Data-driven Computing

The prevalence of myopia is increasing extensively worldwide. The number of people with myopia in 2020 is predicted to be 2.6 billion globally, which is expected to rise up to 4.9 billion (approx. 50% of the global population) by 2050. The number of individuals with high myopia is also increasing substantially and pathological myopia is predicted to become the most common cause of irreversible vision impairment and blindness worldwide. Pathological myopia affects 5-11% of the myopic population and leads to irreversible blindness in an individual’s working lifetime. Our group works on the vascular analysis in retinal images to enable personalised and early myopic degeneration/pathological myopia risk prediction. Using retinal images for the early detection of risk of pathological myopia can significantly reduce the chances of irreversible vision loss caused by pathological myopia and reduce the financial burden on the healthcare to treat the consequences of pathological myopia such as retinal detachments and choroidal neovascularisation. The retinal image data are combined with clinical measurements of refractive error to develop a comprehensive AI based model that evaluates the risk of irreversible vision loss due to pathological myopia. 

Evidence suggests cardiovascular disease (CVD) screening improves medical therapy compliance which can reduce cardiovascular events (CVEs). CVD prevalence, such as carotid artery disease, seems high in general populations but related CVE rate is <2%. I.e although present it causes few strokes/heart attackes etc. Compared with general populations, screening high-risk groups may be beneficial. Patients with Abdominal Aortic Aneurysm (AAA) are considered high risk as they have increased CVD prevalence, such as cardotid artery disease, where the CVE rate is significantly higher (9-17%). We know that carotid artery disease is highly predictive of CVE. Targeting screening focused on high-risk populations, rather than the general population, has been widely reported to be more cost-effective due to the much higher prevalence of CVD detected. Clearly, if screening was undertaken to identify carotid artery disease with the intention of reducing the subsequent risk of all CVE, it is likely that population screening of at least high-risk groups could be justified. CT/MRI are expensive, involve radiation or nephrotoxic contrast and are not suitable for mass screening. Ultrasound is cheaper but not cost-effective as it requires skilled operators, of which there is a shortage. For successful ultrasound-based screening, there must be technological developments that improve value and suitability. Tomographic 3D ultrasound (tUS) captures arterial reconstructions via a simple scanning method taught in 20-minutes. Computational Fluid Dynamic (CFD) and data science modelling could personalize image-based prediction and may be an important predictor of CVE risk. CFD itself is computationally slow, resource-intensive and cannot be directly deployed for usage by clinicians. However, coupling CFD with tUS and machine learning (ML) would enable the introduction of personalized screening that is good value-for-money.