Development of a machine learning method to automa4cally quan4fy calcified intracranial atheroma on CT imaging and assess risk of future neurovascular

Atheroma, or chronic pathological narrowing, of arteries is associated with an increased risk of sudden occlusion of these arteries and subsequent major diseases such as ischaemic stroke and heart attack. Atheroma formation is associated with vascular risk factors that are both modifiable (smoking, diabetes, high cholesterol, high blood pressure, physical inactivity) and non-modifiable (age, race, male sex, family history).1 However, identification of atheroma and subsequent vessel narrowing usually requires dedicated vascular imaging and is therefore often only acquired in patients already known or suspected to have arterial disease (e.g. following mini stroke or chest pain).

Since atheroma is often calcified, it is clearly identifiable on CT imaging, even if that CT was not acquired as an angiogram. Coronary artery calcium scoring is used clinically in this way to predict future risk of heart attack and to plan treatment strategies for patients before they have even a minor coronary event, i.e. to improve their modifiable risk factors. While some evidence suggests calcified intracranial atheroma might be similarly linked to major diseases of the brain such as cognitive impairment, dementia and stroke,2 large-scale population level evidence with longitudinal follow-up is lacking, and thus intracranial calcified atheroma scoring is not yet used in clinical practice.

Due to its speed, patient tolerability and availability, CT is the commonest method for imaging the brain. NHS Scotland performs approximately 120,000 CT brain scans annually for a host of acute (trauma, stroke) and non-acute (cognitive decline, chronic headache) indications. Public Health Scotland (PHS) oversees the collation of and access to pseudonymised national datasets of routinely-collected healthcare data for research in a secure environment.3 These rich datasets include >10 years of imaging linkable to data from primary and secondary care, prescribing, and national statistics and includes specific diagnoses using ICD (International Classification of Diseases) codes.

Aims
1. Develop an automated artificial intelligence method for scoring calcified intracranial atheroma on non-enhanced CT brain scans
2. Use this automated method on a large national imaging dataset linked to clinical data to describe associations between calcified atheroma and future neurovascular disease
3. Define a cranial artery calcification score for neurovascular disease prognosis that can be used in clinical practice

References
1. Banerjee C, Chimowitz MI. Stroke caused by atherosclerosis of the major intracranial arteries. Circulation Research. 2017;120:502-513
2. Chen Y-C, Wei X-E, Lu J, Qiao R-H, Shen X-F, Li Y-H. Correlation between intracranial arterial calcification and imaging of cerebral small vessel disease. Frontiers in Neurology. 2019;10
3. Gao C, McGilchrist M, Mumtaz S, Hall C, Anderson LA, Zurowski J, et al. A national network of safe havens: Scofsh perspective. J Med Internet Res.2022;24:e31684