| dc.description.abstract | Cochlear implants (CIs) use an array of electrodes implanted in the cochlea to directly stimulate the auditory nerve. After CI surgery, patients typically need several appointments with an audiologist over the following months to fine-tune the sound processor's programs for optimal hearing performance. However, few tools exist to assist audiologists in determining the best settings, leading to a time-consuming trial-and-error process that may result in suboptimal outcomes. To address this issue, computational models of the implanted cochlea have been proposed. In this dissertation, we present novel methods for patient-specific modeling of CIs. Firstly, we introduce two model-based applications: auditory nerve fiber health estimation and CI electrode sequence optimization. These proof-of-concept studies provide valuable insights into patient-specific information and have the potential to enhance the hearing outcomes of CI recipients in clinical settings. Secondly, leveraging convolutional neural networks (CNN), we propose novel loss functions and architectures to predict patient-specific electrical parameters and electrical potentials of an electrically evoked cochlea. These CNN models significantly accelerate the modeling process and reduce computational costs for CI users. Thirdly, we develop a transformer-based architecture capable of producing µCT-level tissue label maps of the inner ear using conventional CT scans. This advancement enables more precise modeling of CIs without the need for additional µCT scans. Lastly, we propose a robust and fully-automatic auditory nerve fiber segmentation approach using traditional image-processing techniques, addressing several limitations of a previously used semi-automatic method. In summary, our methods show impressive performance and have made improvements to our current computational models of CIs. | |
