AI deep learning of multimodal imaging for cardiometabolic disease risk assessment

Cardiometabolic diseases remain the leading cause of global morbidity and mortality. Early identification of cardiometabolic risks is pivotal for effective prevention and intervention. Evidence has shown that scalable imaging modalities provide a window into vascular and cardiac health but also encode subtle features of systemic metabolic burden, offering transformative potential for early risk identification. For instance, carotid ultrasound features have been associated with atherosclerotic progression; retinal photography captures microvascular changes correlate with hypertension and diabetes; and ECG waveforms capture cardiac changes linked to sudden cardiac death. 

Artificial intelligence (AI) has demonstrated unparalleled capacity to detect hidden patterns in images. Deep learning models can extract complex features directly from raw inputs, enabling the discovery of novel imaging phenotypes beyond human recognition. Integrating deep learning across accessible imaging modalities could thus yield a comprehensive and scalable cardiometabolic risk profiling tool, especially in settings where access to advanced laboratory diagnostics is limited. 

In the China Kadoorie Biobank of 0.5 M adults, through repeated resurveys of around 20,000 participants, carotid ultrasound images and ECG were collected at two time points (2013-14 and 2020-21), with retinal images of both eyes at 2020-2021. In addition, a wide range of lifestyle, physical characteristic, blood biochemistry and multi-omics data (WGS, proteomics, metabolomics and metagenomics) are available, plus fatal and non-fatal health outcome data collected from electronic health records. 

The project will be based within the CKB group in the Big Data Institute, which hosts a world-class community of population health and data science researchers. The student will be supported by a multi-disciplinary team of healthcare statistician (Wright), machine learning scientist (Zhang) and epidemiologist (Du), and have access to CKB data resources and deep learning computing facilities. 

By the end of the DPhil, the student will be competent in deep learning for large-scale clinical datasets, and develop towards a leadership at the intersection of AI, big data and epidemiology.