal sensitivity has been investigated discretely for four or five fc covering the frequency range between 0.25 and 4 kHz (Pantev et al., 1996; Ross et al., 2000, 2002, 2003) or at most up to 6561 Hz (Wienbruch et al., 2006). In fact, the perception of the lower and higher frequencies might not be critical for daily life. However, it is important for higher cognitive functions such as music appreciation and discrimination of individuals (Hayakawa and Itakura, 1994; Besacier et al., 2000). Further, the stimulus tones are typically delivered at 60 or 70 dB sensation level (SL) above the subjects’ hearing thresholds, with an accuracy of ±4 dB sound pressure level (SPL) (Pantev et al., 1996) or ±10 dB SPL (Ross et al., 2000). Considering the fact that 10-dB amplification would result in a 1.5-fold augmentation of the equivalent current dipole (ECD) moment in a 60–70-dB SL range (Ross et al., 2000), it is considered critical to improve the accuracy of the sound intensity calibration method. In the present study, we examined the frequency characteristics of the neuromagnetic ASSR at the supra-threshold level for a wider frequency range at a finer resolution. Sinusoidally amplitude-modulated (SAM) sweep tones with modulation frequency (fm) of 40 Hz and a carrier frequency (fc) exponentially ascending from 0.1 to 12.5 kHz were presented as stimulus tones, taking advantage of its sweeping characteristics to test the entire frequency range en bloc at the finest resolution. The sound intensity levels of the SAM sweep tones were thoroughly equalized at the fm phasic peaks, in the range of 50–70-dB SPL with an accuracy of ±0.5-dB SPL by applying inverse filtering and real-ear measurement techniques. The magnetoencephalographic (MEG) signal was recorded in response to several kinds of SAM sweep tones that varied in the acoustic and/or presentation parameters. Responses to SAM non-sweeping tones with fixed fc were also examined. The instantaneous strength of the ECD moment at fc sweeping through the frequency/time axes was captured. Corresponding loudness model was simulated by substituting the detected individual hearing threshold into a standard formula for deriving loudness level contour (ISO226:2003(E)). The frequency characteristics of the constant SPL versus the ASSR component, and the ASSR component versus the estimated loudness model, were compared. The differences were quantified and tested statistically.
2. Methods 2.1. Stimulus A stimulus tone was generated by a swept-frequency cosine (chirp) signal ascending exponentially from 0.1 to 12.5 kHz as a function of time with a constant amplitude and a continuous phase. This paradigm was intended to transiently but sequentially activate the frequency-specific auditory cortical neurons located at logarithmic spacing (Romani et al., 1982; Pantev et al., 1988, 1989, 1994, 1995, 1996; Langner et al., 1997), at an equal timing and spacing. The amplitude of the sweep tone was sinusoidally modulated at fm of 40 Hz with 100% modulation depth and 270° of the initial phase. The duration was 5 s. A 100 ms of rise and fall time was applied to ensure a smooth onset and offset to reduce the effect of any other neuronal components such as evoked responses and
791
transient gamma band responses (Ross et al., 2002, 2005). As listed in Table 1, we prepared 17 stimulus tones in total, which were varied in acoustic parameters to examine the effect of sweep direction ([1] ascent, [2] descent), duration/velocity ([3] 5 s, [4] 15 s, and [5] 45 s), SPL ([1] 70-dB SPL, [6] 60-dB SPL, and [7] 50-dB SPL), and fm ([1] 40 Hz, [8] 32 Hz, and [9] 51 Hz) on the frequency characteristics of the ASSR. Reproducibility was tested by comparing [1] first with [3] second sessions. Non-sweeping SAM tones ([10]–[17]) were also prepared at eight specific fc (0.125, 0.25, 0.5, 1, 2, 4, 8, and 10 kHz) for a duration of 200 s. All stimulus tones were synthesized digitally on a PC (Precision M2400; DELL Inc., Round Rock, TX, USA) at a sampling rate of 96 kHz with a 32-bit resolution (Matlab; The MathWorks, Inc., Natick, MA, USA). 2.2. Stimulus delivery and intensity calibration Fig. 1 shows the sound intensity calibration procedure using a test sound, a SAM sweep tone of 45 s in duration with fc sweeping from 0.1 to 12.5 kHz (stimulus tone [9]). The test tone was delivered from a PC (Precision M6400, DELL Inc.) through an audio interface (Quad-Capture UA-55; Roland Corporation, Hamamatsu, Shizuoka, Japan), equalizer (DEQ2496; Behringer GmbH, Kirchardt, Germany), a transducer (ER2; Etymotic Research, Inc., Elk Grove Village, IL, USA), and a plastic tube of 1 m to a tip of a 2-cc cavity simulating a ear canal volume. The sound signals recorded by a probe-tube microphone (ER7c; Etymotic Research, Inc.) at the output of the transducer and at the posterior end of the cavity are shown in Fig. 1A and B, respectively. Attenuation due to the tubing effect was observed in the lower and higher fc for approximately 30-dB SPL at maximum. Fig. 1C shows the impulse response calculated from the signal shown in Fig. 1B, with the spectrum serving as an inverse filter capable of the inverse frequency characteristics of the attenuated sound signal. The inverse filter was convoluted to the signal of Fig. 1A, resulted in generating a sound signal of a concave shape. Fig. 1D and E shows the inverse-filtered sound signal delivered and recorded at the output of the transducer and at the posterior end of the cavity, respectively. The distortion was minimized to a range of ±0.5 dB in the temporal waveform, confirming that thoroughly flat intensity characteristics were achieved for the entire frequency range. The inverse filter was then applied to all the stimulus tones of [1]– [17] listed in Table 1. The stimulus delivery for the subjects was prepared in a magnetically shielded room for the left and right ears. First, a soft silicone rubber probe tube (0.95-mm optical density (OD) 0.58 mm inner diameter (ID) 76 mm long (approximately 3.000 )) (ER7-14C; Etymotic Research, Inc.) was inserted into the ear canals to a depth of approximately 30 mm from the mastoid tip to a position of 2– 3 mm away from the ear drum, where the subjects reported hearing a rustling noise, and it was then fixed with a surgical tape. The outer end of the probe tube was attached to the microphone (ER7c; Etymotic Research, Inc.). Next, 13-mm foam ear pieces (ER1-14A; Etymotic Research, Inc.) attached to the tip of a 1-m plastic tube were squeezed into the ear canals to a depth of approximately 20 mm. Of note, the frequencies of peaks and dips induced by the acoustic interference inside the ear canals were shifted systematically as a function of the distance between the sound output and the recording point (lower at the exterior and higher at the interior locations inside the ear canals), and they were moved above 12.5 kHz, the upper edge of the examined frequency, when the probe tube was inserted to a depth of approximately 30 mm. The lateral ends of the ear canals, together with the probe tubes and ear pieces, were then sealed tightly with soft silicon ear plugs (Insta-putty; Insta-Mold Products, Inc., Oak, PA, USA), which resulted in flattening the global attenuation in the lower frequencies.
792
A. Otsuka et al. / Clinical Neurophysiology 127 (2016) 790–802
1
A
10
5
10
0
10
5
10
0
10
5
10
0
10
5
10
0
0 -1 1
B
0 -1
0
10
20
30
40 45
1
C
-1
0 0.01 0.02 0.03 0.04 0.05
1 0 -1 1
E
Power (a.u.)
D
Intensity (a.u.)
0
0 -1 0
10
20
30
40 45
1
F
0 -1
0
10
20
30
40 45
101
G
Intensity (a.u.)
Time (s)
102
103
104
Frequency (Hz)
0.5 0 -0.5 0
0.1 10
10.1 20
20.1 40
40.1
Time (s) Fig. 1. Sound intensity calibration procedure using a test tone of 45 s in duration (stimulus tone [9]) with fc sweeping from 0.1 to 12.5 kHz. (A and B) Signals in temporal (left column) and spectral (right column) domains recorded by an ER7c probe-tube microphone at the output of the transducer and the posterior end of the cavity, respectively. Attenuation due to the tubing effect was observed in the lower and higher fc for approximately 30-dB SPL at maximum. (C) Impulse response calculated from (B), which served as a reference to compute an inverse filter. (D and E) Inverse-filtered signal recorded at the output of the transducer and the posterior end of the cavity. Thoroughly flat intensity characteristics were achieved for the entire frequency range. (F) Signal recorded inside of a subject’s ear canal after inverse filtering and manual fine-tuning. (G) Excerpts of 100-ms window width extracted from (F). Thick gray lines delineate the envelope amplitude-modulated at the fm of 40 Hz. No critical phasic distortion was observed in either fc or fm, except for the initial rise period of the first 100 ms.
