Algorithm turns out to be able to detect small bleeding in the brain

The latest news about an algorithm capable of detecting small hemorrhages in the brain.

AI has so far been able to find medical conditions with a high degree of accuracy. However, to see bleeding in the brain is very challenging and slows things down. Even minor bleeding can be deadly.

UC Berkeley and UCSF researchers have created an algorithm that detects brain hemorrhages with greater accuracy than two out of four radiologists in a test. The key is very detailed algorithm training data.

According to UCSF, this process relies on a convolutional neural network examining more than 4,396 CT scans. That's a relatively small number of samples, but those abnormalities are broken down at the pixel level.

In other words, they are less likely to interpret noise and other errors as bleeding. This technique also has AI training on parts of the image at once, reducing the chances of making wrong assumptions based on very small changes.

Like other AI-based detection systems, it will not completely replace doctors. It only takes about a second to provide a report, and can automatically classify different types of bleeding.

This can save doctors valuable time in an emergency, and can ensure they catch hard-to-find bleeding that can be fatal in the worst case.

While scientists are still testing the algorithm against CT scans from trauma centers, there may come a time when it is used to quickly screen patients and help doctors focus on saving lives.