Development of an automated CT-brain analysis tool
LEAD SUPERVISOR: Professor Sarah Pendlebury, Nuffield Department of Clinical Neurosciences
Co-supervisor: Professor Mark Jenkinson, Nuffield Department of Clinical Neurosciences
Commercial partner: Brainomix, Oxford
CT-brain imaging is the standard brain imaging modality used in the NHS and globally and is cheaper and better tolerated than MRI particularly in older, frail, multimorbid patients in whom MRI may be contraindicated. We have shown that small vessel disease (SVD) and atrophy on routinely acquired brain imaging can predict delirium and dementia occurring up to 5-years later (Pendlebury et al, Age Ageing,2021; Pendlebury & Rothwell, Lancet Neurology,2019). Routinely acquired CT-brain scans therefore have clinical utility beyond the immediate indication. However, CT-brain data are not fully exploited in research or in clinical risk prediction because the lack of automated CT-tools to quantify SVD and atrophy necessitates time-consuming visual ratings subject to inter-rater variability. In the current project, the student would develop a reliable, user friendly and validated CT-brain analysis tool for quantifying SVD and atrophy building on our team’s established and widely used MR brain imaging tools. Such a tool would be transformational in unlocking the data on brain ageing and pathology contained in CT-brain scans with wide application to research and clinical practice.
To develop and validate a user-friendly research ready CT-brain analysis tool for quantification of WMC and atrophy using real-world clinical data.
- quantify global and regional brain atrophy relative to intracranial volume using an adaptation of our MRI tissue-type segmentation method;
- quantify SVD using adaptations of our MRI-based methods (BIANCA-Griffanti et al 2016);
- evaluation of tool performance using i) visual ratings ii) MRI data from paired CT/MRI scans, iii) manual segmentations iv) correlations with clinical variables.
To achieve the project objectives, the student will work on CT-brain scans from our previously assembled, large, well characterised cohorts of older patients with unplanned hospital admission including for stroke: OCS-Tablet/Recovery (n>850); Oxford Cognitive Co-morbidity, Frailty and Ageing Research Database (ORCHARD n>1000), Oxford Vascular Study (n>1,000). To create the CT-brain analysis tool, a range of machine learning methods (deep learning and classical methods) will be used as already developed for MRI brain (Sundaresan et al, OHBM 2018, 2019, MICCAI 2020) as well as promising new architectures (e.g. transformers). Use of data from different cohorts/scanners will ensure the CT-tool will be generalizable to a variety of settings.
Apply using course: DPhil in Clinical Neurosciences