As a test sound, an inverse-filtered chord composed of 31 pure tones with frequencies corresponding to the 31 frequency bands of the equalizer was presented (Audition 1.0; Adobe Systems Inc., San Jose, CA, USA). The remaining deviations of a few dB SPL, presumably caused by the morphological difference of the individual ear canals and experimental settings, were fine-tuned manually using the 31-band equalizer (DEQ2496; Behringer GmbH, Kirchardt, Germany) at the spectral peak power of the 31 frequency components of the chord. Fig. 1F shows the inverse-filtered fine-tuned SAM sweep tone of 45 s (stimulus tone [9]) recorded inside the subject’s ear canal. Enlarged excerpts of the signal are shown in Fig. 1G, confirming visually that no critical phasic distortion was introduced to the fc and fm by the calibration procedure, except for the initial rise period of the first 100 ms. The overall sound intensity was adjusted to 70 ± 0.5-dB SPL. The transmission delay
of 7.83 ms was identified by the time point of zero crossing of the Hilbert-transformed positive and negative envelopes, and it was compensated for by a shift of the trigger signal. The ER7c microphone was detached and removed out of the magnetically shielded room after the calibration. 2.3. Subjects Eight right-handed individuals with a mean age of 38.5 (33–47, SD = 4.79) years, normal hearing (evaluated by the standard health examination, where pure-tone audiometric test was performed at 1 and 4 kHz), and no history of otological or neurological disorder participated in the experiment as subjects. The experiment was conducted in accordance with the Declaration of Helsinki, and it was approved by the Ethics Committee of The National Institute
A. Otsuka et al. / Clinical Neurophysiology 127 (2016) 790–802
793
Table 1 List of stimulus tones and their acoustic/presentation parameters.
The tones with shaded properties were categorized to test the corresponding parameter. For instance, the direction parameter was tested by comparing [1] and [2]. Stimulus tones [1]–[9] were presented repetitively for 40 trials. Stimulus tones [10]–[17] were presented once. Recording time was calculated by (duration + 5 s of prestimulus period) times the number of trial repetition.
of Advanced Industrial Science and Technology (AIST), Japan. Informed consent was obtained from each subject after the purpose and procedures of the experiment were fully explained. 2.4. Hearing threshold measurement Before the neuromagnetic measurement, subjects’ hearing threshold was recorded. We referred to an audiometric procedure standardized by The Japan Audiological Society (1990), with some modifications to be comparable with the ASSR signal. Nine test tones were extracted from the inverse-filtered 45-s SAM sweep tone (stimulus tone [9]) for a duration of 200 ms containing eight AM cycles with fc starting at 0.125, 0.25, 0.5, 1, 2, 4, 8, and 10 kHz. These test tones were presented binaurally from 70-dB SPL down to 10-dB SPL, then up to 70-dB SPL in 2-dB steps at an interstimulus interval (ISI) of 0.5 s, and it was repeated for three sessions. The presentation order of the test tones was randomized per subject. The subjects were required to respond by moving the index finger every time they perceived the tone. 2.5. MEG measurement The MEG signals were recorded in a magnetically shielded room using a whole-head SQUID system (Neuromag-122™; Elekta Neuromag Oy, Helsinki, Finland) containing 122 gradiometers that detect orthogonal planar gradients (DBz/Dx, DBz/Dy) at 61 distinct sensor elements distributed in a helmet-shaped array over the head. The signals were recorded at 0.03–100 Hz of an analog frequency band with a sampling frequency of 512 Hz with a 24-bit resolution. Each of the SAM sweep tones was presented repetitively 40 times at an ISI of 5 s. The non-sweeping SAM tones were presented once for a duration of 200 s. The total recording times for the first session ([1]–[2], [6]–[17]) and second session ([3]–[5]) were 67.3 and 53.3 min, respectively. The second session was performed on a different day. During the recording, the subjects were requested to watch a silent film with subtitles of their own choice, and to ignore the stimulus tones in order to reduce the effect of attentional modulations (Romani et al., 1982; Linden et al., 1985; Jerger et al., 1986; Plourde and Picton, 1990; Cohen et al., 1991; Ross et al., 2004; Elberling et al., 2007; Lazzouni et al., 2009). 2.6. Signal processing The recorded MEG signals were first spatiotemporally filtered using the temporal signal space separation (tSSS) method (Taulu et al., 2005; Taulu and Simola, 2006; Taulu and Hari, 2009) (Maxfilter 2.1; Elekta Neuromag Oy). The signals were transformed
into magnetostatic multipole moments, the amplitude coefficients of harmonic basis functions capable of separating the internal and external subspaces divided by the sensor array. After the spatial decomposition, external interference was suppressed by omitting the harmonic function components of the external origin. The signals in the internal and intermediate geometrical areas were decomposed by principle component analysis (PCA) at each of the temporal subspaces of the 5-s window. The artifacts and interferences nearby the sensor array were further projected out in the temporal domain when higher correlation coefficients (p > 0.98) were obtained between the PCA components. The signals were then band-passed using a fourth-order bidirectional Butterworth filter with cutoff frequencies centered at fm ±6 Hz width and segmented into 40 epochs consisting of the stimulus presentation period (either 5, 15, or 45 s) plus 5-s prestimulus and poststimulus periods. The epochs were averaged, and the baseline was corrected using the signals of the 5-s prestimulus period. Finally, the PCA components computed from the 5-s prestimulus period were spatially projected out as noise from the entire epoch. The signals for the SAM tones of 200 s with fixed fc were segmented per fm of 25 ms, resulted in computing averaged signals of 8000 epochs. The ECD was estimated for the entire stimulus presentation period at each time sample of 2.5-ms interval by a single moving dipole estimation method with a spherical volume conductor model using all 61 sensors that covered the left or right hemispheres (Curry 6.0.20; Compumedics Neuroscan, El Paso, TX, USA). The ECD parameters were evaluated in terms of goodness of fit (>80%) and 95% confidence volume ( 0.7, p < 0.05), in order to compare the neuronal and perceptual frequency characteristics by a constant fraction of 1:1. The normalized frequency characteristics were statistically compared using a three-way repeated measures ANOVA, with factors of measure (3: ECD moment for left and right hemispheres and loudness model), fc (8), and SPL (3), and a post hoc Scheffe’s test. Finally, the difference between the neuronal and perceptual frequency characteristics was quantified by subtraction (loudness model minus ECD moment and/or ECD moment minus loudness model), yielding the frequency characteristics index of the ECD moment when the loudness model was flat, or the frequency characteristic index of the loudness model when the strength of the ECD moment was flat. The difference was quantified by logarithmic curves.
