New AI Helps Doctors Mark Bone Scans Three Times Faster

Ultrasound uses sound waves to see inside the body. When doctors need to guide a needle or check a bone, they often have to trace the edges of the bone by hand on the screen. This is slow, and experts can spend a minute or more on a single image. A new research system called ExiL makes this job much faster by learning from the doctors who do it.

ExiL starts with a rough, imperfect outline and lets the expert fix it, just like doctors naturally do. But instead of throwing that correction away, the AI remembers it and gets better. Over time, it matches that expert's style. It also runs in real time — processing each image in 10 to 50 milliseconds, or about 20 times faster than the blink of an eye.

In tests with bone ultrasound data, ExiL cut average annotation time from 60 seconds to 20 seconds per image — a 66% reduction. Accuracy remained strong: its best system scored 0.87 on a standard 0-to-1 overlap scale (1 is perfect) and kept the outline within about 3 pixels of the true edge. It also improved accuracy by roughly 0.045 compared to older methods that don't learn progressively.

The true benefit is practical. If you are a busy radiologist or orthopedic surgeon, this means less eye strain and more time with patients. If you are a patient, faster and more consistent ultrasound analysis could mean quicker diagnosis and safer procedures. The catch? The method was tested on cadaver and volunteer data, so real-world clinical validation is still needed before it appears in hospitals.