VRISTA – Voxel-based Retinal Imaging and Systemic-risk Translation with AI across UK and Japanese Cohorts

Retinal imaging offers a non-invasive window on the body's vascular and neural health. This fellowship will develop and validate a multimodal AI framework. At its core is voxel-based morphometry applied to three-dimensional OCT to map subtle, layer-specific structural deviations. These maps will be combined with features from colour fundus photographs and a biological ageing biomarker, the retinal age gap. The objective is early, interpretable prediction of cardiovascular and cerebrovascular risk with performance that is robust across devices, ethnicities, and clinical settings. We will adapt VBM, a well established technique in brain imaging, to OCT volumes to produce anatomically aligned deviation maps that quantify where the retina differs from expected patterns. These OCT features will be fused with macro-scale fundus cues within a deep learning model that estimates continuous, calibrated risk and time-to-event outcomes. To enable screening in settings without OCT, we will use knowledge distillation so that a lightweight fundus-only model inherits performance from the full multimodal system. Self-supervised pretraining and foundation-model backbones will support generalisability and data efficiency. The work leverages harmonised UK and Japanese cohorts totalling more than 200,000 images, enabling rigorous external validation and fairness analysis by sex, refractive status, and ethnicity. Transparency will be promoted through saliency visualisation, concept-based probes, and counterfactual testing. By delivering an accurate, explainable, and deployable tool that works with existing eye-care infrastructure, the project addresses European priorities on healthy ageing and prevention, strengthens EU-Japan collaboration, and lays the groundwork for scalable, equitable detection of silent vascular disease in ageing populations.