2.7. Relationship between neuronal and perceptual frequency characteristics 3. Results The relationship between neuronal and perceptual frequency characteristics was examined in the following procedure. First, a regression analysis was performed per subject to compare the detected hearing threshold with the normal equal-loudness-level contours (Table B.1, Annex B; ISO226:2003(E)). If the hearing threshold of an individual was detected at 26.0-dB SPL at 1 kHz, his/her threshold curvature was compared with the normal equal-loudness-level contours of 26.0 phon. This analysis was performed to evaluate statistically whether the detected hearing threshold was well explained in terms of the standard model. Second, the loudness levels for 50–70-dB SPL were simulated by substituting the detected hearing threshold levels into formula 4.1 (1) of the normal equal-loudness-level contours (ISO226:2003(E)). The loudness level LN of a frequency f, which has an SPL Lp, was calculated by:
Lp ¼
10 lgAf dB LU 94 dB af
where
Af ¼ 4:47 103 ð100:025LN 1:15Þ 0:4 10
T f LU 9 10
af
af is the exponent for loudness perception and LU is a magnitude of the linear transfer function normalized at 1 kHz. af and LU were defined according to Table 1 of ISO226:2003(E), while Tf, the hearing threshold, was obtained in the present study from each subject. The SPL Lp was fixed to 50-, 60-, and 70-dB SPL, as was the case for the strength of the ECD moment of the ASSR component. A regression analysis was performed per subject to compare the estimated loudness models with the normal (unequal-) loudness-level contours for 50-, 60-, and 70-dB SPL (Table B.2, Annex B; ISO226:2003(E)) per subject. Third, the amplitude growth functions of the ECD moment of the ASSR component were calculated by fitting exponential regression curves to the grand-mean ECD moment (for the left and right hemispheres inclusively) and (1) the given sound intensity levels (50-, 60-, or 70-dB SPL) and (2) the loudness models, per fc (0.125, 0.25, 0.5, 1, 2, 4, 8, and 10 kHz) and per consecutive fc width (one octave (0.125–0.25, 0.25–0.5, . . ., 8–10 kHz), two-octaves (0.125–0.5, 0.25–1, . . ., 4–10 kHz), and up to 6.25 octaves (0.125– 10 kHz)). The analyses for the fc widths were performed to evaluate the interactive function of sound intensity level or loudness model with fc. The regression curves were evaluated by adjusted R2 and significance level. Further, the grand-mean ECD moment and loudness models were normalized by fc width, where a significantly higher
Fig. 2 shows the MEG signals at each stage of the analysis procedure. The temporal courses of the raw, tSSS filtered, and band-pass filtered signals of one sensor from the right temporal area of one subject in response to the stimulus tone [1] are shown in Fig. 2A–C, respectively. The fast-Fourier transformed spectra are shown in the right column, which indicate that the ASSR component was clearly elicited at the fm in the spectral domain. The 40-epochs averaged signal is shown in Fig. 2D, confirming the augmentation of the ASSR component in the temporal domain along with the fc of the stimulus tone sweeping from 0.1 to 12.5 kHz. The PCA-filtered averaged signal is shown in Fig. 2E. A marked reduction of the noise level in the prestimulus and poststimulus periods and no morphological distortions in the stimulus presentation period were confirmed by visual inspection. Details of the PCA-filtered signal are shown in Fig. 2F, displaying a 100-ms window width containing four sinusoidal cycles modulating periodically at a 25-ms interval (i.e., at the fm of 40 Hz). Fig. 3A shows the results of the source estimation for the stimulus tone [1] overlaid on the MR image of one subject. All the estimated ECDs for the right hemisphere are displayed with gray dots. The evaluation of the ECD parameters resulted in rejecting 81.07% (SD = 12.3) of the estimated ECD that failed to satisfy the acceptance criteria. The valid ECDs were mostly localized around the auditory cortices, as displayed with the white circles. The tonotopic organization (Romani et al., 1982; Pantev et al., 1988, 1989, 1994, 1995, 1996; Talavage et al., 2004; Wienbruch et al., 2006) was not clearly captured in our results. Fig. 3B shows the time course of the source activity computed by a template ECD source model. Each dot represents the strength of the ECD moment estimated at each time sample, representing the instantaneous total activation of the neuronal ensembles responding in the phase-locking manner to the fm with fc sweeping through the time (0–5 s) and frequency (0.1–12.5 kHz). The low-pass filtered curvature is overlaid with the thick line. The ECD moments showed an increase after the stimulus onset at 0.1 kHz (0 ms), which reached a maximum at 0.5 kHz (1668 ms) and then decreased linearly toward the higher frequencies on the logarithmic scale. Fig. 4 shows the overlay of the grand-mean (n = 8) ECD moment of the ASSR component at the eight fc obtained for the stimulus tones categorized to examine the effects of (A) sweep direction ([1] ascent and [2] descent), (B) duration or sweep velocity ([3] 5 s, [4] 15 s, and [5] 45 s), (C) reproducibility ([1] first and [3] second sessions), (D) SPL ([1] 70-dB SPL, [6] 60-dB SPL, and [7] 50-dB SPL), (E) fm ([1] 40 Hz, [8] 32 Hz, and [9] 51 Hz) and (F) sweeping fc versus fixed fc ([1] SAM sweep tone and [10]–[17] SAM fixed tones). The horizontal axis represents the fc of the stimulus tones
795
A. Otsuka et al. / Clinical Neurophysiology 127 (2016) 790–802
A
10
3
10
6
3
0 6 10
0 -10 3 10
B
Field Power (fT/cm/√Hz)
0
D
Field (fT/cm)
3
C
-10 3 10 0 -10 25
3
0 -25 25
E
0 6 10
0 5 10
0 5 10
0 -25
0
-5
0
5
10
0
Field (fT/cm)
F
20
40
60
Frequency (Hz)
Time (s) 25 0 -25 0
0.1 1
1.1 2
2.1 3
3.1 4
4.1
Time (s) Fig. 2. MEG signal analysis procedure. (A–E) Raw, tSSS-filtered, band-pass filtered, 40-trials averaged and PCA-filtered signals from one sensor in the right temporal area of one subject. A clearly elicited ASSR component was observed at the fm (i.e., 40 Hz) in response to the SAM sweep tone in temporal (left column) and spectral (right column) domains. (F) Excerpts of 100-ms window width extracted from (E) showing the periodic modulation at the fm of 25-ms interval.
sweeping exponentially in time and frequency. The vertical axis represents the sensitivity of the ASSR component as a function of fc. Irrespective of the stimulus acoustic or presentation parameters, the ECD moment showed a general increase from 0.125 kHz, which reached the maximum at 0.5 kHz and then decreased toward the higher frequencies linearly on the logarithmic scale. The mean attenuation rates were 2.5 (SD = 0.87)/2.3 (0.27) from the maximum at 0.5 kHz down to 4 kHz and 2.1 (0.36)/2.0 (0.25) from 4 kHz down to 10 kHz for the left/right hemispheres. The results of the three-way repeated measures ANOVA with factors of hemisphere (2), fc (8), and stimulus tone (2 or 3 depending on the stimulus parameter) showed a main effect of hemisphere in all the examined stimulus parameters (F(1, 7) = 5.39–12.85; p < 0.01–0.05), with the right hemisphere being larger than the left hemisphere, as indicated with asterisks between the left and right columns of Figs. 4A–F. The main effect of fc was observed in all the stimulus parameters (F(7, 49) = 14.26–34.11; p < 0.01), with 0.25, 0.5, and 1 kHz being larger than 0.125, 2, 4, 8, and 10 kHz (p < 0.01–0.02). An interaction effect of hemisphere by fc was also observed in all the stimulus parameters (F(7, 49) = 2.43–3.56; p < 0.01–0.05), with post hoc significances indicated above each figure. However, no significant main effect of stimulus tone was observed for the stimulus parameters of sweep direction (A) and sweep duration/velocity (B). The reproducibility (C) was confirmed by the non-significant difference between the first and second sessions. Further, sweeping fc and fixed fc (F) was not statistically different. On the other hand, the main effect of stimulus tone was significant for SPL (F(2, 14) = 5.15; p < 0.03), with 70-dB
SPL being larger than 50-dB SPL (D), and for fm (F(2, 14) = 5.82; p < 0.02), with 40 Hz being larger than 51 Hz (E). There was a significant interaction effect of fc by fm (F(14, 98) = 4.25; p < 0.01), with a post hoc significance at 0.5 kHz between 40 and 51 Hz (p < 0.01). Notably, although the ECD moment showed the overall decrease as a function of the SPL and fm, the relative frequency characteristics remained consistent. Fig. 5 illustrates the procedure to examine the relationship between the neuronal and perceptual frequency characteristics in 50–70-dB SPL range. Fig. 5A1 shows the grand-mean (n = 8) hearing thresholds overlaid on the normal equal-loudness-level contours (ISO226:2003(E)). The hearing threshold for 1 kHz was detected at 26.0-dB SPL on average (SD = 6.2-dB SPL). The best perceptual sensitivity was obtained at 2 kHz with 25.5-dB SPL (SD = 7.8-dB SPL) and worst at 0.125 kHz with 47.7-dB SPL (SD = 4.7-dB SPL). The regression analysis between the detected hearing threshold and the normal equal-loudness-level contours showed significant regression coefficients (adjusted R2 = 0.77– 0.97, p = 0.01) in all subjects except for one, who was thus excluded from the subsequent analyses. Fig. 5A2 shows the grand-mean (n = 7) loudness levels for 50-, 60-, and 70-dB SPL, simulated by substituting the detected hearing thresholds into formula 4.1 (1) of the normal equal-loudness-level contours (ISO226:2003(E)). The individual loudness levels at eight fc were estimated in the range of 14.32–51.0 (SD = 9.88) phon, 29.09–61.36 (SD = 9.63) phon, and 48.43–71.42 (SD = 7.59) phon, respectively, when the frequency characteristics of the sound intensity level were flat at 50-, 60-, and 70-dB SPL. To note, the
796
A. Otsuka et al. / Clinical Neurophysiology 127 (2016) 790–802
Fig. 3. ECD source estimation of the ASSR component. (A) All (gray dots) and valid (white circles) results of the ECD estimation overlaid on the MR image of one subject. The mean of the valid ECD source location is indicated by a cross, which served as a template ECD source model to calculate the temporal course of the source activity per hemisphere per subject. L = left, R = right, S = superior, I = inferior, A = anterior, P = posterior. (B) Time course of the source activity (gray dot) and low-pass filtered curvature (black line). The strength of the ECD moment showed an increase and a decrease along with the fc sweeping through the time (0–5 s) and frequency (0.1–12.5 kHz).
