Deriving the internal bony structure of the cochlea from high-resolution µCT images for translation to low resolution image-based construction of person-specific computational models of cochlear implants

Abstract

To investigate cochlear implant (CI) performance, geometric computational models of the cochlea have been used to assess and optimise electrode insertion strategies and to investigate current flow through the cochlear volume as a result of intra-cochlear stimulation. Most of these models are derived low-resolution computed tomography (CT) and radiographic scans of humans or high-resolution histological sections of cochleae that are not viable for in vivo studies. Often these models lack a significant set of detail, still use a generic shape of the inner structures of the cochlea or obscured structures and are not clinically translatable. A method for the predication of obscured landmarks from reference landmarks is needed to generate user-specific computational models of the cochlea if the data source is of low quality. A standard set of prediction polynomial functions derived from high-resolution μCT scans needs to be developed and applied to clinically available CT images of the cochlea. Although histological sections of the human cochlea provide the best resolution of the cochlear structures, midmodialar sequential sectioning of the cochlea is not possible. μCT scans provide a solution, as the images are still of high quality and allow for detailed measurement of cochlear parameters on midmodiolar sections. Secondly, the more recent construction of a knowledge-based automated landmark computational model needs to be refined. The search fields that the automated models template uses to place a landmark need to be standardised and should have the ability to morph the cochlear shape together with the inner bony structures. Such models are of great clinical importance, as they can be generated much more quickly to inform CI surgeons on the individual cochlear anatomy of a CI patient and maintenance of CI. Lastly, the effect that taxonomic class has on the functional implications of an implanted electrode array has yet to be determined. The cochlear geometry that best predicts the location of the electrode array is important, as it has a significant implication for hearing outcomes. This thesis assesses the anatomical geometric factors that affect inter-person variations at the peripheral-electrode interface by developing a pre-operative approach to person-specific model design for implant candidates. This approach aims to increase the accuracy and details of geometric parameters that are available for model construction and integrate the image data into three-dimensional (3D) computational volume conduction models. The study used a landmark-based approach to measure the cochlear parameters that contribute to cochlear variation, as well as the development of algorithms to derive obscured landmarks from consistently available cochlear landmarks. A workflow in the form of a custom script UPCochlea.m that describes the technical aspects of landmark analysis was created to describe each cochlea algorithmically and to extract spiral trajectories that describe cochlear anatomy. Polynomial algorithms for the description of each spiral were created for use as standard for determining each cochlear class and the prediction of obscured spirals on clinically available data. This is the first study of its kind to describe all eight spirals that constitute the cochlea and spiral lamina. Automatic generation of user-specific landmark-based 3D computational models is a rapid process that can easily be translated into a clinical tool that may inform surgeons, manufacturers of CI’s and bio-engineers on the maintenance of such models. By refining the search fields for the template that landmark-based automated cochlear computational models search for a landmark to be placed, more accurate automated computational models could be generated. Psychometric data from CI users are correlated with the anatomical dimensions, their taxonomic classification and electrode locations derived from postoperative patient scans to determine the factors, if any, that may affect electrode array locations and thus the functional outcomes of CI users. The factors that contribute to speech and hearing outcomes may be used to optimise the parameter settings for CI user device programming