Realising the potential of cochlear implants

A capacity to adapt to altered inputs is critical for maintaining perceptual abilities following sensory impairments. We have shown that the auditory system can compensate for changes in the balance of inputs between the two ears, both during development and in later life, and maintain accurate sound localization in spite of reduced hearing in one ear. Our research in animals and humans has demonstrated that this can be achieved either through adaptive shifts in neuronal sensitivity to the altered binaural cues or by reweighting different spatial cues according to their relative reliability. The degree to which each strategy for adaptation is used depends on the way that subjects are trained, suggesting that targeted training strategies may be helpful for restoring sound localization in clinical populations according to their residual hearing ability and therefore the localization cues that remain available. Importantly, our recent work suggests that adaptation to asymmetric hearing loss generalizes to untrained sounds and to more challenging listening conditions. Furthermore, we have shown that visual information can influence auditory spatial judgments and the weighting of different spatial cues, and that a multisensory training paradigm can help to improve auditory spatial processing and perception following bilateral cochlear implantation. 

Congenital deafness affects the functional and anatomical properties of the auditory system (Kral et al, 2019, Ann Rev Neurosci). It has been suggested that congenital deafness affects predominantly corticocortical functional connections within but also beyond the auditory system (the connectome model of deafness, Kral et al, 2016, Lancet Neurol). This may lead to increased risk of cognitive deficits in deaf children. Indeed, so-called induced responses, indicative of corticocortical interactions, were reduced in the auditory cortex of congenitally deaf cats (Yusuf et al, 2017, Brain). Here Dr Kral’s group directly investigated effective connectivity between primary and secondary areas of congenitally deaf cats (CDC).

In adult hearing cats (HC) and CDCs, responses to acoustic and electric stimulation (through a cochlear implant) were compared. Recordings were in the primary auditory field (A1) and the higher order posterior auditory field (PAF) using multielectrode arrays. Penetrations were histologically reconstructed. For effective connectivity pairwise phase consistency, weighted phase-lag index and nonparametric Granger causality were determined and compared. Additionally, spike-field coherence was computed between different recording sites.

CDCs demonstrated a substantially reduced stimulus-related corticocortical coupling in the connectivity measures used. Largest deficits were observed in sensory-related top-down interactions, in the alpha and beta band (Yusuf et al, 2021, Front Neurosci). Furthermore, spike-field coherence revealed a decoupling of supragranular and infragranular layers in A1 (Yusuf et al, 2022, Front Syst Neurosci), likely responsible for dystrophic changes in the deep cortical layers (Berger et al, 2017, J Comp Neurol). The data document that corticocortical interactions are dependent on developmental hearing experience. These observations suggests that the congenitally deaf brain cannot incorporate top-down prediction information into auditory processing and thus have a deficient predictive processing.

To successfully design devices for the human body, engineers often view the body itself as the ideal design template. Similarly, for individuals missing a limb, the development of artificial prosthetic limbs often centers on embodiment as the goal: focusing device design and control on becoming more like our biological bodies. But ultimately, the success of artificial limb will critically depend on its neural representation in our brains. Importantly, neurocognitive resources might differ radically, depending on the user’s life experiences and needs. Here I will present a series of studies where we investigated the neural basis of artificial limb use for both substitution and augmentation technologies. We find that contrary to folk wisdom, the brain does not assimilate neural representations for the artificial limb with those for the biological body, creating opportunities for novel technological interfaces. Collectively, these studies suggest that although, in principle, opportunities exist for harnessing hand neural and cognitive resources to control artificial limbs, alternative non-biomimetic approaches could be also well suited for successful human-device interface.  

Cochlear implant (CI) users with a prelingual onset of hearing loss show poor sensitivity to interaural time differences (ITDs), an important cue for sound localization and speech reception in noise. Similarly, neural ITD sensitivity in the inferior colliculus (IC) of neonatally-deafened animals is degraded compared with animals deafened as adults. Here, neonatally-deafened rabbits were provided with chronic bilateral CI stimulation to investigate whether auditory experience through bilateral CIs during development can reverse the effect of early-onset deafness on ITD sensitivity. In the early-deaf rabbits that received chronic stimulation during development, the prevalence of ITD sensitive neurons was restored to the level of adult-deaf rabbits. Also neural ITD sensitivity in the early-deaf and stimulated rabbits improved partially. In contrast, chronic CI stimulation did not improve temporal coding in early-deaf rabbits. These findings highlight the importance of auditory experience during development on the maturation of binaural circuitry.

