Optimising a computational model of human auditory cortex with an evolutionary algorithm.

We demonstrate how the structure of auditory cortex can be investigated by combining computational modelling with advanced optimisation methods. We optimise a well-established auditory cortex model by means of an evolutionary algorithm. The model describes auditory cortex in terms of multiple core, belt, and parabelt fields. The optimisation process finds the optimum connections between individual fields of auditory cortex so that the model is able to reproduce experimental magnetoencephalographic (MEG) data. In the current study, this data comprised the auditory event-related fields (ERFs) recorded from a human subject in an MEG experiment where the stimulus-onset interval between consecutive tones was varied. The quality of the match between synthesised and experimental waveforms was 98%. The results suggest that neural activity caused by feedback connections plays a particularly important role in shaping ERF morphology. Further, ERFs reflect activity of the entire auditory cortex, and response adaptation due to stimulus repetition emerges from a complete reorganisation of AC dynamics rather than a reduction of activity in discrete sources. Our findings constitute the first stage in establishing a new non-invasive method for uncovering the organisation of the human auditory cortex.

Stimulation-history effects on the M100 revealed by its differential dependence on the stimulus onset interval.

The amplitudes of the most prominent component of auditory evoked magnetic fields and electrical potentials, the M100 and N100, recorded from the human scalp depend on the duration of the stimulus onset interval (SOI). Here, we show, using magnetoencephalography, that the SOI dependence of the M100 amplitude strongly depends upon whether stimuli with different SOIs are presented in a conventional block design or in a random manner. This differential dependence reveals that the M100 is affected not only by the stimulus evoking it and by its predecessor, but by a longer-term history of stimulation. We provide a parsimonious model that accounts for our findings with both designs in a quantitative manner. It assumes a transient, temporally asymmetric reduction in the excitability of a fraction of potentially excitable neurons. A rather stereotyped response function may therefore underlie the stimulation-history effects in the human auditory cortex.

The M100 component of evoked magnetic fields differs by scaling factors: implications for signal averaging.

MEG and EEG studies of event-related responses often involve comparisons of grand averages, requiring homogeneity of the variances. Here, we examine the possibility, implied by the nature of neural sources and the measuring principles involved, that the M100 component of auditory-evoked magnetic fields of different subjects, hemispheres, to different stimuli, and at different sensors differs by scaling factors. Such a multiplicative model predicts a linear increase in the standard deviation with the mean, and thus would have important implications for averaging and comparing such data. Our analyses, at the sensor and the source level, clearly show that the multiplicative model applies. We therefore propose geometric, rather than arithmetic, averaging of the M100 component across subjects and suggest a novel and superior normalization procedure. Our results question the justification of the common practice of subtracting arithmetic grand averages.

Effects of the task of categorizing FM direction on auditory evoked magnetic fields in the human auditory cortex.

We examined effects of the task of categorizing linear frequency-modulated (FM) sweeps into rising and falling on auditory evoked magnetic fields (AEFs) from the human auditory cortex, recorded by means of whole-head magnetoencephalography. AEFs in this task condition were compared with those in a passive condition where subjects had been asked to just passively listen to the same stimulus material. We found that the M100-peak latency was significantly shorter for the task condition than for the passive condition in the left but not in the right hemisphere. Furthermore, the M100-peak latency was significantly shorter in the right than in the left hemisphere for the passive and the task conditions. In contrast, the M100-peak amplitude did not differ significantly between conditions, nor between hemispheres. We also analyzed the activation strength derived from the integral of the absolute magnetic field over constant time windows between stimulus onset and 260 ms. We isolated an early, narrow time range between about 60 ms and 80 ms that showed larger values in the task condition, most prominently in the right hemisphere. These results add to other imaging and lesion studies which suggest a specific role of the right auditory cortex in identifying FM sweep direction and thus in categorizing FM sweeps into rising and falling.

A multilevel and cross-modal approach towards neuronal mechanisms of auditory streaming.

We report first results of a multilevel, cross-modal study on the neuronal mechanisms underlying auditory sequential streaming, with the focus on the impact of visual sequences on perceptually ambiguous tone sequences which can either be perceived as two separate streams or one alternating stream. We combined two psychophysical experiments performed on humans and monkeys with two human brain imaging experiments which allow to obtain complementary information on brain activation with high spatial (fMRI) and high temporal (MEG) resolution. The same acoustic paradigm based on the pairing of tone sequences with visual stimuli was used in all human studies and, in an adapted version, in the psychophysical study on monkeys. Our multilevel approach provides experimental evidence that the pairing of auditory and visual stimuli can reliably introduce a bias towards either an integrated or a segregated perception of ambiguous sequences. Thus, comparable to an explicit instruction, this approach can be used to control the subject's perceptual organization of an ambiguous sound sequence without the need for the subject to directly report it. This finding is of particular importance for animal studies because it allows to compare electrophysiological responses of auditory cortex neurons to the same acoustic stimulus sequence eliciting either a segregated or integrated percept.