Large-scale electrophysiology and deep learning reveal distorted neural signal dynamics after hearing loss.

Listeners with hearing loss often struggle to understand speech in noise, even with a hearing aid. To better understand the auditory processing deficits that underlie this problem, we made large-scale brain recordings from gerbils, a common animal model for human hearing, while presenting a large database of speech and noise sounds. We first used manifold learning to identify the neural subspace in which speech is encoded and found that it is low-dimensional and that the dynamics within it are profoundly distorted by hearing loss. We then trained a deep neural network (DNN) to replicate the neural coding of speech with and without hearing loss and analyzed the underlying network dynamics. We found that hearing loss primarily impacts spectral processing, creating nonlinear distortions in cross-frequency interactions that result in a hypersensitivity to background noise that persists even after amplification with a hearing aid. Our results identify a new focus for efforts to design improved hearing aids and demonstrate the power of DNNs as a tool for the study of central brain structures.

Harnessing the Power of Artificial Intelligence in Otolaryngology and the Communication Sciences.

Use of artificial intelligence (AI) is a burgeoning field in otolaryngology and the communication sciences. A virtual symposium on the topic was convened from Duke University on October 26, 2020, and was attended by more than 170 participants worldwide. This review presents summaries of all but one of the talks presented during the symposium; recordings of all the talks, along with the discussions for the talks, are available at https://www.youtube.com/watch?v=ktfewrXvEFg and https://www.youtube.com/watch?v=-gQ5qX2v3rg . Each of the summaries is about 2500 words in length and each summary includes two figures. This level of detail far exceeds the brief summaries presented in traditional reviews and thus provides a more-informed glimpse into the power and diversity of current AI applications in otolaryngology and the communication sciences and how to harness that power for future applications.

Compression and amplification algorithms in hearing aids impair the selectivity of neural responses to speech.

In quiet environments, hearing aids improve the perception of low-intensity sounds. However, for high-intensity sounds in background noise, the aids often fail to provide a benefit to the wearer. Here, using large-scale single-neuron recordings from hearing-impaired gerbils-an established animal model of human hearing-we show that hearing aids restore the sensitivity of neural responses to speech, but not their selectivity. Rather than reflecting a deficit in supra-threshold auditory processing, the low selectivity is a consequence of hearing-aid compression (which decreases the spectral and temporal contrasts of incoming sound) and amplification (which distorts neural responses, regardless of whether hearing is impaired). Processing strategies that avoid the trade-off between neural sensitivity and selectivity should improve the performance of hearing aids.

Why Do Hearing Aids Fail to Restore Normal Auditory Perception?

Hearing loss is a widespread condition that is linked to declines in quality of life and mental health. Hearing aids remain the treatment of choice, but, unfortunately, even state-of-the-art devices provide only limited benefit for the perception of speech in noisy environments. While traditionally viewed primarily as a loss of sensitivity, hearing loss is also known to cause complex distortions of sound-evoked neural activity that cannot be corrected by amplification alone. This Opinion article describes the effects of hearing loss on neural activity to illustrate the reasons why current hearing aids are insufficient and to motivate the use of new technologies to explore directions for improving the next generation of devices.

Midbrain adaptation may set the stage for the perception of musical beat.

The ability to spontaneously feel a beat in music is a phenomenon widely believed to be unique to humans. Though beat perception involves the coordinated engagement of sensory, motor and cognitive processes in humans, the contribution of low-level auditory processing to the activation of these networks in a beat-specific manner is poorly understood. Here, we present evidence from a rodent model that midbrain preprocessing of sounds may already be shaping where the beat is ultimately felt. For the tested set of musical rhythms, on-beat sounds on average evoked higher firing rates than off-beat sounds, and this difference was a defining feature of the set of beat interpretations most commonly perceived by human listeners over others. Basic firing rate adaptation provided a sufficient explanation for these results. Our findings suggest that midbrain adaptation, by encoding the temporal context of sounds, creates points of neural emphasis that may influence the perceptual emergence of a beat.

Stimulus relevance modulates contrast adaptation in visual cortex.

A general principle of sensory processing is that neurons adapt to sustained stimuli by reducing their response over time. Most of our knowledge on adaptation in single cells is based on experiments in anesthetized animals. How responses adapt in awake animals, when stimuli may be behaviorally relevant or not, remains unclear. Here we show that contrast adaptation in mouse primary visual cortex depends on the behavioral relevance of the stimulus. Cells that adapted to contrast under anesthesia maintained or even increased their activity in awake naïve mice. When engaged in a visually guided task, contrast adaptation re-occurred for stimuli that were irrelevant for solving the task. However, contrast adaptation was reversed when stimuli acquired behavioral relevance. Regulation of cortical adaptation by task demand may allow dynamic control of sensory-evoked signal flow in the neocortex.

