An X-ray machine takes a scan, and the doctor sees only numbers — bone density, fat percentage. But the image holds far more. A team of researchers created a model called LeDXA that reads these scans like a map of hidden bodily signals.
The model was trained on 11,540 scans from the Human Phenotype Project with no labels at all. It didn't reconstruct the image; instead, it predicted its internal structure, like guessing an object's shape from its shadow. Then it was tested on 47,400 scans from the UK Biobank. LeDXA predicted future diseases more accurately than standard measurements: over 4.3 years of follow-up, the risk of type 2 diabetes and hip osteoarthritis was notably better assessed. For osteoarthritis, 66% of those who developed it fell into the top risk quartile according to the model, versus 41% using standard metrics.
The model also estimated biological age — with an error of just 2.9 years. People whose scan-based age was far ahead of their chronological age had a 45% higher risk of death. And in women who started hormone therapy, the gap narrowed. This suggests aging can not only be measured but also slowed.
Genes confirmed it too: the DNA regions linked to LeDXA's predictions are mostly already known as regulators of body composition. So a routine scan became a window into systemic health.
The LeDXA model was trained on 150,000 times fewer images than the famous DINOv3, yet it predicted diseases more accurately — saving both data and energy.
In Philip K. Dick's novel 'Minority Report,' crimes are prevented based on predictions from mutants — LeDXA does something similar for diseases, but without mutants, simply by reading old scans.