Machine learning helps identify “hidden fats” with daily bone scans

Researchers at Edith Cowan University (ECU) have developed an artificial intelligence algorithm that can detect potential vertebral fractures and estimate visceral fat levels using captured bone density scans, providing a quick, painless, and affordable detection method.

“Have you ever heard of despicable fat that hides deep in your stomach and is wrapped around your organs? It's visceral fat. It's a real troublemaker that is strongly linked to serious health issues such as heart disease, diabetes, and cancer,” says Ph.D. Student Arobamacsad.

“Obesity poses a serious threat to global health and is a major cause of morbidity and mortality worldwide. Beyond the cost of health, the economic burden is incredible and places a major strain on the health system and the national economy.

“In Australia, the economic costs in 2019 were $39 billion, and was projected to reach $228 billion by 2060. This is 3.5% of Australia's gross domestic product, and it's not just money. It's estimated that there will be 3.7 million obesity-related deaths each year.”

Dangerous visceral fat wrapped in organs is currently estimated using methods such as body mass index, waist circumference, waist to hip ratios. However, Maqsood noted that these measures are limited and do not distinguish between different types of body fat.

“This simplification contributes to the contradictions in assessing obesity and its complications, highlighting the need for a more accurate approach to measuring obesity,” she added.

“Imaging techniques such as MRI and CT scans can accurately measure visceral fat, but their large prices are often a limiting factor. CT also exposes patients to higher levels of radiation.”

Identify vertebral fractures using transverse spine dual energy x-ray absorption measurement (DXA) scans. These images can be reused for opportunistic screening of visceral fat, providing new and valuable health insights without the need for additional testing.

ECU trains machine learning algorithms on these scans to accurately predict the amount of visceral fat present in a person simply by looking at a transverse spine DXA scan.

“The machine learning model is trained on thousands of images. The next step is to incorporate additional datasets from around the world, so you can learn from the largest and most diverse cohorts possible and be as effective as possible.”

Maqsood will present her research at the international conference on Medical Imaging Computing and Computer-Assisted Interventions (Miccai 2025) held in Korea from September 23-27.