| dc.description.abstract | The cochlear implant (CI) is a widely successful neural prosthetic device and is the preferred method of treatment for severe to profound hearing loss. CIs use an array of electrodes surgically implanted in the cochlea to directly stimulate the auditory nerve, inducing the sensation of hearing. However, patient outcomes remain highly variable, and there is a substantial fraction of individuals who experience poor speech recognition outcomes with CIs. Image-guided procedures for improving outcomes, such as better pre-operative planning or post-operative CI programming, are highly dependent on proper segmentation of the relevant anatomical structures. Frequently however, adequate training sets for a direct application of state-of-the-art deep learning techniques are difficult and time-consuming to obtain. The goal of this work is to look at these scenarios and use active shape models (ASM)s and other machine learning approaches for analyzing cochlear anatomy and other small shape libraries. This is accomplished in 4 aims: 1) Study relationships between anatomical landmarks and optimal CI insertion depth for pre-curved electrode arrays. 2) Improve automatic chorda tympani segmentation over the current existing method. 3) Validate the ASM for use in intra-cochlear anatomy segmentation. 4) Develop a novel extension to the ASM, called “multi-element ASM”, for use in small shape library situations. We found the round window to be a reliable landmark for optimal insertion depth, with the recommendation of a generic marker for alignment with this landmark placed at 2.28mm. Our proposed chorda segmentation method achieves sub-millimeter mean accuracy and outperforms the existing technique by 49%. We further found optimal parameters for an intra-cochlear ASM and found it to be viable for clinical use. Lastly, we found our multi-element ASM provided benefits to segmentation performance in a variety of scenarios and is a viable technique for small sample size applications. These results are beneficial in providing more accurate image-guided techniques and therefore outcomes for CI recipients. | |
