Cardiac risk factors could be better detected earlier by radiomic analysis using machine learning, say researchers

A new position paper supported by the London Medical Imaging & AI Centre for Value Based Healthcare proposes the use of machine learning-enabled cardiovascular magnetic resonance (CMR) radiomics analysis. 

CMR radiomics is an emerging technique that extracts features from radiographic medical images using data-characterisation algorithms for deeper cardiac phenotyping. CMR may be used to detect early changes in the shape and tissue texture of the heart due to risk factors including cigarette smoking (past and current), hypertension, high cholesterol, and diabetes. 

Co-authored by BCN MedTech and Queen Mary’s University London among other five partners, the paper, published in Frontiers in Cardiovascular Medicine, evaluates five risk factor groups, using radiomics analysis to extract features from heart MRI images and uncover disease characteristics in the heart.

The research was based on over 5,000 scans, making it the largest and most comprehensive study to demonstrate the feasibility and performance of CMR radiomics for identifying cardiovascular risk factors to date.

Existing MRI methods use detailed scans and simple measurements of the hearts main pumping chambers but may not always detect early changes related to disease. Radiomics analysis of images is highly detailed and produces hundreds of numerical measurements of the heart.

AI Centre Professor of Cardiovascular Medicine and second author, Steffen Petersen, said the paper demonstrates radiomics analysis of heart magnetic resonance imaging (MRI) scans is better at identifying people with cardiac risk factors such as cigarette smoking, hypertension, high cholesterol, and diabetes, compared with the current methods of analysing scans.

"The work suggests that radiomics analysis will have applications on a wider range of heart conditions and may have potential to identify individuals at higher risk of serious heart problems, such as heart attacks” he said. 

In the future, following additional research, radiomics analysis could be used for early interventions such as preventative medication or closer monitoring.

Professor Petersen said the use of machine learning algorithms was key to making sense of such a large number of measures and building optimal clinical models.

“The collaboration on this paper was essential to address important questions for patients and clinicians using sophisticated machine learning approaches,” he said.