Background
Stroke is a leading cause of death and disability globally. Providing better information on the risks of health outcomes like disability or impaired cognitive function after a stroke is a priority for stroke researchers and people affected by stroke.
Novelty and Importance
All stroke patients admitted to hospital in the UK will undergo brain imaging scans soon after admission to inform diagnosis and treatment. Features found on brain scans have been associated with worsened stroke outcome; in particular tissue hypoattenuation, large lesion size, swelling, hyper-attenuated artery, atrophy and leukoariosis. However, these features represent only a tiny fraction of the wealth of data captured by routine imaging. Artificial Intelligence (AI) based image analysis methods have the potential to phenotype both the stroke and the overall health of a patient’s brain in tremendous detail and so provide invaluable information to better understand the risk of health outcomes following stroke. These models could then be implemented into clinical care alongside the broader patient data to optimise risk prognostication and inform treatment decision making.
Aims and objectives
1. Create a linked database including routinely collected brain imaging data from two South London hospitals linked to structured data from the South London Stroke Register of over >8000 patients
2. Use the created database to develop prediction models (eg Bayesian Inference Models, Generative Adversarial Networks, CNNs) for complications and outcomes after acute stroke (mortality, stroke recurrence, disability, cognitive impairment)
3. Investigate the utility of adding AI based imaging analysis to conventional statistical modelling

