Automatic assimilation of particle velocimetry data into computational blood flow models

As the worldwide prevalence of diabetes mellitus continues to increase, diabetic retinopathy (DR) remains the most common vascular complication in diabetic patients. Despite advances in treatment, DR remains a leading cause of visual loss in working-aged people worldwide. New approaches are necessary in order to better understand how to prevent vision loss from diabetic eye complications.

In this context, early DR detection is a promising approach to avoiding retinal damage and vision loss. Previous studies have found changes in blood flow in the diabetic eyes preceding the appearance of vascular lesions, which are currently the main clinical signs for diagnosis. Therefore, we hypothesise that DR early detection can be achieved via monitoring of early blood flow changes.

In a recent study, we proposed the first-ever non-invasive method for the assessment of blood flow in the parafoveal region of the retina (of paramount importance for central sharp vision). We validated our approach by comparing the blood velocity predicted by our computational flow models with in vivo blood velocity measurements obtained by tracking blood cell aggregations (BCA) visible in the retinal scans. Despite the accuracy in the model estimates, we could not achieve statistical significance in the comparison between the DR and control groups. We hypothesise that this is primarily caused by an important limitation in the definition of our flow models: the use of non-patient-specific flow boundary conditions, which we anticipate differing substantially in both groups.

In the current proposal we aim to address this model limitation by estimating the model boundary conditions via BCA velocity assimilation. This approach will allow us to define patient-specific models based on a small set of particle tracking experimental readouts. The proposed approach is based on constructing a constrained optimisation problem, whereby the model solution fields (fluid pressure and velocity in this case) are diagnosed by minimising a measure of the mismatch between the model output velocity and the cell tracking data.

As part of the project, we will develop a software tool that applies the previous numerical procedure to microvascular network flow models reconstructed from high resolution retinal images of the parafoveal region of the eye and in vivo blood velocity measurements obtained by BCA tracking. The former will rely on methodology previously published by the authors. Both retinal images and cell tracking data already exist at the laboratory of a project partner and were successfully employed in a previous publication.

This grant proposal is part of an international coordinated effort to develop a non-invasive technology for early diabetic retinopathy (DR) detection that can be integrated into the next generation of medical imaging devices. In terms of technological development pathway, our recent study (Lu et al. 2016) summarises our progress bringing this novel technology to technology readiness level (TRL) 1. We have identified an unmet clinical need (i.e. early DR detection), proposed a novel approach to address it (i.e. detection of haemodynamic changes preceding clinically observable structural changes), and demonstrated the technical feasibility of combining Adaptive Optics Scanning Laser Ophthalmoscopy (AOSLO) imaging and Computational Fluid Dynamics (CFD) modelling for this purpose. In the current proposal, we seek funding to undertake the first stages towards TRL 2, which will be achieved when we have delivered a proof of concept of haemodynamic-based DR detection based on retrospective data in a non-clinical environment. The first step in the TRL 2 process is the improvement of our CFD modelling technology to achieve patient specificity in our CFD boundary conditions as described in the Case for Support. At the end of the grant, we expect to have all the technological components in place necessary to design a proof-of-principle study based on existing retrospective data in a research environment, which will conclude TRL 2.

The two main stakeholder groups that we will be engaging with are healthcare professionals and industry. With regards to the former, this proposal will help to consolidate the PI's ongoing collaboration with Drs Jennifer Sun and Lloyd P. Aiello at the Joslin Diabetes Center, Harvard Medical School. Since 2014, we have collaborated in the development of the initial stages of the DR early diagnosis technology described in this proposal. We have budgeted into the current proposal a week-long visit to the Joslin Diabetes Center to engage with clinical colleagues and patients, and gather requirements about how the proposed technology could integrate into existing clinical work flows. With regards to industry engagement, this proposal seeks to consolidate an ongoing dialogue with the UK device manufacturing company Optos plc. Our project will benefit from early interactions with an industrial partner in order to detect early and to try to avoid, if possible, the most typical pitfalls on the route to commercialisation. Our long-term goal in this domain would be the integration of our technology into a prototype adaptive optics ophthalmoscopy device and to move towards an early proof of concept demonstration in a small scale clinical trial.

If the proposal is successful, the PI will engage with Edinburgh Research and Innovation (ERI), the research commercialisation arm of The University of Edinburgh, to discuss the most suitable approaches to managing the Intellectual Property arising from this project. ERI will carry out a preliminary market assessment of the proposed technology free of charge to the project. The purpose of this study will be to understand how the technology could be utilised in a clinical setting as part of a care pathway programme for patients. More precisely, we will answer the following questions: a) how would the new technology fit into a decision tree that affects patient care?, b) does it shorten or improve diagnostics or prognostic care?, and c) does it enable patients to be put on the right treatment path quicker?. Furthermore, the PI will undertake training in Health Economics through the National Institute for Health and Care Excellence.