Modelling neural circuits involved in detecting signals from background noise in the ventral cochlear nucleus

One of the computational challenges for the brain is to interpret the sound waveform reaching the ear in terms of the events that produced it. This is a formidable problem, because the sound waveform is typically made up of a mixture of sounds originating from a variety of different sources. The ability of the auditory system to segregate sounds into separate 'auditory objects' such as trying to listen to a single musical instrument in an orchestra or a person speaking in a noisy café is often referred to as a 'scene analysis' problem. Indeed, in the natural environment listeners rarely, if ever, have to detect sounds in quiet and the ability to distinguish signals from background noise has important survival consequences e.g. the detection of a predator. Interestingly background noise is often not random but slowly fluctuates in time and it appears that the auditory system has evolved to make use of these slow fluctuations. For example, we are much better at detecting a signal when it is accompanied by a fluctuating noise than a non-fluctuating noise. However, it is equally important that these fluctuations are coherent in time across different frequency regions of the cochlea. When discussing this 'scene analysis' problem Bregman (1990) distinguishes between primitive grouping, which involves the (probably innate) use of general properties of sound to determine the source of acoustic elements, and schema-based grouping, which is generally learned and therefore dependent on the listener's specific experience. Both processes influence the grouping of complex stimuli like speech. We hypothesize that the primitive grouping is a low-level phenomenon, taking place in the early stages of the auditory pathway. It occurs before perception takes place and so we regard this as perceptual pre-processing. Until recently we have had very little idea about how the brain's circuits may contribute to auditory scene analysis, however, we have recently demonstrated that simple circuits in the cochlear nucleus, the earliest stage in the central auditory pathway, are involved in the enhanced detection of signals in noise (see Pressnitzer et al., 2001; Meddis et al., 2002; Verhey et al., 2003; Neuert et al., 2004). Our work, so far, has identified several potential new and important processing systems in the ventral cochlear nucleus. In this project we will realize a more complete understanding of this part of the brain by producing computational models that replicate the neural responses of the circuits involved. Our proposal offers the close interaction between researchers with established track records in the physiology of the auditory brainstem (Winter), and auditory modeling (O'Mard). The modeling will both complement and constrain future physiological studies, providing a theoretical framework within which to interpret the neural data. Existing physiological data (recorded by Winter) will inform the development of a neural computational model of perceptual pre-processing in the mammalian ventral cochlear nucleus. This kind of integrative approach to investigating auditory processing offers the best opportunity for significant advances in our understanding of the segregation of signals from noise. It will permit the testing of a large array of stimuli that would not be feasible using current neurophysiological recording techniques and provide a realistic output from the circuit which can be used as input to higher levels of processing.

We have identified a neural circuit in the mammalian ventral cochlear nucleus (VCN) that is ideally positioned to play a role in segregating signals from noise. This circuit is comprised of both ascending and descending inputs as well as intrinsic neural interconnections. We will produce computational models of the individual elements of this circuit and then model their interaction. The output of the circuit will be validated using the known responses of the circuit elements to complex sounds such as those used in physiological recordings in the applicant's laboratory. The computer modelling will take place utilising the developmental system for auditory modelling (DSAM) produced by the Centre for the Neural Basis of Hearing at Essex (by O'Mard). DSAM provides a flexible modelling environment of the auditory periphery and we will continue this approach to modelling by producing a model of a circuit within the VCN. The neural circuit in the mammalian VCN is well positioned to play a role in the detection of signals in noise under a variety of conditions. The output of the circuit is through multipolar cells that are known to project to the inferior colliculus. The multipolar cells receive excitatory input from auditory nerve fibres but also from the collaterals of the olivocochlear bundle. We propose that the function of these collaterals is to enhance the detection of signals in noise. Multipolar cells also receive input from wideband inhibitors which act to reduce the response to modulated masking noise and play a role in co-modulation masking release. Although classically viewed as a monaural nucleus it is clear that the circuit in the ventral cochlear nucleus is in receipt from input from the contralateral cochlear nucleus. It is possible that this input comes from wideband inhibitor neurons and could provide a binaural enhancement in signal detectability at a level as early as the VCN.