REVAMMAD · Retinal Vascular Modeling, Measurement and Diagnosis""

Retinal Vascular Modeling, Measurement and Diagnosis

The REtinal VAscular Modeling, Measurement And Diagnosis (REVAMMAD) project addressed the problem of diagnosing some of the most globally significant chronic medical conditions and vascular diseases including diabetes, hypertension, dementia and stroke, by focusing on the detailed analyses of changes caused to the blood vessels by utilising images taken from the retina. The project engaged a wide range of expertise, from mathematical modellers and clinicians to computer vision experts, who collaboratively worked across countries and across the boundaries of individual disciplines to train a new generation of interdisciplinary scientists. As a result of their participation in the REVAMMAD project, our Early Stage Researchers (ESRs) have been given the opportunities to gain new skills which have enabled them to effectively translate the latest vascular modelling theory and computerised image analysis techniques into effective interventions. Our researchers investigated changes to the walls of the blood vessels of the retina caused by vascular disease, engaging with patients who have pre-diabetic retinopathy to track the changes caused to the retina vessels, effectively mapping the journey of progression to retinopathy through studying these vessels. This gave our researchers the opportunity to develop a range of new techniques which enabled them to both accurately detect the retinal vasculature, and to extract precise measurements from it. New techniques in mathematical modelling were developed and validated by our researchers, including the modelling of myogenic response gradients throughout the retinal vessel network, modelling how vessel walls respond to changes in blood flow under stimuli, and how whole vessel networks behave. These advances have been synergistically combined with our advancements in computer vision techniques which themselves have shed new light on how we are able to detect the individual parts of the retina, generating significant new data. This bringing together of disciplines has enabled us to utilise the overlap of mathematical modelling and computer vision in the study of flow dynamics within retinal blood vessels, which in turn increases our understanding of how these vessels are actually connected. We have also developed “risk models” to optimise the screening intervals for patients with diabetic retinopathy, involving input from clinicians in determining the role of computerised analysis of the retina in support of this advancement. A fully automated system to estimate the degree of narrowing, bulging or tortuosity (‘curving’) of blood vessels was developed, both in terms of modelling the curvilinear structures and the learning context filters. Using a novel approach to accelerated convolutional sparse coding filter learning, this new methodology is expected to make filter learning much more discriminative and will result in a more robust segmentation module. The tortuosity plane has been proposed as a better tool to quantify and interpret tortuosity. The curvature enhancement method has allowed for a new set of tortuosity metrics to be created. The progression from diabetes to diabetic retinopathy is associated with changes in retinal haemodynamics (i.e. the dynamics of blood flow as it adjusts to the conditions within the body and its environment). We analysed longitudinal studies of 24 subjects for the three years preceding the appearance of diabetic retinopathy, estimated fluidynamic parameters using a simple haemodynamic model, and established statistically significant changes in some estimated haemodynamic parameters associated with the development of diabetic retinopathy. We developed a new algorithm for the automatic detection of optic nerve hypoplasia (ONH), a congenital optical nerve anomaly resulting in the underdevelopment of the optic nerve. Our algorithm was demonstrated to be the fastest and second best performing in the world compared to current literature, when tested on the MESSIDOR dataset - a collection of diabetic retinopathy examinations, each consisting of two macula-centred eye fundus images (one per eye). Qualitative results are shown to improve with an elliptical approximation of the ONH boundary; this was enabled by modelling the photograph distortion caused by the quasi-spherical shape of the eye. We also developed probabilistic models which relate the location of retinal lesions to clinical risk, using a prospective database of 60 patients recorded over 10 years, with 900 images. Special analysis tools were developed in the project which have shown a strong relationship between the area where micro aneurysms appear, and the development of disease. Vascular disease and chronic medical conditions such as Alzheimer’s, diabetes, stroke and coronary heart disease account for an increasingly large proportion of healthcare costs across European member states. As a result of the research carried out in the REVAMMAD project, the new and improved methodologies and algorithms will feed into the future development of computerised screening systems to enable earlier diagnosis of disease, and improvements to both treatment and ultimately outcomes, ultimately benefitting both the economy and society as a whole. The project generated a significant number of software tools and datasets which can house relevant data, including images and clinical assessments to be deposited by clinicians and scientists working in the field, together with software components for mathematical modelling and computerised imagery analysis. In addition to making these software components publicly available to support the advancement of research in this field, the projects partners are looking at other ways to commercially exploit this software. The programme was most notably successful in achieving its core objective to provide a high quality interdisciplinary training programme. As a result of their participation in the project, the researchers developed into a strong cohort with a good, shared ethos and deep levels of collaboration and co-operation across the project. The core of the programme was the series of five extensive workshops held during the project: two at the University of Lincoln; one at FORTH in Crete; one at Charite in Berlin, and one at Padova in Italy. Each of the workshops combined taught activities, some of which were introduced by leading international experts in the field from outside the consortium, and additional skills such as media training (for outreach activities, and career preparation), research presentations and identification of research goals. These workshops were complemented by an extensive series of secondments, participated in by all of the researchers (most of whom experienced more than one secondment). These secondments enabled knowledge and expertise transfer and facilitated the development of collaboration between laboratories. The cohort’s connectivity was reinforced by regular events such as e-seminars and discussions and by other joint dissemination and information-sharing tools, such as blogs, which were organised and operated by the ESRs themselves. Our ESRs, having benefited from the intense training and supervision in REVAMMAD, have gone on to establish scientific and academic careers as lecturers and researchers and in scientific companies. Almost all registered for PhDs, either having already completed, or are approaching completion, of their studies. Websites: Project Website: https://revammad.blogs.lincoln.ac.uk/ Project Outreach blog (maintained by ESRs): https://revammad.wordpress.com/ Contact details: REVAMMAD Coordinator: Professor Andrew Hunter, Deputy Vice Chancellor Research and Innovation, University of Lincoln REVAMMAD Project Manager: Pilar Pousada Solino, University of Lincoln (ppousadasolino@lincoln.ac.uk) 01522 835087

Data: CORDIS, © European Union