Approved Research

Multimodal Integration for Deep Learning-Enabled Screening for Cardiovascular Diseases

Lay summary

Cardiovascular diseases such as myocardial infarction and stroke are a major cause for illness and death. The early identification of individuals at risk of developing cardiovascular diseases is crucial as it allows for an early lifestyle and medical intervention, so that irreversible damage can be averted. Yet, existing screening tools often are not reliable enough or are too expensive to be used for assessing large parts of the population. To overcome this challenge, we aim to use cutting-edge technologies and develop artificial intelligence-based screening models to identify individuals at risk. We will especially focus on cardiovascular diseases such as heart failure, atrial fibrillation, arterial hypertension, and stroke. We will build on the thorough assessment of the participants of the UK Biobank and utilize imaging data, e.g. from magnetic resonance imaging, laboratory markers, electrocardiograms (ECGs), and other examinations for the development of our artificial intelligence models. We expect that the integration of results from multiple diagnostic methods into one model will lead to an increase in its performance and that the developed models will outperform existing screening tools for cardiovascular diseases. This study will contribute to the implementation of artificial intelligence into clinical practice, as it could lay the ground for building artificial intelligence-based screening tools that are cheap, easy to use, and show a good accuracy in identifying individuals at risk in the general population. We expect the project's duration to be three years.

Scope extension:

(1): Exploration of the benefits of multi modal integration of ECGs, cardiac imaging (Echo & MRI), and other biomarkers in deep learning models

(2): Creation of AI-based risk-scores for the development of cardiovascular diseases (heart failure, atrial fibrillation, coronary artery disease, stroke, vascular dementia, chronic kidney disease) and screening tools

(3): Validation of these novel scores and screening tools in independent cohorts.

(4): Comparison of newly created AI-based risk scores and AI-based screening tools with published risk and screening scores

(5): Explore possibilities to use available genomic, proteomic, and metabolomic data to enhance fidelity of developed AI risk scores