Niki Katerina Vavatzanidis

Objective. Electroencephalography (EEG) is a key tool for studying auditory processing in cochlear implant (CI) users. In particular, EEG recordings obtained during continuous speech are becoming increasingly important for assessing speech and language processing in CI users, and may be utilized for neurofeedback. However, CIs also induce strong stimulation artifacts that are time-locked to the stimulus and mask the neural responses that have smaller magnitudes. Existing artifact reduction methods are typically based on event-related potentials or require manual component selection, making them unsuitable for naturalistic listening conditions or large datasets. Approach. We develop CORSICA (CORrelation-baSed ICA artifact rejection), a reproducible, parameter-efficient method for CI artifact reduction in EEG responses to continuous speech. CORSICA operates on independent components (ICs) obtained through Infomax ICA and requires no manual component labeling, with performance governed by a single tunable threshold. It exploits the observation that CI artifacts temporally follow the audio signal without delay, whereas neural responses have an inherent lag due to auditory pathway latencies. For each IC, CORSICA computes the cross-correlation with the speech stimulus. Artifacts are identified by a high signal-to-noise ratio (SNR) of the correlation peak near zero lag, and the component is rejected if this SNR exceeds a threshold. To benchmark CORSICA, we evaluate two alternatives: a TRF-based SNR method, in which temporal response functions are fitted to each IC and artifact-driven peaks near zero lag are used for rejection, and a variant replacing ICA with second-order blind identification (SOBI) as the source separation step. Main results. CORSICA effectively suppressed CI artifacts while preserving neural activity, enabling recovery of physiologically plausible TRFs with only 2% of ICs rejected. Both benchmark methods confirmed the validity of the SNR-based rejection framework, but CORSICA outperformed the TRF-based alternative in artifact suppression quality. Replacing ICA with SOBI as the source separation step required more ICs to be rejected, further supporting ICA as the preferred backbone for CORSICA. Significance. CORSICA provides a fully objective, label-free approach to identifying CI artifacts in speech-evoked EEG data, with no manual intervention required. By centering artifact rejection on a single interpretable threshold, it offers a reproducible preprocessing standard for future EEG studies on speech processing in CI users. Conclusion. Our findings demonstrate that objective CI artifact suppression in speech-evoked EEG data is feasible on the basis of the IC’s temporal response patterns.

Objective: Electroencephalography (EEG) is a key tool for studying auditory processing in cochlear implant (CI) users. In particular, EEG recordings obtained during continuous speech are becoming increasingly important for assessing speech and language processing in CI users, and may be utilized for neurofeedback. However, CIs also induce strong stimulation artifacts that are time-locked to the stimulus and mask the neural responses that have smaller magnitudes. Existing artifact reduction methods are typically based on event-related potentials (ERPs) or require manual component selection, making them unsuitable for naturalistic listening conditions or large datasets. Approach: We develop CORSICA (CORrelation-baSed ICA artifact rejection), a reproducible, parameter-efficient method for CI artifact reduction in EEG responses to continuous speech. CORSICA operates on independent components (ICs) obtained through Infomax ICA and requires no manual component labelling, with performance governed by a single tunable threshold. It exploits the observation that CI artifacts temporally follow the audio signal without delay, whereas neural responses have an inherent lag due to auditory pathway latencies. For each IC, CORSICA computes the cross-correlation with the speech stimulus. Artifacts are identified by a high signal-to-noise ratio (SNR) of the correlation peak near zero lag, and the component is rejected if this SNR exceeds a threshold. To benchmark CORSICA, we evaluate two alternatives: a TRF-based SNR method, in which temporal response functions are fitted to each IC and artifact-driven peaks near zero lag are used for rejection, and a variant replacing ICA with second-order blind identification (SOBI) as the source separation step. Main results: CORSICA effectively suppressed CI artifacts while preserving neural activity, enabling recovery of physiologically plausible TRFs with only 2% of ICs rejected. Both benchmark methods confirmed the validity of the SNR-based rejection framework, but CORSICA outperformed the TRF-based alternative in artifact suppression quality. Replacing ICA with SOBI as the source separation step required more ICs to be rejected, further supporting ICA as the preferred backbone for CORSICA. Significance: CORSICA provides a fully objective, label-free approach to identifying CI artifacts in speech-evoked EEG data, with no manual intervention required. By centering artifact rejection on a single interpretable threshold, it offers a reproducible preprocessing standard for future EEG studies on speech processing in CI users. Conclusion: Our findings demonstrate that objective CI artifact suppression in speech-evoked EEG data is feasible on the basis of the IC’s temporal response patterns.