Inhibitory control of correlated intrinsic variability in cortical networks.

Cortical networks exhibit intrinsic dynamics that drive coordinated, large-scale fluctuations across neuronal populations and create noise correlations that impact sensory coding. To investigate the network-level mechanisms that underlie these dynamics, we developed novel computational techniques to fit a deterministic spiking network model directly to multi-neuron recordings from different rodent species, sensory modalities, and behavioral states. The model generated correlated variability without external noise and accurately reproduced the diverse activity patterns in our recordings. Analysis of the model parameters suggested that differences in noise correlations across recordings were due primarily to differences in the strength of feedback inhibition. Further analysis of our recordings confirmed that putative inhibitory neurons were indeed more active during desynchronized cortical states with weak noise correlations. Our results demonstrate that network models with intrinsically-generated variability can accurately reproduce the activity patterns observed in multi-neuron recordings and suggest that inhibition modulates the interactions between intrinsic dynamics and sensory inputs to control the strength of noise correlations.

Periodotopy in the gerbil inferior colliculus: local clustering rather than a gradient map.

Periodicities in sound waveforms are widespread, and shape important perceptual attributes of sound including rhythm and pitch. Previous studies have indicated that, in the inferior colliculus (IC), a key processing stage in the auditory midbrain, neurons tuned to different periodicities might be arranged along a periodotopic axis which runs approximately orthogonal to the tonotopic axis. Here we map out the topography of frequency and periodicity tuning in the IC of gerbils in unprecedented detail, using pure tones and different periodic sounds, including click trains, sinusoidally amplitude modulated (SAM) noise and iterated rippled noise. We found that while the tonotopic map exhibited a clear and highly reproducible gradient across all animals, periodotopic maps varied greatly across different types of periodic sound and from animal to animal. Furthermore, periodotopic gradients typically explained only about 10% of the variance in modulation tuning between recording sites. However, there was a strong local clustering of periodicity tuning at a spatial scale of ca. 0.5 mm, which also differed from animal to animal.

An Overrepresentation of High Frequencies in the Mouse Inferior Colliculus Supports the Processing of Ultrasonic Vocalizations.

Mice are of paramount importance in biomedical research and their vocalizations are a subject of interest for researchers across a wide range of health-related disciplines due to their increasingly important value as a phenotyping tool in models of neural, speech and language disorders. However, the mechanisms underlying the auditory processing of vocalizations in mice are not well understood. The mouse audiogram shows a peak in sensitivity at frequencies between 15-25 kHz, but weaker sensitivity for the higher ultrasonic frequencies at which they typically vocalize. To investigate the auditory processing of vocalizations in mice, we measured evoked potential, single-unit, and multi-unit responses to tones and vocalizations at three different stages along the auditory pathway: the auditory nerve and the cochlear nucleus in the periphery, and the inferior colliculus in the midbrain. Auditory brainstem response measurements suggested stronger responses in the midbrain relative to the periphery for frequencies higher than 32 kHz. This result was confirmed by single- and multi-unit recordings showing that high ultrasonic frequency tones and vocalizations elicited responses from only a small fraction of cells in the periphery, while a much larger fraction of cells responded in the inferior colliculus. These results suggest that the processing of communication calls in mice is supported by a specialization of the auditory system for high frequencies that emerges at central stations of the auditory pathway.

Nonlinear transfer of signal and noise correlations in cortical networks.

Signal and noise correlations, a prominent feature of cortical activity, reflect the structure and function of networks during sensory processing. However, in addition to reflecting network properties, correlations are also shaped by intrinsic neuronal mechanisms. Here we show that spike threshold transforms correlations by creating nonlinear interactions between signal and noise inputs; even when input noise correlation is constant, spiking noise correlation varies with both the strength and correlation of signal inputs. We characterize these effects systematically in vitro in mice and demonstrate their impact on sensory processing in vivo in gerbils. We also find that the effects of nonlinear correlation transfer on cortical responses are stronger in the synchronized state than in the desynchronized state, and show that they can be reproduced and understood in a model with a simple threshold nonlinearity. Since these effects arise from an intrinsic neuronal property, they are likely to be present across sensory systems and, thus, our results are a critical step toward a general understanding of how correlated spiking relates to the structure and function of cortical networks.

State-dependent population coding in primary auditory cortex.

