CNS*2020 Online has ended
Welcome to the Sched instance for CNS*2020 Online! Please read the instruction document on detailed information on CNS*2020.
Back To Schedule
Sunday, July 19 • 7:00pm - 8:00pm
P130: Unifying information theory and machine learning in a model of cochlear implant electrode discrimination

Log in to save this to your schedule, view media, leave feedback and see who's attending!

Feedback form is now closed.
Currently having an internet disruption at home, if you have any question please send me an email xiao.gao@unimelb.edu.au, I am more than happy to answer your questions. 

Xiao Gao
, David Grayden, Mark McDonnell

Despite the success of cochlear implants (CIs) over more than three decades, wide inter-subject variability in speech perception is reported [1]. The key factors that cause variability between users are unclear. We previously developed an information theoretic modelling framework that enables estimation of the optimal number of electrodes and quantification of electrode discrimination ability [2, 3]. However, the optimal number of electrodes was estimated based only on statistical correlations between channel outputs and inputs, and the model did not quantitatively model psychophysical measurements and study inter-subject variability.

Here, we unified information theoretic and machine learning techniques to investigate the key factors that may limit the performance of CIs. The framework used a neural network classifier to predict which electrode was stimulated for a given simulated activation pattern of the auditory nerve, and mutual information was then estimated between the actual stimulated electrode and the predicted one.

Using the framework, electrode discrimination was quantified with a range of parameter choices, as shown in Fig. 1. The columns from left to right show how the distance between electrodes and auditory nerve fibres, _r_ , the number of surviving fibres, _N_ , the maximum current level (modelled as the percentage of surviving fibres, _N_ , that generate action potentials for a given stimulated electrode), and the attenuation in electrode current, _A_ , affect the model performance, respectively. The parameters were chosen to reflect the key factors that are believed to limit the performance of CIs. The model shows sensitivity to parameter choices, where smaller _r_ , larger _N_ , __ and higher attenuation in current lead to higher mutual information and improved classification.

This approach provides a flexible framework that may be used to investigate the key factors that limit the performance of cochlear implants. We aim to investigate its application to personalised configurations of CIs.


This work is supported by a McKenzie Fellowship, The University of Melbourne.


[1] Holden LK, Finley CC et al. “Affecting Open-Set word recognition in adults with cochlear implants”, Ear Hear, Vol. 34, 342-360, 2013

[2] Gao X, Grayden DB, McDonnell MD, “Stochastic information transfer from cochlear implant electrodes to auditory nerve fibers”, _Physical Review E_ 90 (2014) 022722.

[3] Gao X, Grayden DB, McDonnell MD, “Modeling electrode place discrimination in cochlear implant stimulation _”, IEEE Transactions on Biomedical Engineering_ 64 (2017) 2219–2229.

avatar for (Demi) Xiao Gao

(Demi) Xiao Gao

McKenzie Research Fellow, Department of Biomedical Engineering, University of Melbourne
Demi Xiao Gao is a Mckenzie research fellow in the Department of Biomedical Engineering with an honorary appointment in the School of Physics, University of Sydney. Demi received her Bachelor degree in Computer Science and Master in Biology, and completed her Ph.D in Information Technology... Read More →

Sunday July 19, 2020 7:00pm - 8:00pm CEST
Slot 11