|Abstract: ||The modern cochlear implant is one of the most successful neural stimulation devices, which partially mimics the workings of the auditory periphery. In the last few decades it has created a paradigm shift in hearing restoration of the deaf population, which has led to more than 324,000 cochlear implant users today. Despite its great success there is great disparity in patient outcomes without clear understanding of the aetiology of this variance in implant performance. Furthermore speech recognition in adverse conditions or music appreciation is still not attainable with today's commercial technology. This motivates the research for the next generation of cochlear implants that takes advantage of recent developments in electronics, neuroscience, nanotechnology, micro-mechanics, polymer chemistry and molecular biology to deliver high fidelity sound.
The main difficulties in determining the root of the problem in the cases where the cochlear implant does not perform well are two fold: first there is not a clear paradigm on how the electrical stimulation is perceived as sound by the brain, and second there is limited understanding on the plasticity effects, or learning, of the brain in response to electrical stimulation. These significant knowledge limitations impede the design of novel cochlear implant technologies, as the technical specifications that can lead to better performing implants remain undefined.
The motivation of the work presented in this thesis is to compare and contrast the cochlear implant neural stimulation with the operation of the physiological healthy auditory periphery up to the level of the auditory nerve. As such design of novel cochlear implant systems can become feasible by gaining insight on the question `how well does a specific cochlear implant system approximate the healthy auditory periphery?' circumventing the necessity of complete understanding of the brain's comprehension of patterned electrical stimulation delivered from a generic cochlear implant device.
A computational model, termed Digital Cochlea Stimulation and Evaluation Tool (‘DiCoStET’) has been developed to provide an objective estimate of cochlear implant performance based on neuronal activation measures, such as vector strength and average activation. A patient-specific cochlea 3D geometry is generated using a model derived by a single anatomical measurement from a patient, using non-invasive high resolution computed tomography (HRCT), and anatomically invariant human metrics and relations. Human measurements of the neuron route within the inner ear enable an innervation pattern to be modelled which joins the space from the organ of Corti to the spiral ganglion subsequently descending into the auditory nerve bundle. An electrode is inserted in the cochlea at a depth that is determined by the user of the tool. The geometric relation between the stimulation sites on the electrode and the spiral ganglion are used to estimate an activating function that will be unique for the specific patient's cochlear shape and electrode placement. This `transfer function', so to speak, between electrode and spiral ganglion serves as a `digital patient' for validating novel cochlear implant systems. The novel computational tool is intended for use by bioengineers, surgeons, audiologists and neuroscientists alike.
In addition to ‘DiCoStET’ a second computational model is presented in this thesis aiming at enhancing the understanding of the physiological mechanisms of hearing, specifically the workings of the auditory synapse. The purpose of this model is to provide insight on the sound encoding mechanisms of the synapse. A hypothetical mechanism is suggested in the release of neurotransmitter vesicles that permits the auditory synapse to encode temporal patterns of sound separately from sound intensity.
DiCoStET was used to examine the performance of two different types of filters used for spectral analysis in the cochlear implant system, the Gammatone type filter and the Butterworth type filter. The model outputs suggest that the Gammatone type filter performs better than the Butterworth type filter. Furthermore two stimulation strategies, the Continuous Interleaved Stimulation (CIS) and Asynchronous Interleaved Stimulation (AIS) have been compared. The estimated neuronal stimulation spatiotemporal patterns for each strategy suggest that the overall stimulation pattern is not greatly affected by the temporal sequence change. However the finer detail of neuronal activation is different between the two strategies, and when compared to healthy neuronal activation patterns the conjecture is made that the sequential stimulation of CIS hinders the transmission of sound fine structure information to the brain.
The effect of the two models developed is the feasibility of collaborative work emanating from various disciplines; especially electrical engineering, auditory physiology and neuroscience for the development of novel cochlear implant systems. This is achieved by using the concept of a `digital patient' whose artificial neuronal activation is compared to a healthy scenario in a computationally efficient manner to allow practical simulation times.|