normal equal-loudness-level contours shown in Fig. 5A1 and A2 delineate the same data, but with different concepts, either to show the SPL required to maintain the equal loudness level or the loudness level in the unit of phon perceived when sounds were delivered at an equal SPL. The modeled and normal (unequal-) loudness-level contours at 50–70-dB SPL showed significant regression coefficients (adjusted R2 = 0.87–0.99, p = 0.01) in all the examined subjects (n = 7 out of 8), supporting the idea that the estimated loudness models were well explained in terms of the standard model. Fig. 5B1 and B2 show the grand-mean ECD moment on the y-axis (=y-axis of Fig. 4D) and either the given sound intensity level (50-, 60-, or 70-dB SPL) or the estimated loudness model (=y-axis of Fig. 5A2) on the x-axis, with exponential regression curves fitted to represent the amplitude growth functions of the ASSR component by intensity (IGF) or modeled loudness level (loudness growth function (LGF)), per fc. Significantly higher regression coefficients were obtained for 0.125–8 kHz for IGF (adjusted R2 = 0.72, 0.79, 0.92, 0.86, 0.90, 0.80, 0.70; p = 0.01–0.04) and for LGF (adjusted R2 = 0.70, 0.83, 0.93, 0.86, 0.90, 0.81, 0.71; p = 0.01–0.04), indicating that the ECD moment of the ASSR component in 50–70-dB SPL range could be explained by the exponential model when IGF or LGF was examined per fc. Although slightly better fitting results were obtained for LGF than for IGF (mean adjusted R2 = 0.82/0.81, respectively), the frequency characteristics of the amplitude growth functions of the ASSR component were consistent in both measures, with 0.5 kHz exhibiting a steeper slope than for the lower and higher fc. Fig. 5B3 shows the grand-mean ECD moment on the y-axis (=y-axis of Fig. 4D) and the modeled loudness level (=y-axis of Fig. 5A2) on the x-axis, but with exponential regression curves fitted to the two-octaves fc width, where the significantly higher regression coefficients were obtained (adjusted R2 = 0.82; p < 0.01 for 0.125–0.5 kHz and R2 = 0.70; p < 0.01 for 4–10 kHz). The one-octave fc width of 0.125–0.25, 0.25–0.5, and 4–8 kHz also showed significantly higher regression coefficients (adjusted R2 = 0.73, 0.89, 0.81, respectively; p < 0.01). No significant regression coefficients were obtained for fc width of more than two octaves. Interestingly, the ECD moment between 0.5 and
4 kHz decreased, while the loudness models were mostly equal at around 50, 60, and 70 phon, respectively, indicating that the strength of the ECD moment of the ASSR component could not be explained by the single exponential function of loudness, but interactively with frequency. Fig. 5C1 and C2 show two possible interpretations for how the ECD moment and loudness model could be overlaid when they were normalized by fc width, where significantly higher regression coefficients were obtained between the neuronal and perceptual frequency characteristics. In the case where the ECD moment and loudness model were adjusted by 0.125–0.5 kHz (R2 = 0.82; p < 0.01), a large difference was induced in the higher frequencies because of the steep decrease of the ECD moment after 0.5 kHz, in contrast to the equal loudness model sustained up to 4 kHz. A significant interaction effect was observed between the ECD moment and loudness model by fc (F(14, 280) = 16.12; p < 0.01), with the ECD moment being significantly smaller than the loudness model at 2–10 kHz (p < 0.01–0.05) in both hemispheres. Conversely, in the case where the ECD moment and loudness model were adjusted by 4–10 kHz (R2 = 0.70; p < 0.01), a large difference was induced in the lower frequencies because of the precipitous growth of the ECD moment at 0.5 kHz, in contrast to the loudness model plateauing between 0.5 and 4 kHz. A significant interaction effect was observed between the ECD moment and loudness model by fc (F(14, 280) = 11.81; p < 0.01), with the ECD moment being significantly larger than the loudness model at 0.25–1 kHz (p < 0.01) in both hemispheres. Fig. 5C3 shows the possible differences between the neuronal and perceptual frequency characteristics obtained by subtractions. The ECD moment was augmented in the lower frequencies peaking at 0.5 kHz when the frequency characteristics of the loudness model were flat. By contrast, the loudness model was augmented in the higher frequencies peaking at 4 kHz when the frequency characteristics of the ECD moment were flat. The difference as a function of fc was linear between 0.5 and 4 kHz on logarithmic scale, and it was quantified by the regression curves as y = 3.68log (fc) + 2.18 (R2 = 0.98) (i.e., ECD moment when the loudness model was flat) and y = 1.5log (fc) 0.44 (R2 = 0.99) (i.e., loudness model when ECD moment was flat), respectively.
797
A. Otsuka et al. / Clinical Neurophysiology 127 (2016) 790–802
Left Hemisphere
A
Right Hemisphere
Left Hemisphere
D
Sweep Direction *** ~ *
***
***
***
0.8
***
***
SPL ***
***~**
Right Hemisphere
***
***
*** ***
***
*
***
***
[1] 70 dB [6] 60 dB [7] 50 dB
*
[1] ascent [2] descent
0
E
Duration ***
**~* ***~**
0.8
***
*
fm ***
***
*** **~*
***
*
***
[1] 40 Hz [8] 32 Hz [9] 51 Hz
***
[3] 5 sec [4] 15 sec [5] 45 sec ***
Normalized ECD Moment (a.u.)
B
0
C
F
Session ***
***
*** ***
0.8
*
***
Sweep / Fix ***
***
***
***
*** ***
*
***
***
[1] 1st [3] 2nd
***
[1] SAM chirp [10] SAM Fixed
0 102
103
104 102
103
104
102
Frequency (Hz)
103
104 102
103
104 *** ** *
p
Contents lists available at ScienceDirect
Clinical Neurophysiology journal homepage: www.elsevier.com/locate/clinph
Frequency characteristics of neuromagnetic auditory steady-state responses to sinusoidally amplitude-modulated sweep tones Asuka Otsuka a,b, Masato Yumoto b, Shinya Kuriki c, Takuya Hotehama a, Seiji Nakagawa a,⇑ a
Biomedical Research Institute, The National Institute of Advanced Industrial Science and Technology (AIST), Osaka, Japan Graduate School of Medicine, The University of Tokyo, Tokyo, Japan c Research Center for Science and Technology, Tokyo Denki University, Chiba, Japan b
a r t i c l e
i n f o
Article history: Accepted 2 May 2015 Available online 11 May 2015 Keywords: Magnetoencephalography (MEG) Auditory steady-state response (ASSR) Loudness modeling Sound intensity calibration Frequency characteristics Sinusoidally amplitude-modulated (SAM) sweep tone
h i g h l i g h t s Frequency characteristics of neuromagnetic auditory steady-state response (ASSR) were captured for
0.1–12.5 kHz. Strength of the ASSR obtained at constant SPL was maximum at 0.5 kHz. Corresponding loudness model was plateaued between 0.5 and 4 kHz.
a b s t r a c t Objective: This study aimed to capture the neuronal frequency characteristics, as indexed by the auditory steady-state response (ASSR), relative to physical characteristics of constant sound pressure levels (SPLs). Relationship with perceptual characteristics (loudness model) was also examined. Methods: Neuromagnetic 40-Hz ASSR was recorded in response to sinusoidally amplitude-modulated sweep tones with carrier frequency covering the frequency range of 0.1–12.5 kHz. Sound intensity was equalized at 50-, 60-, and 70-dB SPL with an accuracy of ±0.5-dB SPL at the phasic peak of the modulation frequency. Corresponding loudness characteristics were modeled by substituting the detected individual hearing thresholds into a standard formula (ISO226:2003(E)). Results: The strength of the ASSR component was maximum at 0.5 kHz, and it decreased linearly on logarithmic scale toward lower and higher frequencies. Loudness model was plateaued between 0.5 and 4 kHz. Conclusions: Frequency characteristics of the ASSR were not equivalent to those of SPL and loudness model. Factors other than physical and perceptual frequency characteristics may contribute to characterizing the ASSR. Significance: The results contribute to the discussion of the most efficient signal summation for the generation of the ASSR at 0.5 kHz and efficient neuronal processing at higher frequencies, which require less energy to retain equal perception. Ó 2015 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
1. Introduction Auditory steady-state response (ASSR) is an electrical component of neuronal activity encoding information of periodic modulations of a sound (Langner, 1992; Picton et al., 2003; Joris et al.,
⇑ Corresponding author at: Biomedical Research Institute, The National Institute of Advanced Industrial Science and Technology (AIST), 1-8-31 Midorigaoka, Ikeda-shi, Osaka 563-8577, Japan. Tel.: +81 (0)72 751 8785; fax: +81 (0)72 751 9517. E-mail address: [email protected] (S. Nakagawa).
