Analysis of magnetic source localization of P300 using the multiple signal classification algorithm


Dr Norihito Yamada, MD, Department of Neuropsychiatry, Okayama University Medical School, 2–5−1 Shikata-cho, Okayama 700–8558, Japan. Email:


Abstract  The authors studied the localization of P300 magnetic sources using the multiple signal classification (MUSIC) algorithm. Six healthy subjects (aged 24–34 years old) were investigated with 148-channel whole-head type magnetencephalography using an auditory oddball paradigm in passive mode. The authors also compared six stimulus combinations in order to find the optimal stimulus parameters for P300 magnetic field (P300m) in passive mode. Bilateral MUSIC peaks were located on the mesial temporal, superior temporal and parietal lobes. Interestingly, all MUSIC peaks in these regions emerged earlier in the right hemisphere than in the left hemisphere, suggesting that the right hemisphere has predominance over the left in the processing activity associated with P300m. There were no significant differences among the six stimulus combinations in evoking those P300m sources. The results of the present study suggest that the MUSIC algorithm could be a useful tool for analysis of the time-course of P300m.


The sources of a late positive event-related potential (ERP) component (P300) are controversial, although a few researchers have sought it using intracranial electroencephalography (EEG), topographical distributions of the scalp-recorded P300, functional magnetic resonance imaging (fMRI), and magnetencephalography (MEG).1–9 Multiple sources are thought to be associated with P300. The conventional auditory version of P300 uses two different tones with the target ‘oddball’ stimulus presented less frequently than the non-target or standard stimulus. The subject is required to distinguish between the two tones by responding to the target stimulus (e.g. mentally counting, pressing a button etc.). However, when subjects are either uncooperative or unconscious, P300 must be measured in passive mode, in which subjects do not count or respond to the target stimuli. The authors’ final research aim is to study magnetic sources of P300 in patients with head trauma to evaluate how much cognitive function they potentially have. Although P300 is known to be relatively unaffected by changes in stimulus parameters, it is still unknown if the conventional stimulus paradigm is truly the optimal method for those who are unconscious. The authors tested six different pairings of target and non-target tones in order to find out whether the conventional stimulus paradigm is optimal for evoking P300 in passive mode. The authors made the assumption that manipulation of the task influenced the activated regions for the P300 component.

Most previous studies have used an equivalent current dipole (ECD) as a source for analysis with MEG. However, considering that multiple electrical sources for over several hundreds of milliseconds are involved in P300, analysis of P300 sources requires a method that can detect multiple sources simultaneously. The multiple signal classification (MUSIC) algorithm is a method of localizing multiple dipole sources.10–15 Ninomiya et al. suggested the usefulness of MUSIC for analysis of pain-related somatosensory-evoked magnetic fields (SEF), showing the time course of multiple sources.16 In short, the MUSIC algorithm is a multidipole directed search in which a single dipole is scanned through a grid confined to a 3-D head, the forward model is computed from the EEG or MEG data at each grid point, and a relatively dense grid of dipolar source locations is created. It enables the user to time-dependent changes of multiple magnetic/electrical sources with relatively little time or effort for the analysis. Also, it does not require the number of dipoles to be known in advance. Therefore, the authors analyzed the magnetic sources of P300 in passive mode in normal volunteers with the MUSIC algorithm.



Six healthy normal subjects with no psychiatric or neurological disorders were recruited for this study (ages, 24–34 years; four males and two females). All subjects were right-handed. Informed consent was obtained from all subjects before the study.

Experimental paradigm

The subjects lay on a bed in a quiet, magnetically shielded room and were requested to keep their eyes closed and remain still during the measurement. An auditory oddball paradigm in passive mode was employed as a stimulus sequence. The stimulus tone was delivered via an airtube. All stimuli were presented at 85 dB. The stimulus duration was 50 ms (rise time 10 ms, plateau time 30 ms, fall time 10 ms) with a 2000 ± 500 ms interstimulus interval (ISI). The authors used six different kinds of oddball paradigm in which tone frequency for the target and non-target was varied (Table 1). The occurrence rate was 20% for the target. Each session included 160 stimuli. A total of 32 trials with the target stimuli were averaged to obtain P300. The subjects received two sessions for each stimulus paradigm, consequently undergoing 12 sessions in total. The order of stimulus paradigm was randomized.