Our lab studies children and adults who receive bilateral cochlear implants (BiCI), or adults with single-sided deafness receiving a CI in the deaf ear (SSD-CI). Bilateral hearing typically improves localization of sounds and segregation of speech from background noise compared with unilateral hearing. However, patients typically perform worse than normal hearing listeners. We use several approaches to understand mechanisms driving gaps in performance, including asymmetry in sensitivity to monaural information such as modulation detection. Further interaural mismatch in place of stimulation along the cochlea, age at onset of deafness and age at implantation contribute to recovery of binaural hearing. Because CI processors do not preserve binaural cues with fidelity, we use research processors to generate multi-channel binaural stimulation strategies that introduce different rates of stimulation across the electrode arrays, thereby preserving rates that are important for both binaural sensitivity and speech understanding. In addition, eye gaze studies reveal developmental factors that in decision-making that are not observed with measures of threshold. Finally, pupillometry studies provide insight into the impact of integrating inputs from two ears, whereby in some instances improved performance with two ears can be “costly” in the listening effort domain.

Bilateral cochlear implants (CIs) greatly improve the spatial hearing of CI users compare to those with only one implant. However substantial gaps still exist between bilateral CI users and normal hearing listeners in different aspects, such as their lateralization and localization performance. Computer models are expected to partially reveal the possible reasons behind such gaps, by dissecting the specific contribution of each stage along the whole processing chain. In order to provide a model framework for predicating bilateral CI users’ lateralization or localization performance in a variety of experimental scenarios, here a recently published electrically stimulated single excitation-inhibition (EI) model neuron (Hu et al. 2022, JARO) was extend to population-EI-model neurons. The proposed model framework includes (a) parameter initialization, (b) binaural signal generating, (c) switchable CI processing, (d) auditory nerve (AN)- and EI- neuron processing, and (e) decision model stages. For demonstration purposes, several bilateral CIs experiments were simulated, such as pulse rate limitation of ITD sensitivity; left/right discrimination or lateralization; free-field localization with and without two independent automatic gain controls (AGCs). In general, the model framework could capture the average performance of all selected experiments even with the same EI-model units and a very simplified decision model.

Bilateral cochlear implants (CIs) have provided hearing advantages over use of a single CI in children with deafness in both ears. Yet, the goal of promoting normal binaural/spatial hearing development in these children remains elusive.  Evidence of remaining abnormalities in children using bilateral CIs comes from behavioral measures of impaired sound localization and sensitivity to binaural cues. These challenges may reflect abnormalities in integration of bilateral CI input in the auditory brainstem and cortex, including disruptions in functional connectivity, as measured with electrophysiology in children. We suggest that mismatches in bilateral CI input contribute to binaural/spatial hearing challenges by creating an aural preference for one ear when implantation of the opposite ear is delayed or allowing ongoing asymmetries or mismatches in stimulation of the two ears. While the former problem can be addressed through early access to CI in each candidate ear, there is no present protocol to harmonize the separate program parameters and processing of independent CIs.  Present studies seek to identify these sources of mismatched bilateral CI input and to provide methods that might be used to resolve these challenges to binaural hearing development.

The provision of a cochlear implant to patients who are single-sided deaf has allowed researchers to estimate the sound quality, or voice, of a cochlear implant. Clean signals can be directed to the implanted ear and then signals that might sound like the CI (candidate signals) can be directed to the normal hearing ear. Patients can then tell the researcher how close the candidate signals come to the sound of the CI. To create close approximations to CI sound quality, the group currently choose from the following operations: 1) flatten the intonation contour, 2) upshift F0, 3) upshift formant frequencies, 4) smear the signal in the frequency domain, 5) add a flange effect, 6) combine natural speech signals and noise or sine vocoders, 7) smear the signal in the time domain and 8) high pass, low pass or band-pass the signal. Altering natural speech signals using subsets of these operations has allowed us to create close matches to CI sound quality for many patients. A muffled sound quality dominates the matches for many patients while a higher ‘pitch’ created by an upshifted formant pattern or F0 dominates the matches for others. A small number of patients match to signals that are almost clean – an outcome that seems very unlikely.