The AI software can effectively recognize the amount of fat surrounding the heart in MRI images, as well as assess a patient’s diabetes risk, according to the researchers.
A team lead by Queen Mary University of London scientists has created a novel artificial intelligence (AI) tool that can assess the amount of fat surrounding the heart from MRI scans.
Using the new technique, the researchers were able to show that having more fat around the heart is linked to a higher risk of diabetes, regardless of a person’s age, gender, or BMI.
The study was funded by the CAP-AI initiative, which is run by Barts Life Sciences, a research and innovation collaboration between Queen Mary University of London and Barts Health NHS Trust. The study was published in the journal Frontiers in Cardiovascular Medicine.
The way fat is distributed in the body has an impact on a person’s risk of acquiring certain diseases. Body mass index (BMI) is a frequently used metric that represents fat buildup under the skin rather than around the internal organs.
Fat buildup around the heart, specifically, has been related to a variety of diseases, including atrial fibrillation, diabetes, and coronary artery disease, and may be a predictor of heart disease.
“Unfortunately, manual measurement of the amount of fat around the heart is challenging and time-consuming. For this reason, to date, no-one has been able to investigate this thoroughly in studies of large groups of people,” said principal investigator Dr. Zahra Raisi-Estabragh from Queen Mary University of London.
“To address this problem, we’ve invented an AI tool that can be applied to standard heart MRI scans to obtain a measure of the fat around the heart automatically and quickly, in under three seconds. This tool can be used by future researchers to discover more about the links between the fat around the heart and disease risk, but also potentially in the future, as part of a patient’s standard care in hospital.”
The researchers put the AI system to the test using pictures from cardiac MRI scans of over 45,000 people, including participants in the UK Biobank, a collection of health data from over half a million people throughout the UK.
The researchers discovered that the AI program could properly detect the amount of fat surrounding the heart in those pictures, as well as calculate a patient’s diabetes risk.
“The AI tool also includes an in-built method for calculating uncertainty of its own results, so you could say it has an impressive ability to mark its own homework,” leader of technical development, Dr. Andrew Bard, stated.
Professor Steffen Petersen from Queen Mary University of London, one of the project’s supervisors weighed in stating that “the novel tool has high utility for future research and, if clinical utility is demonstrated, may be applied in clinical practice to improve patient care. This work highlights the value of cross-disciplinary collaborations in medical research, particularly within cardiovascular imaging.”
The study was published in the journal, Frontiers in Cardiovascular Medicine, on July 7th, 2021.
Results. Model performance remained high in the whole UKB Imaging cohort and in the external dataset, with medium–good quality segmentation in 94.3% (mean Dice score = 0.77) and 94.4% (mean Dice score = 0.78), respectively. There was high correlation between CMR and CCT PAT measures (Pearson r = 0.72, p-value 5.3 ×10−18). Larger CMR PAT area was associated with significantly greater odds of diabetes independent of age, sex, and body mass index.
Conclusions: We present a novel fully automated method for CMR PAT quantification with good model performance on independent and external datasets, high correlation with reference standard CCT PAT measurement, and expected clinical associations with diabetes.