2004; Stapells et al., 2004). The ASSR component is specifically characterized by its sensitivity to carrier frequencies (fc) (Galambos et al., 1981; Pantev et al., 1996; Ross et al., 2000, 2003; Wienbruch et al., 2006), providing a useful measure with respect to evaluating the perceptual hearing characteristics per fc, particularly at threshold levels (Aoyagi et al., 1994, 1996, 1999; Lins et al., 1996; Perez-Abalo et al., 2001; Dimitrijevic et al., 2002; Herdman and Stapells, 2003; Picton et al., 2005; Vander Werff and Brown, 2005; Scherf et al., 2006; Ahn et al., 2007). However, the relationship between the frequency characteristics of the ASSR component and hearing characteristics at the
http://dx.doi.org/10.1016/j.clinph.2015.05.002 1388-2457/Ó 2015 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All rights reserved.
A. Otsuka et al. / Clinical Neurophysiology 127 (2016) 790–802
supra-threshold levels remains unclear (Vander Werff and Brown, 2005; Ménard et al., 2008; Zenker Castro et al., 2008), despite its possible application, for instance, to adjust the hearing aids objectively and automatically. The key technical issues concerned with the investigation of the frequency characteristics of the neuromagnetic ASSR component at the supra-threshold level involve the frequency width and frequency resolution to be tested and the accuracy in the calibration of the sound intensity at the output. In the neuromagnetic studies so far, the neuronal sensitivity has been investigated discretely for four or five fc covering the frequency range between 0.25 and 4 kHz (Pantev et al., 1996; Ross et al., 2000, 2002, 2003) or at most up to 6561 Hz (Wienbruch et al., 2006). In fact, the perception of the lower and higher frequencies might not be critical for daily life. However, it is important for higher cognitive functions such as music appreciation and discrimination of individuals (Hayakawa and Itakura, 1994; Besacier et al., 2000). Further, the stimulus tones are typically delivered at 60 or 70 dB sensation level (SL) above the subjects’ hearing thresholds, with an accuracy of ±4 dB sound pressure level (SPL) (Pantev et al., 1996) or ±10 dB SPL (Ross et al., 2000). Considering the fact that 10-dB amplification would result in a 1.5-fold augmentation of the equivalent current dipole (ECD) moment in a 60–70-dB SL range (Ross et al., 2000), it is considered critical to improve the accuracy of the sound intensity calibration method. In the present study, we examined the frequency characteristics of the neuromagnetic ASSR at the supra-threshold level for a wider frequency range at a finer resolution. Sinusoidally amplitude-modulated (SAM) sweep tones with modulation frequency (fm) of 40 Hz and a carrier frequency (fc) exponentially ascending from 0.1 to 12.5 kHz were presented as stimulus tones, taking advantage of its sweeping characteristics to test the entire frequency range en bloc at the finest resolution. The sound intensity levels of the SAM sweep tones were thoroughly equalized at the fm phasic peaks, in the range of 50–70-dB SPL with an accuracy of ±0.5-dB SPL by applying inverse filtering and real-ear measurement techniques. The magnetoencephalographic (MEG) signal was recorded in response to several kinds of SAM sweep tones that varied in the acoustic and/or presentation parameters. Responses to SAM non-sweeping tones with fixed fc were also examined. The instantaneous strength of the ECD moment at fc sweeping through the frequency/time axes was captured. Corresponding loudness model was simulated by substituting the detected individual hearing threshold into a standard formula for deriving loudness level contour (ISO226:2003(E)). The frequency characteristics of the constant SPL versus the ASSR component, and the ASSR component versus the estimated loudness model, were compared. The differences were quantified and tested statistically.
2. Methods 2.1. Stimulus A stimulus tone was generated by a swept-frequency cosine (chirp) signal ascending exponentially from 0.1 to 12.5 kHz as a function of time with a constant amplitude and a continuous phase. This paradigm was intended to transiently but sequentially activate the frequency-specific auditory cortical neurons located at logarithmic spacing (Romani et al., 1982; Pantev et al., 1988, 1989, 1994, 1995, 1996; Langner et al., 1997), at an equal timing and spacing. The amplitude of the sweep tone was sinusoidally modulated at fm of 40 Hz with 100% modulation depth and 270° of the initial phase. The duration was 5 s. A 100 ms of rise and fall time was applied to ensure a smooth onset and offset to reduce the effect of any other neuronal components such as evoked responses and
791
transient gamma band responses (Ross et al., 2002, 2005). As listed in Table 1, we prepared 17 stimulus tones in total, which were varied in acoustic parameters to examine the effect of sweep direction ([1] ascent, [2] descent), duration/velocity ([3] 5 s, [4] 15 s, and [5] 45 s), SPL ([1] 70-dB SPL, [6] 60-dB SPL, and [7] 50-dB SPL), and fm ([1] 40 Hz, [8] 32 Hz, and [9] 51 Hz) on the frequency characteristics of the ASSR. Reproducibility was tested by comparing [1] first with [3] second sessions. Non-sweeping SAM tones ([10]–[17]) were also prepared at eight specific fc (0.125, 0.25, 0.5, 1, 2, 4, 8, and 10 kHz) for a duration of 200 s. All stimulus tones were synthesized digitally on a PC (Precision M2400; DELL Inc., Round Rock, TX, USA) at a sampling rate of 96 kHz with a 32-bit resolution (Matlab; The MathWorks, Inc., Natick, MA, USA). 2.2. Stimulus delivery and intensity calibration Fig. 1 shows the sound intensity calibration procedure using a test sound, a SAM sweep tone of 45 s in duration with fc sweeping from 0.1 to 12.5 kHz (stimulus tone [9]). The test tone was delivered from a PC (Precision M6400, DELL Inc.) through an audio interface (Quad-Capture UA-55; Roland Corporation, Hamamatsu, Shizuoka, Japan), equalizer (DEQ2496; Behringer GmbH, Kirchardt, Germany), a transducer (ER2; Etymotic Research, Inc., Elk Grove Village, IL, USA), and a plastic tube of 1 m to a tip of a 2-cc cavity simulating a ear canal volume. The sound signals recorded by a probe-tube microphone (ER7c; Etymotic Research, Inc.) at the output of the transducer and at the posterior end of the cavity are shown in Fig. 1A and B, respectively. Attenuation due to the tubing effect was observed in the lower and higher fc for approximately 30-dB SPL at maximum. Fig. 1C shows the impulse response calculated from the signal shown in Fig. 1B, with the spectrum serving as an inverse filter capable of the inverse frequency characteristics of the attenuated sound signal. The inverse filter was convoluted to the signal of Fig. 1A, resulted in generating a sound signal of a concave shape. Fig. 1D and E shows the inverse-filtered sound signal delivered and recorded at the output of the transducer and at the posterior end of the cavity, respectively. The distortion was minimized to a range of ±0.5 dB in the temporal waveform, confirming that thoroughly flat intensity characteristics were achieved for the entire frequency range. The inverse filter was then applied to all the stimulus tones of [1]– [17] listed in Table 1. The stimulus delivery for the subjects was prepared in a magnetically shielded room for the left and right ears. First, a soft silicone rubber probe tube (0.95-mm optical density (OD) 0.58 mm inner diameter (ID) 76 mm long (approximately 3.000 )) (ER7-14C; Etymotic Research, Inc.) was inserted into the ear canals to a depth of approximately 30 mm from the mastoid tip to a position of 2– 3 mm away from the ear drum, where the subjects reported hearing a rustling noise, and it was then fixed with a surgical tape. The outer end of the probe tube was attached to the microphone (ER7c; Etymotic Research, Inc.). Next, 13-mm foam ear pieces (ER1-14A; Etymotic Research, Inc.) attached to the tip of a 1-m plastic tube were squeezed into the ear canals to a depth of approximately 20 mm. Of note, the frequencies of peaks and dips induced by the acoustic interference inside the ear canals were shifted systematically as a function of the distance between the sound output and the recording point (lower at the exterior and higher at the interior locations inside the ear canals), and they were moved above 12.5 kHz, the upper edge of the examined frequency, when the probe tube was inserted to a depth of approximately 30 mm. The lateral ends of the ear canals, together with the probe tubes and ear pieces, were then sealed tightly with soft silicon ear plugs (Insta-putty; Insta-Mold Products, Inc., Oak, PA, USA), which resulted in flattening the global attenuation in the lower frequencies.
