The advent of non-invasive imaging techniques such as Cornputeci Tomography (CT) haS revolutionised modern diagnostic medicine, resulting in a massive increase in the data available to clinicians. This data needs to be analysed and interpreted, requiring the development of tools and techniques specific to the application. This thesis develops tracking, modelling and registration techniques for anatomical shape analysis applied to a case study, the human cochlea. These algorithms are validated against generated data (with added noise), and applied to clinical CT image data, which is then validated by an surgeon.
The tubular extraction algorithm tracks along the centreline of the cochlea, revealing the characteristic path and providing a unique perspective on the cross-section. An analysis of the curvature of the cochlea shows its outline can be described by a logarithmic spiral model. Extending this model into 3D, a parametric shape model of the cochlea is developed, with model parameters derived from clinical descriptions. The Parametric Model-Image Registration method is applied to fit this shape model to clinical data, resulting in metrics that are consistent with other published figures. This work shows how application-specific models can give insights into anatomy and have explanatory power beyond that of general models.