Sensory function is mediated by interactions between external stimuli and intrinsic cortical dynamics that are evident in the modulation of evoked responses by cortical state. A number of recent studies across different modalities have demonstrated that the patterns of activity in neuronal populations can vary strongly between synchronized and desynchronized cortical states, i.e., in the presence or absence of intrinsically generated up and down states. Here we investigated the impact of cortical state on the population coding of tones and speech in the primary auditory cortex (A1) of gerbils, and found that responses were qualitatively different in synchronized and desynchronized cortical states. Activity in synchronized A1 was only weakly modulated by sensory input, and the spike patterns evoked by tones and speech were unreliable and constrained to a small range of patterns. In contrast, responses to tones and speech in desynchronized A1 were temporally precise and reliable across trials, and different speech tokens evoked diverse spike patterns with extremely weak noise correlations, allowing responses to be decoded with nearly perfect accuracy. Restricting the analysis of synchronized A1 to activity within up states yielded similar results, suggesting that up states are not equivalent to brief periods of desynchronization. These findings demonstrate that the representational capacity of A1 depends strongly on cortical state, and suggest that cortical state should be considered as an explicit variable in all studies of sensory processing.

The neural representation of interaural time differences in gerbils is transformed from midbrain to cortex.

Interaural time differences (ITDs) are the dominant cue for the localization of low-frequency sounds. While much is known about the processing of ITDs in the auditory brainstem and midbrain, there have been relatively few studies of ITD processing in auditory cortex. In this study, we compared the neural representation of ITDs in the inferior colliculus (IC) and primary auditory cortex (A1) of gerbils. Our IC results were largely consistent with previous studies, with most cells responding maximally to ITDs that correspond to the contralateral edge of the physiological range. In A1, however, we found that preferred ITDs were distributed evenly throughout the physiological range without any contralateral bias. This difference in the distribution of preferred ITDs in IC and A1 had a major impact on the coding of ITDs at the population level: while a labeled-line decoder that considered the tuning of individual cells performed well on both IC and A1 responses, a two-channel decoder based on the overall activity in each hemisphere performed poorly on A1 responses relative to either labeled-line decoding of A1 responses or two-channel decoding of IC responses. These results suggest that the neural representation of ITDs in gerbils is transformed from IC to A1 and have important implications for how spatial location may be combined with other acoustic features for the analysis of complex auditory scenes.

Independent population coding of speech with sub-millisecond precision.

To understand the strategies used by the brain to analyze complex environments, we must first characterize how the features of sensory stimuli are encoded in the spiking of neuronal populations. Characterizing a population code requires identifying the temporal precision of spiking and the extent to which spiking is correlated, both between cells and over time. In this study, we characterize the population code for speech in the gerbil inferior colliculus (IC), the hub of the auditory system where inputs from parallel brainstem pathways are integrated for transmission to the cortex. We find that IC spike trains can carry information about speech with sub-millisecond precision, and, consequently, that the temporal correlations imposed by refractoriness can play a significant role in shaping spike patterns. We also find that, in contrast to most other brain areas, the noise correlations between IC cells are extremely weak, indicating that spiking in the population is conditionally independent. These results demonstrate that the problem of understanding the population coding of speech can be reduced to the problem of understanding the stimulus-driven spiking of individual cells, suggesting that a comprehensive model of the subcortical processing of speech may be attainable in the near future.

Analysis and modelling of variability and covariability of population spike trains across multiple time scales.

As multi-electrode and imaging technology begin to provide us with simultaneous recordings of large neuronal populations, new methods for modelling such data must also be developed. We present a model of responses to repeated trials of a sensory stimulus based on thresholded Gaussian processes that allows for analysis and modelling of variability and covariability of population spike trains across multiple time scales. The model framework can be used to specify the values of many different variability measures including spike timing precision across trials, coefficient of variation of the interspike interval distribution, and Fano factor of spike counts for individual neurons, as well as signal and noise correlations and correlations of spike counts across multiple neurons. Using both simulated data and data from different stages of the mammalian auditory pathway, we demonstrate the range of possible independent manipulations of different variability measures, and explore how this range depends on the sensory stimulus. The model provides a powerful framework for the study of experimental and surrogate data and for analyzing dependencies between different statistical properties of neuronal populations.

Impaired auditory temporal selectivity in the inferior colliculus of aged Mongolian gerbils.

Aged humans show severe difficulties in temporal auditory processing tasks (e.g., speech recognition in noise, low-frequency sound localization, gap detection). A degradation of auditory function with age is also evident in experimental animals. To investigate age-related changes in temporal processing, we compared extracellular responses to temporally variable pulse trains and human speech in the inferior colliculus of young adult (3 month) and aged (3 years) Mongolian gerbils. We observed a significant decrease of selectivity to the pulse trains in neuronal responses from aged animals. This decrease in selectivity led, on the population level, to an increase in signal correlations and therefore a decrease in heterogeneity of temporal receptive fields and a decreased efficiency in encoding of speech signals. A decrease in selectivity to temporal modulations is consistent with a downregulation of the inhibitory transmitter system in aged animals. These alterations in temporal processing could underlie declines in the aging auditory system, which are unrelated to peripheral hearing loss. These declines cannot be compensated by traditional hearing aids (that rely on amplification of sound) but may rather require pharmacological treatment.