Table 1.  Six variations of oddball paradigm used in the experiment
I1 kHz 80%, 2 kHz 20% (standard oddball paradigm)
II1 kHz 20%, 2 kHz 80%
III1 kHz 80%, 1.5 kHz 20%
IV1 kHz 20%, 1.5 kHz 80%
V0.5 kHz 80%, 1 kHz 20%
VI0.5 kHz 20%, 1 kHz 80%

Data acquisition

Electroencephalography and magnetencephalography recording

The head of the subject was positioned in a 148-channel whole-head-type biomagnetometer (Magnes, Biomagnetic Technologies Inc., San Diego, CA, USA). A simultaneous EEG was recorded from electrodes placed on the scalp at Fz, Cz, and Pz according to the International 10–20 System. Ag/AgCl electrodes were referenced to the linked ears. They were fixed to the scalp with collodion, and the impedance of all electrodes was kept at <5 kΩ. The data was sent to an NEC digital EEG recorder (Neuropack 8, Nihon Kohden). Both the MEG and EEG were digitized at a sampling frequency of 2034.5 Hz and filtered with a band-pass of 0.1–400 Hz. Digitized data were stored and analyzed off-line, filtered again through a 1–50 Hz band-pass filter to the MEG data. A GE Signa 1.5 T system was used for magnetic resonance imaging (MRI). The 3-D spoiled gradient pulse sequence (SPGR) images were used for overlays with MUSIC localization and ECD sources detected by MEG. The nasion was identified on MRI with the aid of high-contrast cod liver oil capsules.

The responses to the onset of the target and non-target stimuli were averaged on-line separately. The data acquisition time was between 150 ms before and 550 ms after each stimulus onset. The analysis window for the on-line averaging was from 50 ms before to 550 ms after each stimulus onset. In order to confirm the reproducibility of responses, two sessions for each stimulus paradigm were carried out for each subject. The P300 component was determined as a positive deflection for the EEG response to the target stimuli evoked at Cz or Pz at the latency between 250 and 550 ms, and magnetic deflections for the target were detected during the same time interval in the MEG channels.

Data analysis

The authors applied two different methods to analyze the magnetic source of the brain activities obtained during the tasks: the ECD source analysis and the MUSIC algorithm. First, source analysis of the MUSIC algorithm was performed using Advanced Source Analysis (ASA; Advanced Neuro Technology, Enschede, The Netherlands) on a PC/AT compatible personal computer. The principle of the MUSIC algorithm has been explained in detail in previous papers.10–14 The authors performed the MUSIC analysis every 30 ms ranging 0 ms to 450 ms after stimulation, that is, they made a sequence of 15 source analyses for each measurement. Second, the authors confirmed every source scanned with the MUSIC algorithm using the ECD model.

Statistical analysis

Event-related potential (P300)

One-way anova was performed to study the differences among the six paradigms on the latency and amplitude of the P300.

Event-related magnetic field

A sign test was performed to clarify whether there was laterality in the time-course of MUSIC peaks every 30 ms ranging between 0 ms and 450 ms after stimulation. A sign test was also performed to examine differences in the number of current sources obtained with among the six paradigms.


Event-related potential (P300)

P300 was clearly elicited in the simultaneously recorded EEG in all subjects. Figure 1 shows an example of P300 as well as simultaneously recorded MEG waveforms. The amplitude of these components was greater in recordings based on oddball stimuli than they are in recordings based on frequent stimuli. Overall, raw MEG waveforms also indicated more activity between 200 and 400 ms responding to the oddball stimuli.

Figure 1.

P300 waveform simultaneously recorded in electroencephalography and magnetencephalography (Subject1). P300 was clearly elicited in response to the target stimulus in electroencephalography.

There were no significant changes in either the latency or amplitude of P300 in the EEG recordings by the variation of oddball paradigm (Table 2).

Table 2.  Latency and amplitude of the passive P300 by six variations of oddball paradigm (mean ± SD)
 Latencies (P300) msAmplitude (N100-P300) µV
  1. There were no significant changes in the P300 latency and amplitude by the six variation of oddball paradigm (one-way anova).

I320.6 ± 15.78.9 ± 3.5
II303.8 ± 20.610.4 ± 4.67
III311.8 ± 19.96.65 ± 2.58
IV307.7 ± 24.9 7.5 ± 2.56
V328.5 ± 17.210.4 ± 4.33
VI330.8 ± 13.2 7.6 ± 2.17

Event-related magnetic field

Multiple signal classification and equivalent current dipole localization of P300m

Superimposition of contour maps by the MUSIC algorithm onto MRI in the axial, coronal, and sagittal planes showed that N100m components were located in the bilateral superior temporal lobes and that P300m components were located in the bilateral mesial temporal lobes (MTL), mainly the hippocampus, superior temporal lobes (STL) and parietal lobes (PL; Fig. 2). There were signals in the frontal lobes as for the P300m components, but the authors excluded those signals from the analysis because it was difficult for the MUSIC algorithm to discriminate the real signals from artifacts when signals near the frontal midline were detected bilaterally. These midline peaks could be due to a shortcoming of MUSIC algorithm itself, according to which a midline peak can indicate a wave in any brain area by inverse operation. The sources in the MTL, STL and PL obtained with MUSIC analysis were equivalent to those estimated with ECD (post-hoc estimation, Fig. 2).

