Scientific Rationale:
The “Heart-Brain Axis” represents a complex physiological interplay where cardiovascular dysfunction significantly increases the risk of neurodegenerative diseases and stroke. However, traditional statistical methods often fail to capture the high-dimensional, non-linear interactions between cardiac and cerebral systems. UK Biobank provides an unprecedented resource of paired cardiac and brain MRI, genetics, and longitudinal clinical outcomes. We propose to apply state-of-the-art artificial intelligence and deep learning techniques to this multimodal dataset to uncover novel biomarkers and mechanistic pathways linking heart and brain health.
Research Questions:
1. Can deep learning algorithms extract novel, subclinical imaging phenotypes from cardiac and brain MRI that are superior to conventional metrics in predicting comorbidity risk?
2. How do high-dimensional genetic and metabolic profiles interact with imaging markers to drive the progression of heart-brain comorbidities?
3. Can multimodal deep learning models improve the early prediction of dementia and stroke in patients with cardiovascular disease compared to standard clinical risk scores?
Objectives:
We will develop an end-to-end multimodal AI framework for the heart-brain axis by (i) building and validating computer-vision pipelines for automated segmentation and feature extraction from cardiac and brain MRI, (ii) constructing integrative deep learning models that fuse imaging with genomics and electronic health records to uncover shared, non-linear risk factors linking cardiovascular and neurological conditions, and (iii) applying unsupervised, data-driven clustering to identify distinct patient subgroups with characteristic patterns of heart-brain dysfunction, enabling more precise risk stratification and targeted prevention.