792
A. Otsuka et al. / Clinical Neurophysiology 127 (2016) 790–802
1
A
10
5
10
0
10
5
10
0
10
5
10
0
10
5
10
0
0 -1 1
B
0 -1
0
10
20
30
40 45
1
C
-1
0 0.01 0.02 0.03 0.04 0.05
1 0 -1 1
E
Power (a.u.)
D
Intensity (a.u.)
0
0 -1 0
10
20
30
40 45
1
F
0 -1
0
10
20
30
40 45
101
G
Intensity (a.u.)
Time (s)
102
103
104
Frequency (Hz)
0.5 0 -0.5 0
0.1 10
10.1 20
20.1 40
40.1
Time (s) Fig. 1. Sound intensity calibration procedure using a test tone of 45 s in duration (stimulus tone [9]) with fc sweeping from 0.1 to 12.5 kHz. (A and B) Signals in temporal (left column) and spectral (right column) domains recorded by an ER7c probe-tube microphone at the output of the transducer and the posterior end of the cavity, respectively. Attenuation due to the tubing effect was observed in the lower and higher fc for approximately 30-dB SPL at maximum. (C) Impulse response calculated from (B), which served as a reference to compute an inverse filter. (D and E) Inverse-filtered signal recorded at the output of the transducer and the posterior end of the cavity. Thoroughly flat intensity characteristics were achieved for the entire frequency range. (F) Signal recorded inside of a subject’s ear canal after inverse filtering and manual fine-tuning. (G) Excerpts of 100-ms window width extracted from (F). Thick gray lines delineate the envelope amplitude-modulated at the fm of 40 Hz. No critical phasic distortion was observed in either fc or fm, except for the initial rise period of the first 100 ms.
As a test sound, an inverse-filtered chord composed of 31 pure tones with frequencies corresponding to the 31 frequency bands of the equalizer was presented (Audition 1.0; Adobe Systems Inc., San Jose, CA, USA). The remaining deviations of a few dB SPL, presumably caused by the morphological difference of the individual ear canals and experimental settings, were fine-tuned manually using the 31-band equalizer (DEQ2496; Behringer GmbH, Kirchardt, Germany) at the spectral peak power of the 31 frequency components of the chord. Fig. 1F shows the inverse-filtered fine-tuned SAM sweep tone of 45 s (stimulus tone [9]) recorded inside the subject’s ear canal. Enlarged excerpts of the signal are shown in Fig. 1G, confirming visually that no critical phasic distortion was introduced to the fc and fm by the calibration procedure, except for the initial rise period of the first 100 ms. The overall sound intensity was adjusted to 70 ± 0.5-dB SPL. The transmission delay
of 7.83 ms was identified by the time point of zero crossing of the Hilbert-transformed positive and negative envelopes, and it was compensated for by a shift of the trigger signal. The ER7c microphone was detached and removed out of the magnetically shielded room after the calibration. 2.3. Subjects Eight right-handed individuals with a mean age of 38.5 (33–47, SD = 4.79) years, normal hearing (evaluated by the standard health examination, where pure-tone audiometric test was performed at 1 and 4 kHz), and no history of otological or neurological disorder participated in the experiment as subjects. The experiment was conducted in accordance with the Declaration of Helsinki, and it was approved by the Ethics Committee of The National Institute
A. Otsuka et al. / Clinical Neurophysiology 127 (2016) 790–802
793
Table 1 List of stimulus tones and their acoustic/presentation parameters.
The tones with shaded properties were categorized to test the corresponding parameter. For instance, the direction parameter was tested by comparing [1] and [2]. Stimulus tones [1]–[9] were presented repetitively for 40 trials. Stimulus tones [10]–[17] were presented once. Recording time was calculated by (duration + 5 s of prestimulus period) times the number of trial repetition.
of Advanced Industrial Science and Technology (AIST), Japan. Informed consent was obtained from each subject after the purpose and procedures of the experiment were fully explained. 2.4. Hearing threshold measurement Before the neuromagnetic measurement, subjects’ hearing threshold was recorded. We referred to an audiometric procedure standardized by The Japan Audiological Society (1990), with some modifications to be comparable with the ASSR signal. Nine test tones were extracted from the inverse-filtered 45-s SAM sweep tone (stimulus tone [9]) for a duration of 200 ms containing eight AM cycles with fc starting at 0.125, 0.25, 0.5, 1, 2, 4, 8, and 10 kHz. These test tones were presented binaurally from 70-dB SPL down to 10-dB SPL, then up to 70-dB SPL in 2-dB steps at an interstimulus interval (ISI) of 0.5 s, and it was repeated for three sessions. The presentation order of the test tones was randomized per subject. The subjects were required to respond by moving the index finger every time they perceived the tone. 2.5. MEG measurement The MEG signals were recorded in a magnetically shielded room using a whole-head SQUID system (Neuromag-122™; Elekta Neuromag Oy, Helsinki, Finland) containing 122 gradiometers that detect orthogonal planar gradients (DBz/Dx, DBz/Dy) at 61 distinct sensor elements distributed in a helmet-shaped array over the head. The signals were recorded at 0.03–100 Hz of an analog frequency band with a sampling frequency of 512 Hz with a 24-bit resolution. Each of the SAM sweep tones was presented repetitively 40 times at an ISI of 5 s. The non-sweeping SAM tones were presented once for a duration of 200 s. The total recording times for the first session ([1]–[2], [6]–[17]) and second session ([3]–[5]) were 67.3 and 53.3 min, respectively. The second session was performed on a different day. During the recording, the subjects were requested to watch a silent film with subtitles of their own choice, and to ignore the stimulus tones in order to reduce the effect of attentional modulations (Romani et al., 1982; Linden et al., 1985; Jerger et al., 1986; Plourde and Picton, 1990; Cohen et al., 1991; Ross et al., 2004; Elberling et al., 2007; Lazzouni et al., 2009). 2.6. Signal processing The recorded MEG signals were first spatiotemporally filtered using the temporal signal space separation (tSSS) method (Taulu et al., 2005; Taulu and Simola, 2006; Taulu and Hari, 2009) (Maxfilter 2.1; Elekta Neuromag Oy). The signals were transformed
into magnetostatic multipole moments, the amplitude coefficients of harmonic basis functions capable of separating the internal and external subspaces divided by the sensor array. After the spatial decomposition, external interference was suppressed by omitting the harmonic function components of the external origin. The signals in the internal and intermediate geometrical areas were decomposed by principle component analysis (PCA) at each of the temporal subspaces of the 5-s window. The artifacts and interferences nearby the sensor array were further projected out in the temporal domain when higher correlation coefficients (p > 0.98) were obtained between the PCA components. The signals were then band-passed using a fourth-order bidirectional Butterworth filter with cutoff frequencies centered at fm ±6 Hz width and segmented into 40 epochs consisting of the stimulus presentation period (either 5, 15, or 45 s) plus 5-s prestimulus and poststimulus periods. The epochs were averaged, and the baseline was corrected using the signals of the 5-s prestimulus period. Finally, the PCA components computed from the 5-s prestimulus period were spatially projected out as noise from the entire epoch. The signals for the SAM tones of 200 s with fixed fc were segmented per fm of 25 ms, resulted in computing averaged signals of 8000 epochs. The ECD was estimated for the entire stimulus presentation period at each time sample of 2.5-ms interval by a single moving dipole estimation method with a spherical volume conductor model using all 61 sensors that covered the left or right hemispheres (Curry 6.0.20; Compumedics Neuroscan, El Paso, TX, USA). The ECD parameters were evaluated in terms of goodness of fit (>80%) and 95% confidence volume ( 0.7, p < 0.05), in order to compare the neuronal and perceptual frequency characteristics by a constant fraction of 1:1. The normalized frequency characteristics were statistically compared using a three-way repeated measures ANOVA, with factors of measure (3: ECD moment for left and right hemispheres and loudness model), fc (8), and SPL (3), and a post hoc Scheffe’s test. Finally, the difference between the neuronal and perceptual frequency characteristics was quantified by subtraction (loudness model minus ECD moment and/or ECD moment minus loudness model), yielding the frequency characteristics index of the ECD moment when the loudness model was flat, or the frequency characteristic index of the loudness model when the strength of the ECD moment was flat. The difference was quantified by logarithmic curves.
