Generative design of hydrocephalus shunts

Thesis

The design of medical devices is often a lengthy and costly process, requiring multiple stages of testing through experimental, animal, and human trials. Therefore, computational methods are increasingly used early in the design process both to explore the mechanicalbiological interactions which affect device use, and to identify promising new design options in a process known as generative design.

This thesis focusses on improving the design of ventricular catheters used in the treatment of hydrocephalus, a relatively common but serious condition caused by excess cerebrospinal fluid (CSF) in the brain. Current catheters have high failure rates (≥ 40% [1]), often due to blockage by the Choroid Plexus (ChP), the nearby brain tissue that produces CSF. There is consequently an urgent need to improve the design to reduce this risk.

A 3D computational fluid–structure interaction (FSI) model was developed to assess how catheter designs influence ChP deformation. Designs used in the UK were tested and compared, with superior designs defined as those that caused less ChP deformation and hence reduced the likelihood of blockage.

As 3D simulations take days to run, faster, reduced-order models were developed using geometric simplifications and mechanistic approximations. Combined with a predictive machine learning layer, this created a surrogate model able to predict ChP deformation rapidly and accurately, enabling a large-scale optimisation over design space. Promising new catheter designs suggested by this surrogate-aided optimisation were tested in the full FSI model and showed marked improvements in performance when compared to current devices.

This interdisciplinary thesis brings together methods from mathematics and engineering, and applies them to a critical biomedical problem.

The results here focus on ventricular catheters, however the surrogateaided generative design framework is general, and has the potential to be used more widely. The modular codebases and methodologies developed in this work have all been deliberately designed to ensure they can be applied to other optimisation and design problems, both within the biomedical sphere and beyond.

Authors: Hayman, E

• DOI: 10.5287/ora-xq1yro5r5
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: fluid-structure interaction; medical device design; multi-physics design; hydrocephalus; surrogate-aided optimisation; computational optimisation; mechanistic learning
• Subjects: Mathematics; Engineering