Figure 2.

Superimposition of the equivalent current dipole and multiple signal classification localizations on magnetic resonance imaging. N100m components were located in the bilateral superior temporal lobes (upper). P300m components were located in the bilateral superior temporal, mesial temporal, and parietal lobes (middle and lower). The mesial temporal and parietal lobes were activated simultaneously in multiple signal classification sagittal slices (middle). The arrow points to multiple peaks discriminated at the same time. Note that peaks were observed in the right hemisphere in 300–330 ms (middle) and peaks appeared to move on the left hemisphere in 390–420 ms (bottom). N100m peaks, however, were observed in the both hemisphere simultaneously (upper).

Time-course of the peaks obtained from the multiple signal classification algorithm

1 kHz 80%, 2 kHz 20% (standard oddball paradigm)  The time-course of the peaks obtained from the MUSIC algorithm is presented in Fig. 3. The figure shows the results of MUSIC analysis by every 30 ms period between 0 and 450 ms after the stimulation among the 12 sessions (two sessions of six subjects). The places, where there were MUSIC peaks, were plotted with each session. MUSIC peaks of N100m were observed 60–120 ms after the stimulation in the bilateral temporal cortices. No laterality was found in the appearance of N100m. MUSIC peaks of P300m were estimated in the bilateral MTL, STL and PL, as mentioned above. The authors identified these MUSIC peaks since they were constantly observed and their appearance time coincided with the rise time of either N100 or P300 in EEG. MUSIC peaks in the MTL were observed 240–360 ms after the stimulation in the right hemisphere, and those in the left hemisphere were observed 360–420 ms after the stimulation. MUSIC peaks in the PL were observed 240–360 ms after the stimulation in the right hemisphere, and those in the left hemisphere were observed 360–420 ms after stimulation. In the STL, MUSIC peaks were found 300–360 ms after the stimulation in the right hemisphere, and those in the left hemisphere were observed 360–420 ms after the stimulation. Therefore, the MUSIC peaks of the right hemisphere preceded those of the left hemisphere in the MTL, PL and STL. There was no specific tendency to be earlier in any one of the three areas.

Figure 3.

Time-course of the sources evoked by the standard oddball paradigm (1 kHz 80%, 2 kHz 20%). Analysis using the multiple signal classification (MUSIC) algorithm was performed every 30 ms from 0 s to 450 ms after the stimulation. MUSIC peaks obtained in each session are expressed with filled grids in the figure. The N100m components were estimated in the bilateral temporal lobes. The P300m components were estimated on the mesial temporal lobes, superior temporal lobes, and parietal lobes. The MUSIC peaks emerged earlier in the right hemisphere than in the left hemisphere. * P < 0.05, a sign test was performed to compare the occurrence rate of the MUSIC peaks between the two hemispheres.

Comparison among the six paradigms  The authors compared the time-course of MUSIC peaks of the stimulus paradigm II–VI with that of the paradigm I, which is the most conventional stimulus paradigm. There were no significant differences in the occurrence of MUSIC peaks among the six paradigms. Also, no significant differences were found in the peaks in the MTL, PL and STL (data not shown).


The scalp-recorded P300 is thought to be an overlap of two different subcomponents, the frontocentral P300 and the temporal-parietal P300.17 Halgren et al. found that P300 represents a widespread corticolimbic modulation of the systems responsible for orienting attention toward a rare auditory stimuli (P3a) and the subsequent encoding of that stimulus into a cognitive event (P3b) with an intracranial EEG.5 The P3a was generated in the inferior parietal, cingulate, and dorsolateral prefrontal cortex (system for the orientation of activity), and the P3b was generated in the hippocampus, superior temporal cortex, and parietal cortex (event-encoding system). Early MEG studies using ECD analysis suggested that the magnetic sources were in the hippocampus1 and thalamus.2 Tesche et al. suggested that generators of evoked magnetic responses to oddball stimuli were in the hippocampus based on whole-head MEG using signal-space projection, and that the prominent responses in the hippocampus emerged 200–500 ms after attended oddballs.4 Nishitani et al. detected three generator sources for the auditory P300 on each hemisphere: MTL, STL and inferior parietal areas as a result of analysis using a time-varying multidipole model using the 122 channel whole-head neuromagnetometer.6–8 The present study demonstrated that analysis using the MUSIC algorithm detected three major cerebral current sources: bilateral MTL, STL and PL. The result demonstrated that MUSIC algorithm is able to detect the magnetic sources associated with P3b.

