New AI Can Spot Lung Problems in Chest X-Rays With 99% Accuracy

Researchers in India have built a new AI system that reads chest X-rays — the most common and cheapest medical scan in the world — and flags signs of lung disease. The method first brightens the image so hidden details become visible, then converts it into self-repeating shapes called fractals (the same kind of patterns found in ferns and snowflakes) so it can tell healthy tissue apart from infected tissue. It reported 98.5% to 99.5% accuracy across large public image sets.

Why this matters: chest X-rays are cheap, fast, and available in almost every clinic, but they are genuinely hard to read. Infections hide behind ribs and overlapping tissue, images are often low contrast, and two radiologists can look at the same film and disagree. In places with few specialists, that means delays — and delays are what turn a treatable pneumonia into something serious.

The catch: the system was tested on benchmark datasets, which are curated collections of images, not in live hospitals. Lab accuracy rarely survives first contact with real-world messiness. The authors also stress that this is an unreviewed preprint, so it should not guide your own medical care. A flagged X-ray still needs a human doctor to confirm the finding.

What's next: this kind of tool is designed as a second pair of eyes, not a replacement — a first-pass filter that points a radiologist to the right spot. That could cut reading time and catch things tired eyes miss at the end of a long shift. If it holds up outside the lab, it could bring specialist-level screening to small clinics and rural hospitals that have no lung expert on staff.