2.7. Relationship between neuronal and perceptual frequency characteristics 3. Results The relationship between neuronal and perceptual frequency characteristics was examined in the following procedure. First, a regression analysis was performed per subject to compare the detected hearing threshold with the normal equal-loudness-level contours (Table B.1, Annex B; ISO226:2003(E)). If the hearing threshold of an individual was detected at 26.0-dB SPL at 1 kHz, his/her threshold curvature was compared with the normal equal-loudness-level contours of 26.0 phon. This analysis was performed to evaluate statistically whether the detected hearing threshold was well explained in terms of the standard model. Second, the loudness levels for 50–70-dB SPL were simulated by substituting the detected hearing threshold levels into formula 4.1 (1) of the normal equal-loudness-level contours (ISO226:2003(E)). The loudness level LN of a frequency f, which has an SPL Lp, was calculated by:
Lp ¼
10 lgAf dB LU 94 dB af
where
Af ¼ 4:47 103 ð100:025LN 1:15Þ 0:4 10
T f LU 9 10
af
af is the exponent for loudness perception and LU is a magnitude of the linear transfer function normalized at 1 kHz. af and LU were defined according to Table 1 of ISO226:2003(E), while Tf, the hearing threshold, was obtained in the present study from each subject. The SPL Lp was fixed to 50-, 60-, and 70-dB SPL, as was the case for the strength of the ECD moment of the ASSR component. A regression analysis was performed per subject to compare the estimated loudness models with the normal (unequal-) loudness-level contours for 50-, 60-, and 70-dB SPL (Table B.2, Annex B; ISO226:2003(E)) per subject. Third, the amplitude growth functions of the ECD moment of the ASSR component were calculated by fitting exponential regression curves to the grand-mean ECD moment (for the left and right hemispheres inclusively) and (1) the given sound intensity levels (50-, 60-, or 70-dB SPL) and (2) the loudness models, per fc (0.125, 0.25, 0.5, 1, 2, 4, 8, and 10 kHz) and per consecutive fc width (one octave (0.125–0.25, 0.25–0.5, . . ., 8–10 kHz), two-octaves (0.125–0.5, 0.25–1, . . ., 4–10 kHz), and up to 6.25 octaves (0.125– 10 kHz)). The analyses for the fc widths were performed to evaluate the interactive function of sound intensity level or loudness model with fc. The regression curves were evaluated by adjusted R2 and significance level. Further, the grand-mean ECD moment and loudness models were normalized by fc width, where a significantly higher
Fig. 2 shows the MEG signals at each stage of the analysis procedure. The temporal courses of the raw, tSSS filtered, and band-pass filtered signals of one sensor from the right temporal area of one subject in response to the stimulus tone [1] are shown in Fig. 2A–C, respectively. The fast-Fourier transformed spectra are shown in the right column, which indicate that the ASSR component was clearly elicited at the fm in the spectral domain. The 40-epochs averaged signal is shown in Fig. 2D, confirming the augmentation of the ASSR component in the temporal domain along with the fc of the stimulus tone sweeping from 0.1 to 12.5 kHz. The PCA-filtered averaged signal is shown in Fig. 2E. A marked reduction of the noise level in the prestimulus and poststimulus periods and no morphological distortions in the stimulus presentation period were confirmed by visual inspection. Details of the PCA-filtered signal are shown in Fig. 2F, displaying a 100-ms window width containing four sinusoidal cycles modulating periodically at a 25-ms interval (i.e., at the fm of 40 Hz). Fig. 3A shows the results of the source estimation for the stimulus tone [1] overlaid on the MR image of one subject. All the estimated ECDs for the right hemisphere are displayed with gray dots. The evaluation of the ECD parameters resulted in rejecting 81.07% (SD = 12.3) of the estimated ECD that failed to satisfy the acceptance criteria. The valid ECDs were mostly localized around the auditory cortices, as displayed with the white circles. The tonotopic organization (Romani et al., 1982; Pantev et al., 1988, 1989, 1994, 1995, 1996; Talavage et al., 2004; Wienbruch et al., 2006) was not clearly captured in our results. Fig. 3B shows the time course of the source activity computed by a template ECD source model. Each dot represents the strength of the ECD moment estimated at each time sample, representing the instantaneous total activation of the neuronal ensembles responding in the phase-locking manner to the fm with fc sweeping through the time (0–5 s) and frequency (0.1–12.5 kHz). The low-pass filtered curvature is overlaid with the thick line. The ECD moments showed an increase after the stimulus onset at 0.1 kHz (0 ms), which reached a maximum at 0.5 kHz (1668 ms) and then decreased linearly toward the higher frequencies on the logarithmic scale. Fig. 4 shows the overlay of the grand-mean (n = 8) ECD moment of the ASSR component at the eight fc obtained for the stimulus tones categorized to examine the effects of (A) sweep direction ([1] ascent and [2] descent), (B) duration or sweep velocity ([3] 5 s, [4] 15 s, and [5] 45 s), (C) reproducibility ([1] first and [3] second sessions), (D) SPL ([1] 70-dB SPL, [6] 60-dB SPL, and [7] 50-dB SPL), (E) fm ([1] 40 Hz, [8] 32 Hz, and [9] 51 Hz) and (F) sweeping fc versus fixed fc ([1] SAM sweep tone and [10]–[17] SAM fixed tones). The horizontal axis represents the fc of the stimulus tones
795
A. Otsuka et al. / Clinical Neurophysiology 127 (2016) 790–802
A
10
3
10
6
3
0 6 10
0 -10 3 10
B
Field Power (fT/cm/√Hz)
0
D
Field (fT/cm)
3
C
-10 3 10 0 -10 25
3
0 -25 25
E
0 6 10
0 5 10
0 5 10
0 -25
0
-5
0
5
10
0
Field (fT/cm)
F
20
40
60
Frequency (Hz)
Time (s) 25 0 -25 0
0.1 1
1.1 2
2.1 3
3.1 4
4.1
Time (s) Fig. 2. MEG signal analysis procedure. (A–E) Raw, tSSS-filtered, band-pass filtered, 40-trials averaged and PCA-filtered signals from one sensor in the right temporal area of one subject. A clearly elicited ASSR component was observed at the fm (i.e., 40 Hz) in response to the SAM sweep tone in temporal (left column) and spectral (right column) domains. (F) Excerpts of 100-ms window width extracted from (E) showing the periodic modulation at the fm of 25-ms interval.
