AI-Powered Liver MRI Scan Analysis Can Predict Cardiovascular Risk
A new, promising application of deep learning in medicine has been revealed by recent German research. It is forecasting the risk of cardiovascular disease based on liver MRI scans. Researchers led by Dr. Jakob Nikolas Kather, from the Medical Oncology Department of the National Center for Tumor Diseases at University Hospital Heidelberg, investigated the application of transformer neural networks to interpret liver MRI data.
Their study, published in JHEP Reports on April 22, demonstrated the potential of artificial intelligence in identifying individuals at risk for cardiovascular disease, particularly major adverse cardiac events (MACEs).
Using Liver MRI Data to Assess Cardiovascular Risk
Cardiovascular disease (CVD) often stems from underlying metabolic disorders, and the liver plays a pivotal role in regulating metabolism. Dr. Kather and his team believe that the liver can serve as an indicator for the metabolic changes that precede cardiovascular conditions. Early detection of these changes could lead to better, more personalized prevention strategies, especially for individuals who have not yet shown visible signs of disease. However, identifying these risks noninvasively has long been a challenge.
The researchers aimed to enhance existing biomarkers by developing an imaging-based solution that could predict cardiovascular risk more effectively. Their goal was to create a tool that could efficiently incorporate quantitative data from liver MRI scans to identify metabolic risk factors tied to heart disease.
Also read: Sonata Software Secures $73 Million Deal to Drive AI-Driven Digital Transformation
Transformer Neural Networks and the UK Biobank Data
To develop this model, the researchers used transformer neural networks—a powerful type of machine learning model known for its flexibility. The team trained the model using liver MRI data from the UK Biobank, a large-scale health database. Specifically, they used 44,672 single-slice liver MRIs, which were analyzed to assess cardiovascular risk. The data set included participants with a history of MACE, as well as those without, providing a robust foundation for training the model.
The model's predictive power was then tested by comparing the AI-derived risk scores with real cardiovascular outcomes. The findings were encouraging, indicating that the model had the ability to predict MACE and cardiovascular death with high accuracy.
Improved Predictions Compared to Existing Methods
One of the standout findings of the study was that the AI model outperformed traditional methods like SCORE2, a commonly used risk assessment tool for cardiovascular disease. The researchers also examined specific cardiovascular risk factors such as diabetes, cholesterol levels, systolic blood pressure, smoking status, and sex within the model’s predictions. This helped them understand which factors were more heavily emphasized by the AI system, providing valuable insights into its predictive capabilities.
Despite the promising results, the authors cautioned against using MRI as a widespread screening tool for cardiovascular risk due to its high cost and limited accessibility. Instead, they suggested that this technology would be most effective for high-risk groups or patients undergoing liver imaging for other reasons.
Challenges and Future Directions
There were limitations to the study. It was based on a retrospective assessment of single-slice liver MRIs, which can be a source of bias. Also mentioned was the absence of comprehensive liver metrics, including markers of mild liver disease. Future studies were encouraged by the researchers to incorporate more varied data sets, including whole liver imaging and clinical data, to enhance the model's accuracy.
Kather and colleagues concluded that, with continued refinement, their model might one day be used in clinical practice, providing a non-invasive means of determining cardiovascular risk based on liver health. As AI technologies advance, the promise of enhancing early diagnosis and tailoring prevention strategies in healthcare shines brighter than ever.