Users of hearing aids (HAs) and cochlear implants (CIs) experience significant difficulty understanding a target speaker in multi-talker environments or when other background noise is present. Segregation of a particular voice from background noise occurs partly through enhanced cortical tracking of amplitude fluctuations in the target signal. Measuring a person's cortical tracking allows decoding their focus of attention and may be used for neurofeedback in hearing devices, potentially aiding their users with speech-in-noise comprehension. Most studies on cortical tracking have employed typical hearing (TH) individuals, whereas studies in people with hearing impairment whose cortical tracking may differ are still scarce. The objective of this study was to compare cortical speech tracking of HA (n=29) and CI users (n=24) to that of age-matched TH individuals (n=29). We recorded EEG data while the participants attended one of two competing talkers (one with a female and one with a male voice), in a free-field acoustic environment. Importantly, HA users as well as CI users used their personal, clinically-fitted devices. Cortical speech tracking was assessed through linear backward and forward models that related the EEG data to the speech envelope. For the CI users, electrical artifacts stemming from the implant were addressed through a bespoke method for artifact rejection. We found that the HA group exhibited cortical tracking and attentional modulation that were largely comparable to those of the TH group. CI users also showed successful cortical tracking. However, they displayed a profound deficit in attentional modulation, seen in the significantly poorer neural segregation of the attended vs. the ignored speech streams. These results shed light on a neurobiological mechanism for speech-in-noise comprehension and have implications for neurofeedback in hearing devices.

Objective. Understanding speech in the presence of background noise such as other speech streams is a difficult problem for people with hearing impairment, and in particular for users of cochlear implants (CIs). To improve their listening experience, auditory attention decoding (AAD) aims to decode the target speaker of a listener from electroencephalography (EEG), and then use this information to steer an auditory prosthesis towards this speech signal. In normal-hearing individuals, deep neural networks (DNNs) have been shown to improve AAD compared to simpler linear models. We aim to demonstrate that DNNs can improve attention decoding in CI users too, which would make them the state-of-the-art candidate for a neuro-steered CI. Approach. To this end, we first collected an EEG dataset on selective auditory attention from 25 bilateral CI users, and then implemented both a linear model as well as a convolutional neural network (CNN) for attention decoding. Moreover, we introduced a novel, objective CI-artifact removal strategy and evaluated its impact on decoding accuracy, alongside learnable speaker classification using a support vector machine (SVM). Main results. The CNN outperformed the linear model across all decision window sizes from 1 to 60 s. Removing CI artifacts modestly improved the CNN’s decoding accuracy. With SVM classification, the CNN decoder reached a peak mean decoding accuracy of 74% at the population level for a 60 s decision window. Significance. These results demonstrate the superior potential of CNN-based decoding for neuro-steered CIs, which could improve speech perception of its users in cocktail party situations significantly.

In the present study we explore the implications of acquiring language when relying mainly or exclusively on input from a cochlear implant (CI), a device providing auditory input to otherwise deaf individuals. We focus on the time course of semantic learning in children within the second year of implant use; a period that equals the auditory age of normal hearing children during which vocabulary emerges and extends dramatically. 32 young bilaterally implanted children saw pictures paired with either matching or non-matching auditory words. Their electroencephalographic responses were recorded after 12, 18 and 24 months of implant use, revealing a large dichotomy: Some children failed to show semantic processing throughout their second year of CI use, which fell in line with their poor language outcomes. The majority of children, though, demonstrated semantic processing in form of the so-called N400 effect already after 12 months of implant use, even when their language experience relied exclusively on the implant. This is slightly earlier than observed for normal hearing children of the same auditory age, suggesting that more mature cognitive faculties at the beginning of language acquisition lead to faster semantic learning.

Children with sensorineural hearing loss may (re)gain hearing with a cochlear implant – a device that transforms sounds into electric pulses and bypasses the dysfunctioning inner ear by stimulating the auditory nerve directly with an electrode array. Many implanted children master the acquisition of spoken language successfully, yet we still have little knowledge of the actual input they receive with the implant and specifically which language sensitive cues they hear. This would be important however, both for understanding the flexibility of the auditory system when presented with stimuli after a (life-)long phase of deprivation and for planning therapeutic intervention. In rhythmic languages the general stress pattern conveys important information about word boundaries. Infant language acquisition relies on such cues and can be severely hampered when this information is missing, as seen for dyslexic children and children with specific language impairment. Here we ask whether children with a cochlear implant perceive differences in stress patterns during their language acquisition phase and if they do, whether it is present directly following implant stimulation or if and how much time is needed for the auditory system to adapt to the new sensory modality. We performed a longitudinal ERP study, testing in bimonthly intervals the stress pattern perception of 17 young hearing impaired children (age range: 9-50 months; mean: 22 months) during their first 6 months of implant use. An additional session before the implantation served as control baseline. During a session they passively listened to an oddball paradigm featuring the disyllable “baba”, which was stressed either on the first or second syllable (trochaic vs. iambic stress pattern). A group of age-matched normal hearing children participated as controls. Our results show, that within the first 6 months of implant use the implanted children develop a negative mismatch response for iambic but not for trochaic deviants, thus showing the same result as the normal hearing controls. Even congenitally deaf children show the same developing pattern. We therefore conclude a) that young implanted children have early access to stress pattern information and b) that they develop ERP responses similar to those of normal hearing children.