The authors did not find sources related to P3a, probably because the study was done in passive mode. It is known that passive oddball task does not activate the process of response selection, planning or working memory. It is, therefore, likely that prefrontal cortex was not activated by the task in the present study. Another possibility, as stated above, is that one disadvantage of the MUSIC algorithm is that is could have masked sources located near the midline of the brain. It is possible that sources near the midline such as in the cingulate and prefrontal cortices may have been abandoned since it is difficult to distinguish between sources and artifact. Sources in thalamus18,19 might have also been contaminated by artifact for the same reason.

The single dipole model has widely been used as a source model in the analysis of electrophysiological data. Although the model has high spatial resolution, it is not suitable for analyzing complex processes such as higher brain functions because of its limited accuracy in measuring multiple electrical current sources.14 Mosher et al. introduced the use of the MUSIC algorithm to the MEG inverse model11 as a method of localizing multiple dipole sources.13 The algorithm can be used to locate multiple asynchronus dipolar sources from either EEG or MEG data. It does not require the number of dipoles to be known in advance, which is an advantage in the screening of multiple sources involved in higher brain activity.

Another advantage of the MUSIC algorithm is its ease in revealing the temporal profile of the multiple sources associated with P300m. Interestingly, the MUSIC peaks at the MTL, STL, and PL occurred earlier on the right hemisphere than the left hemisphere. Few studies have reported the time course and laterality of the sources in the oddball paradigm. Downar et al. have demonstrated that multimodally responsive areas comprised a right lateralized network that includes the temporoparietal, inferior frontal gyrus, and insula by fMRI using a modified version of oddball paradigms. This right-lateralized network consists of areas that are thought to underlie P300.20 The time course analysis of P300 by low resolution electromagnetic tomography (LORETA) revealed different time course patterns in the left and right hemisphere with earlier activations in frontal and parietal regions in the right hemisphere using an auditory oddball paradigm.21

The authors’ results clearly show that, in every subject examined, the sources were always first generated in the right hemisphere and moved to the left hemisphere several tens of milliseconds after generation in the right hemisphere. The right hemisphere may have predominance over the left in the processing activity associated with P300m. Another interesting result was that the three brain areas (MTL, STL and PL) appeared to be activated in the same period of time in each hemisphere. Those three areas were seemingly activated at once in the right hemisphere, and then the same areas in the left hemisphere were activated with a delay period of 60–120 ms. Although the reason of this pattern is still unknown, analysis with higher time resolution will give us more information about the issue. Also, more studies of various modes of cognition are needed to clarify whether this laterality is definite and universal. In such studies, the MUSIC algorithm will again be useful for the analysis of the time-dependent manner in magnetic sources. There are a few different algorithms that estimate 3-D multiple sources from the data obtained with EEG or MEG. Recently, LORETA and Minimum Norm Estimation (MNE) have often been used for analyzing the sources of event-related potentials in EEG/MEG. These techniques, including MUSIC, sometimes give different results probably because of their different modeling criteria.22 Since there have been few studies analyzing P300 magnetic sources using either LORETA or MNE, it will be necessary to study the source analysis with these algorithms to evaluate the results in the present study.

Changes in the amplitude and latency of the P300 have been modulated by a variety of experimental manipulations, including stimulus probability, task difficulty, and task demands.23 The probability of occurrence of the target stimuli has an effect on the amplitude of the P300.24,25 Horovitz et al. showed that when the probability of occurrence of the target auditory stimuli decreases, the amplitude of the P300 increases.9 According to Sugg and Polish, varying stimulus frequencies also affects P300 measures in the absence of significant task performance.26 Therefore, 1 kHz 80%, 2 kHz 20% has been widely used as a standard oddball paradigm. However, there has been little evidence that indicated what the optimal frequency of tones for auditory stimuli is. Sugg and Polish reported that the amplitude of the P300 which was obtained for combinations of high frequency stimuli was somewhat larger, although the difference was not significant.26 They used combinations of 250/500 and 1000/2000 Hz (standard/target frequency). The authors, therefore, compared six different combinations of tones keeping the probability of occurrence constant. The amplitude of P300 did not differ among the stimuli in the present study, and there were no significant differences of MUSIC peaks among the stimuli. The variation in the pitch of stimulus tones did not influence either the parameters of P300 in EEG or magnetic sources in MEG.

In conclusion, the authors localized the magnetic sources of P300 in the MTL, STL, and PL using the MUSIC algorithm. The MUSIC peaks of the right hemisphere preceded those of the left hemisphere in these areas. The authors suggest that the MUSIC algorithm could be a useful tool as a complementary method for analysis when multiple dipoles exist simultaneously.


This study was supported in part by a grant from the Zikei Institute of Psychiatry, Okayama, Japan.