sweeping exponentially in time and frequency. The vertical axis represents the sensitivity of the ASSR component as a function of fc. Irrespective of the stimulus acoustic or presentation parameters, the ECD moment showed a general increase from 0.125 kHz, which reached the maximum at 0.5 kHz and then decreased toward the higher frequencies linearly on the logarithmic scale. The mean attenuation rates were 2.5 (SD = 0.87)/2.3 (0.27) from the maximum at 0.5 kHz down to 4 kHz and 2.1 (0.36)/2.0 (0.25) from 4 kHz down to 10 kHz for the left/right hemispheres. The results of the three-way repeated measures ANOVA with factors of hemisphere (2), fc (8), and stimulus tone (2 or 3 depending on the stimulus parameter) showed a main effect of hemisphere in all the examined stimulus parameters (F(1, 7) = 5.39–12.85; p < 0.01–0.05), with the right hemisphere being larger than the left hemisphere, as indicated with asterisks between the left and right columns of Figs. 4A–F. The main effect of fc was observed in all the stimulus parameters (F(7, 49) = 14.26–34.11; p < 0.01), with 0.25, 0.5, and 1 kHz being larger than 0.125, 2, 4, 8, and 10 kHz (p < 0.01–0.02). An interaction effect of hemisphere by fc was also observed in all the stimulus parameters (F(7, 49) = 2.43–3.56; p < 0.01–0.05), with post hoc significances indicated above each figure. However, no significant main effect of stimulus tone was observed for the stimulus parameters of sweep direction (A) and sweep duration/velocity (B). The reproducibility (C) was confirmed by the non-significant difference between the first and second sessions. Further, sweeping fc and fixed fc (F) was not statistically different. On the other hand, the main effect of stimulus tone was significant for SPL (F(2, 14) = 5.15; p < 0.03), with 70-dB
SPL being larger than 50-dB SPL (D), and for fm (F(2, 14) = 5.82; p < 0.02), with 40 Hz being larger than 51 Hz (E). There was a significant interaction effect of fc by fm (F(14, 98) = 4.25; p < 0.01), with a post hoc significance at 0.5 kHz between 40 and 51 Hz (p < 0.01). Notably, although the ECD moment showed the overall decrease as a function of the SPL and fm, the relative frequency characteristics remained consistent. Fig. 5 illustrates the procedure to examine the relationship between the neuronal and perceptual frequency characteristics in 50–70-dB SPL range. Fig. 5A1 shows the grand-mean (n = 8) hearing thresholds overlaid on the normal equal-loudness-level contours (ISO226:2003(E)). The hearing threshold for 1 kHz was detected at 26.0-dB SPL on average (SD = 6.2-dB SPL). The best perceptual sensitivity was obtained at 2 kHz with 25.5-dB SPL (SD = 7.8-dB SPL) and worst at 0.125 kHz with 47.7-dB SPL (SD = 4.7-dB SPL). The regression analysis between the detected hearing threshold and the normal equal-loudness-level contours showed significant regression coefficients (adjusted R2 = 0.77– 0.97, p = 0.01) in all subjects except for one, who was thus excluded from the subsequent analyses. Fig. 5A2 shows the grand-mean (n = 7) loudness levels for 50-, 60-, and 70-dB SPL, simulated by substituting the detected hearing thresholds into formula 4.1 (1) of the normal equal-loudness-level contours (ISO226:2003(E)). The individual loudness levels at eight fc were estimated in the range of 14.32–51.0 (SD = 9.88) phon, 29.09–61.36 (SD = 9.63) phon, and 48.43–71.42 (SD = 7.59) phon, respectively, when the frequency characteristics of the sound intensity level were flat at 50-, 60-, and 70-dB SPL. To note, the
796
A. Otsuka et al. / Clinical Neurophysiology 127 (2016) 790–802
Fig. 3. ECD source estimation of the ASSR component. (A) All (gray dots) and valid (white circles) results of the ECD estimation overlaid on the MR image of one subject. The mean of the valid ECD source location is indicated by a cross, which served as a template ECD source model to calculate the temporal course of the source activity per hemisphere per subject. L = left, R = right, S = superior, I = inferior, A = anterior, P = posterior. (B) Time course of the source activity (gray dot) and low-pass filtered curvature (black line). The strength of the ECD moment showed an increase and a decrease along with the fc sweeping through the time (0–5 s) and frequency (0.1–12.5 kHz).
normal equal-loudness-level contours shown in Fig. 5A1 and A2 delineate the same data, but with different concepts, either to show the SPL required to maintain the equal loudness level or the loudness level in the unit of phon perceived when sounds were delivered at an equal SPL. The modeled and normal (unequal-) loudness-level contours at 50–70-dB SPL showed significant regression coefficients (adjusted R2 = 0.87–0.99, p = 0.01) in all the examined subjects (n = 7 out of 8), supporting the idea that the estimated loudness models were well explained in terms of the standard model. Fig. 5B1 and B2 show the grand-mean ECD moment on the y-axis (=y-axis of Fig. 4D) and either the given sound intensity level (50-, 60-, or 70-dB SPL) or the estimated loudness model (=y-axis of Fig. 5A2) on the x-axis, with exponential regression curves fitted to represent the amplitude growth functions of the ASSR component by intensity (IGF) or modeled loudness level (loudness growth function (LGF)), per fc. Significantly higher regression coefficients were obtained for 0.125–8 kHz for IGF (adjusted R2 = 0.72, 0.79, 0.92, 0.86, 0.90, 0.80, 0.70; p = 0.01–0.04) and for LGF (adjusted R2 = 0.70, 0.83, 0.93, 0.86, 0.90, 0.81, 0.71; p = 0.01–0.04), indicating that the ECD moment of the ASSR component in 50–70-dB SPL range could be explained by the exponential model when IGF or LGF was examined per fc. Although slightly better fitting results were obtained for LGF than for IGF (mean adjusted R2 = 0.82/0.81, respectively), the frequency characteristics of the amplitude growth functions of the ASSR component were consistent in both measures, with 0.5 kHz exhibiting a steeper slope than for the lower and higher fc. Fig. 5B3 shows the grand-mean ECD moment on the y-axis (=y-axis of Fig. 4D) and the modeled loudness level (=y-axis of Fig. 5A2) on the x-axis, but with exponential regression curves fitted to the two-octaves fc width, where the significantly higher regression coefficients were obtained (adjusted R2 = 0.82; p < 0.01 for 0.125–0.5 kHz and R2 = 0.70; p < 0.01 for 4–10 kHz). The one-octave fc width of 0.125–0.25, 0.25–0.5, and 4–8 kHz also showed significantly higher regression coefficients (adjusted R2 = 0.73, 0.89, 0.81, respectively; p < 0.01). No significant regression coefficients were obtained for fc width of more than two octaves. Interestingly, the ECD moment between 0.5 and
4 kHz decreased, while the loudness models were mostly equal at around 50, 60, and 70 phon, respectively, indicating that the strength of the ECD moment of the ASSR component could not be explained by the single exponential function of loudness, but interactively with frequency. Fig. 5C1 and C2 show two possible interpretations for how the ECD moment and loudness model could be overlaid when they were normalized by fc width, where significantly higher regression coefficients were obtained between the neuronal and perceptual frequency characteristics. In the case where the ECD moment and loudness model were adjusted by 0.125–0.5 kHz (R2 = 0.82; p < 0.01), a large difference was induced in the higher frequencies because of the steep decrease of the ECD moment after 0.5 kHz, in contrast to the equal loudness model sustained up to 4 kHz. A significant interaction effect was observed between the ECD moment and loudness model by fc (F(14, 280) = 16.12; p < 0.01), with the ECD moment being significantly smaller than the loudness model at 2–10 kHz (p < 0.01–0.05) in both hemispheres. Conversely, in the case where the ECD moment and loudness model were adjusted by 4–10 kHz (R2 = 0.70; p < 0.01), a large difference was induced in the lower frequencies because of the precipitous growth of the ECD moment at 0.5 kHz, in contrast to the loudness model plateauing between 0.5 and 4 kHz. A significant interaction effect was observed between the ECD moment and loudness model by fc (F(14, 280) = 11.81; p < 0.01), with the ECD moment being significantly larger than the loudness model at 0.25–1 kHz (p < 0.01) in both hemispheres. Fig. 5C3 shows the possible differences between the neuronal and perceptual frequency characteristics obtained by subtractions. The ECD moment was augmented in the lower frequencies peaking at 0.5 kHz when the frequency characteristics of the loudness model were flat. By contrast, the loudness model was augmented in the higher frequencies peaking at 4 kHz when the frequency characteristics of the ECD moment were flat. The difference as a function of fc was linear between 0.5 and 4 kHz on logarithmic scale, and it was quantified by the regression curves as y = 3.68log (fc) + 2.18 (R2 = 0.98) (i.e., ECD moment when the loudness model was flat) and y = 1.5log (fc) 0.44 (R2 = 0.99) (i.e., loudness model when ECD moment was flat), respectively.
797
A. Otsuka et al. / Clinical Neurophysiology 127 (2016) 790–802
Left Hemisphere
A
Right Hemisphere
Left Hemisphere
D
Sweep Direction *** ~ *
***
***
***
0.8
***
***
SPL ***
***~**
Right Hemisphere
***
***
*** ***
***
*
***
***
[1] 70 dB [6] 60 dB [7] 50 dB
*
[1] ascent [2] descent
0
E
Duration ***
**~* ***~**
0.8
***
*
fm ***
***
*** **~*
***
*
***
[1] 40 Hz [8] 32 Hz [9] 51 Hz
***
[3] 5 sec [4] 15 sec [5] 45 sec ***
Normalized ECD Moment (a.u.)
B
0
C
F
Session ***
***
*** ***
0.8
*
***
Sweep / Fix ***
***
***
***
*** ***
*
***
***
[1] 1st [3] 2nd
***
[1] SAM chirp [10] SAM Fixed
0 102
103
104 102
103
104
102
Frequency (Hz)
103
104 102
103
104 *** ** *
p
