Skip to main content

LIDA Seminar Series 31st October 2019

LIDA Seminar
Thursday 31 October 2019, 12:30pm - 1:30pm
Room 8.43Y, Level 8, Worsley Building, University of Leeds, Clarendon Way, Leeds, LS2 9NL

This seminar will be held in the Room 8.43Y, Worsley Building, at 12.30 on Thursday 31st October.

Seminars are free and open for all to attend. No prior booking is required.

Each presentation will be followed by a short Q&A session.



Presentation: Combining empirical and mechanistic models for diagnosis and treatment planning of vascular disease

By: Dr Toni Lassila (University of Leeds)


Mathematical and statistical models have been proposed to assist the diagnosis and treatment of cardiovascular diseases. We identify two main dichotomies for such models: mechanistic vs. empirical and deterministic vs. stochastic. While different types of models have been successfully applied to study cardiovascular disease, they are often developed by different communities that do not interact. The aim of this talk is to provide examples where we have combined mechanistic and deterministic models with empirical and data-driven statistical models to obtain results that could not have been reached with either type of model alone.

Our first example concerns the minimally-invasive endovascular treatment of cerebral aneurysms by flow diversion. The question to answer is why in some cases the aneurysms rupture after treatment. To this end, we have developed mechanistic models for blood flow and thrombus formation that predict post-operative flow reduction, wall shear stress, and platelet content. These indicators have all been linked with post-treatment rupture. At the same time, we argue that the physiological flow variability at the internal carotid artery needs to be included to make the flow predictions more robust. We have developed data-driven boundary condition models based on a Gaussian process model applied to carotid ultrasound measurements. Using this model, we show that one commonly used rupture predictor, oscillatory shear index, is very sensitive to flow variability and that other more reliable indicators of rupture should be used.

Our second example concerns the diagnosis of dementia. Machine learning classifiers applied to neuropsychological test scores and brain MRI scans have shown promise in dementia classification. However, the classifiers based on brain imaging biomarkers do not perform particularly well in the early diagnosis of Alzheimer's disease (AD). Within the VPH-DARE@IT EU project, we have processed multi-modal data including cardiac and carotid ultrasound and 24-hour Holter blood pressures. The objective was to use systemic flow variability to provide additional diagnostic information, since AD has been linked with variability in the arterial pulse pressure. The data were used to calibrate lumped-parameter models for the systemic circulation and to predict patient-specific variability of carotid blood flow. We demonstrated that by adding the model-predicted flow and pressure information to regression-based dementia classifiers, we improved the classification accuracy in distinguishing mild cognitive impairment cases from healthy controls.