| Popis: |
The cochlea is the spiral-shaped organ of hearing within the inner ear which spatially separates sound waves based on frequency. Sensory hair cells distributed along the cochlear spiral convert the frequency-separated vibrations within the cochlea to electrical impulses for neural interpretation and sound perception. When the hair cells along the cochlea are not functioning and electrical impulses are not being generated, the nerves along the cochlear spiral can be directly stimulated using a cochlear implant to restore sound perception. A cochlear implant consists of an array of electrode contacts surgically implanted within the cochlea. Cochlear implants are programmed with a frequency map, which dictates how the frequency content of sound is distributed across the implanted electrode contacts for stimulation. Each individual cochlea has a unique spatial distribution of frequencies and post-operative electrode contact locations can vary, however cochlear implants have traditionally been programmed with a generalized frequency map. When a cochlear implant is programmed to stimulate with a generalized frequency map, individual anatomy and post-operative electrode contact locations are not taken into consideration which can result in reduced hearing outcomes. In the initial steps of this thesis, previously published techniques for individual cochlear modelling were validated using novel imaging data of cadaveric cochlea. A detailed analysis of cochlear micro-anatomy involved in spatial frequency distributions was then performed based on the identified limitations of current models. A patient-specific cochlear implant frequency mapping equation was developed which allows for the individualization of frequency mapping using angular measurements which can be obtained in clinical computed-tomography scans. A pilot clinical trial study was conducted to compare speech recognition performance when using the developed patient-specific frequency mapping tool and a default generalized approach. Preliminary benefit was observed in recipients using the patient-specific frequency map. The final steps of the thesis involved the development of a convolutional neural network for automatic cochlear segmentation in clinical imaging data in order to automate the measurements required for patient-specific frequency mapping. |
