Authors: C. Garcia1, T. Goehring1, S. Cosentino1, R.E. Turner2, J.M. Deeks1, T. Brochier, T. Rughooputh1, M. Bance3, R.P. Carlyon1
1MRC Cognition & Brain Sciences Unit, University of Cambridge, Cambridge, CB2 7EF
2Department of Engineering, University of Cambridge, Cambridge, CB2 1PZ
3Department of Clinical Neurosciences, University of Cambridge, Cambridge, CB2 0SP
Background: While cochlear implants (CIs) have had much success in restoring hearing to otherwise deaf individuals, there is a lot of variability in outcomes and many recipients still struggle with open-set speech recognition. Having knowledge of patient-specific neural excitation patterns of CIs can provide important information for optimizing efficacy and improving speech perception of users. The Panoramic ECAP Method (or PECAP) is an objective tool that is uses the forward masking technique for recording Electrically evoked Compound Action Potentials (ECAPs) from every possible combination of masker and probe electrodes in a CI. It leverages a non-linear optimisation algorithm to provide a patient-specific model of the electrode-neuron interface, considering the combined contributions of current spread and neural health and their variations along the length of the cochlea.
Methods: The accuracy of PECAP and its robustness to noise were first evaluated with computer simulations. Ten different current spread and neural health scenarios were simulated at various signal-to-noise ratios (SNRs) and the resultant ECAP measurement matrix was submitted to the PECAP algorithm. This provided an opportunity to evaluate the ability of PECAP to recreate ‘ground truth’ current spread and neural health estimates and its degradation with decreasing SNR. Secondly, data was provided from DeVries et al (2016) enabling comparison between PECAP estimates of current spread and neural health with electrode-to-modiolus distances and focused thresholds in 9 human Advanced Bionics users. Finally, local areas of neural death were simulated in 7 human Cochlear Corporation users, allowing a within-participant evaluation of PECAP’s ability to identify the simulated dead region when compared to a control condition.
Results / Conclusions: PECAP was able to reconstruct computer simulated neural excitations patterns with less than 10% error at SNRs down to 10dB. A moderate but significant negative correlation was found between PECAP’s estimate of neural health and focused thresholds, suggesting consistency with other estimates of neural health in the literature. No significant correlation was found between PECAP’s estimate of current spread and electrode-to-modiolus distances. However, PECAP was able to identify all of the simulated dead regions in human CI data while maintaining largely unchanged estimates of current spread between conditions. These results support the hypothesis that PECAP is able to accurately model variations in current spread and neural health along the cochlea in CI users. While based on ECAPs and therefore only representative of peripheral, synchronous neural activity in the cochlea, the results suggest that PECAP may provide useful patient-specific information about electrode-neural interfaces to inform and optimize CI stimulation strategies in clinical settings.
Acknowledgements: Supported by the Armstrong Fund and by Action on Hearing Loss (UK, grant number 82)
References: DeVries et al (2016) JARO 17, 237-252.
Dear authors,
This is a really nice approach and very interesting indeed.
My question is: What’s the measurement of “neural health” used in this study (estimate of fiber count, eCAP amplitude, Polarity Sensitivity, etc.)?
Hi Igancio. This is a great question – thank you. The ‘neural health’ estimate in this study is the output of an optimisation algorithm that fits neural activation patterns based entirely on relative ECAP amplitudes. It does not include measurements of polarity sensitivity, nor does it aim to give an estimate of the actual fibre count. As ECAPs are understood to reflect synchronous, peripheral neural responsiveness within the cochlea, it can be interpreted to represent the health of the peripheral processes. The actual unit is slightly irrelevant: the point of the algorithm is to give an estimate of relative peripheral neural health along the length of a specific cochlear implant user’s cochlea. The intention is for the metric to be used in a precision medicine type approach, for example to make decisions about tailoring CI processing strategies to individual patients based on their specific pattern of electrode-neuron-interface, as opposed to making comparisons between different patients.
Hi Charlotte! Thank you so much for your answer!