Differential connectivity and response dynamics of excitatory and inhibitory neurons in visual cortex.

Neuronal responses during sensory processing are influenced by both the organization of intracortical connections and the statistical features of sensory stimuli. How these intrinsic and extrinsic factors govern the activity of excitatory and inhibitory populations is unclear. Using two-photon calcium imaging in vivo and intracellular recordings in vitro, we investigated the dependencies between synaptic connectivity, feature selectivity and network activity in pyramidal cells and fast-spiking parvalbumin-expressing (PV) interneurons in mouse visual cortex. In pyramidal cell populations, patterns of neuronal correlations were largely stimulus-dependent, indicating that their responses were not strongly dominated by functionally biased recurrent connectivity. By contrast, visual stimulation only weakly modified co-activation patterns of fast-spiking PV cells, consistent with the observation that these broadly tuned interneurons received very dense and strong synaptic input from nearby pyramidal cells with diverse feature selectivities. Therefore, feedforward and recurrent network influences determine the activity of excitatory and inhibitory ensembles in fundamentally different ways.

The effects of interaural time difference and intensity on the coding of low-frequency sounds in the mammalian midbrain.

We examined how changes in intensity and interaural time difference (ITD) influenced the coding of low-frequency sounds in the inferior colliculus of male gerbils at both the single neuron and population levels. We found that changes in intensity along the positive slope of the rate-level function (RLF) evoked changes in spectrotemporal filtering that influenced the overall timing of spike events but preserved their precision across trials such that the decoding of single neuron responses was not affected. In contrast, changes in ITD did not trigger changes in spectrotemporal filtering, but did have strong effects on the precision of spike events and, consequently, on decoder performance. However, changes in ITD had opposing effects in the two brain hemispheres and, thus, canceled out at the population level. These results were similar with and without the addition of background noise. We also found that the effects of changes in intensity along the negative slope of the RLF were different from the effects of changes in intensity along the positive slope in that they evoked changes in both spectrotemporal filtering and in the precision of spike events across trials, as well as in decoder performance. These results demonstrate that, at least at moderate intensities, the auditory system employs different strategies at the single neuron and population levels simultaneously to ensure that the coding of sounds is robust to changes in other stimulus features.

Modeling Population Spike Trains with Specified Time-Varying Spike Rates, Trial-to-Trial Variability, and Pairwise Signal and Noise Correlations.

As multi-electrode and imaging technology begin to provide us with simultaneous recordings of large neuronal populations, new methods for modeling such data must also be developed. Here, we present a model for the type of data commonly recorded in early sensory pathways: responses to repeated trials of a sensory stimulus in which each neuron has it own time-varying spike rate (as described by its PSTH) and the dependencies between cells are characterized by both signal and noise correlations. This model is an extension of previous attempts to model population spike trains designed to control only the total correlation between cells. In our model, the response of each cell is represented as a binary vector given by the dichotomized sum of a deterministic "signal" that is repeated on each trial and a Gaussian random "noise" that is different on each trial. This model allows the simulation of population spike trains with PSTHs, trial-to-trial variability, and pairwise correlations that match those measured experimentally. Furthermore, the model also allows the noise correlations in the spike trains to be manipulated independently of the signal correlations and single-cell properties. To demonstrate the utility of the model, we use it to simulate and manipulate experimental responses from the mammalian auditory and visual systems. We also present a general form of the model in which both the signal and noise are Gaussian random processes, allowing the mean spike rate, trial-to-trial variability, and pairwise signal and noise correlations to be specified independently. Together, these methods for modeling spike trains comprise a potentially powerful set of tools for both theorists and experimentalists studying population responses in sensory systems.

Incremental mutual information: a new method for characterizing the strength and dynamics of connections in neuronal circuits.

Understanding the computations performed by neuronal circuits requires characterizing the strength and dynamics of the connections between individual neurons. This characterization is typically achieved by measuring the correlation in the activity of two neurons. We have developed a new measure for studying connectivity in neuronal circuits based on information theory, the incremental mutual information (IMI). By conditioning out the temporal dependencies in the responses of individual neurons before measuring the dependency between them, IMI improves on standard correlation-based measures in several important ways: 1) it has the potential to disambiguate statistical dependencies that reflect the connection between neurons from those caused by other sources (e.g. shared inputs or intrinsic cellular or network mechanisms) provided that the dependencies have appropriate timescales, 2) for the study of early sensory systems, it does not require responses to repeated trials of identical stimulation, and 3) it does not assume that the connection between neurons is linear. We describe the theory and implementation of IMI in detail and demonstrate its utility on experimental recordings from the primate visual system.