Researchers have taken an important step toward data-driven precision cardiology by integrating cardiac imaging into an AI-powered knowledge graph. The new platform, called CardioKG, links detailed heart images to genetic data and drug information, significantly improving the ability to identify disease-related genes and potential treatment options.
The research was led by Dr. Khaled Rjoob and Professor Declan O’Regan from the Computational Cardiac Imaging Group at the MRC Laboratory of Medical Sciences. Their work addresses a long-standing limitation of biomedical knowledge graphs: while these systems excel at connecting genes, diseases and drugs, they traditionally lack information on how organs actually look and function at the individual patient level.
Integrating imaging features with data
To build CardioKG, the team analyzed cardiac imaging data from 4,280 participants in the UK Biobank with conditions such as atrial fibrillation, heart failure and myocardial infarction, alongside 5,304 healthy individuals. This resulted in more than 200,000 image-derived traits capturing variation in heart structure and function.
These imaging features were then integrated with data from 18 biological databases, covering genes, molecular pathways, diseases and drugs. Using AI, the model learned how imaging-derived phenotypes connect to genetic mechanisms and treatment opportunities. “Knowledge graphs already combine information on genes, diseases and drugs,” explains O’Regan. “By adding cardiac imaging, we dramatically improved how accurately we can identify new disease genes and potential therapies.”
New gene-disease associations
The enriched knowledge graph enabled the researchers to predict novel gene–disease associations and identify existing drugs that could be repurposed for cardiovascular conditions. Among the most notable findings: methotrexate, commonly used to treat rheumatoid arthritis, showed potential benefits for heart failure, while gliptins, diabetes medications, may be effective in atrial fibrillation.
The model also revealed an unexpected association between caffeine intake and a protective effect in patients with atrial fibrillation who experience rapid, irregular heart rhythms. While preliminary, these findings align with emerging evidence from other studies and highlight the value of combining real-world imaging data with AI-driven analysis.
Beyond the heart
CardioKG serves as a proof of concept for a broader class of imaging-enabled knowledge graphs. The same approach could be applied to brain scans, body composition imaging or other organ systems, opening new research avenues in areas such as dementia, metabolic disease and obesity.
For pharmaceutical companies, such models offer a powerful starting point for drug discovery. By rapidly prioritizing high-confidence genes and pathways, imaging-based knowledge graphs can help focus experimental validation and shorten development timelines.
Looking ahead, the researchers aim to evolve CardioKG into a dynamic, patient-centered platform. “Our next step is to capture disease trajectories over time,” says Rjoob. “That will allow us to predict not only which treatments may work, but also when disease is likely to develop, bringing us closer to truly personalized medicine.”
AI-driven cardocascular ultrasound
Last year, Philips introduced a new AI-driven platform for cardiovascular ultrasound that automates image analysis and accelerates clinical workflows. Integrated into the EPIQ CVx and Affiniti CVx systems and cleared by the US FDA, the platform reduces workload for clinicians while improving productivity and diagnostic consistency. AI automates key measurements, enabling faster and more reliable detection and monitoring of heart conditions such as heart failure, which affects millions worldwide.
Cardiovascular ultrasound is a non-invasive and widely used diagnostic tool, and embedding AI enhances its clinical value by standardizing image interpretation and shortening exam times. New data presented at ASE2024 also show that Philips’ AI algorithms can accurately detect regional wall motion abnormalities, an important indicator of cardiac events. Overall, the AI-enabled solutions support earlier diagnosis, improve reproducibility and help cardiologists and sonographers deliver more efficient, high-quality care.
