A weak magnet in an MRI machine is like a dim bulb in a dark room: you get a picture, but details drown in grain. These scanners operate at fields below 0.1 tesla (a typical hospital machine runs at 3 tesla, 30 times stronger), yet they're lightweight, inexpensive, and don't require a special room. The catch: noise eats away at quality.

The neural network stepped in as a restorer. Trained on pairs of noisy and clean images, it learned to strip away random noise without touching real tissue details. Tests on scanners with fields of 88 and 72 millitesla showed that after processing, images become as sharp as those from pricey 3-tesla machines.

But there's no such thing as a free lunch. If the original image is too dark (the signal is very weak), the network can't pull out what isn't there. And if it was trained on one type of image and then shown another (say, a different organ or a different scanner) it might make mistakes. That's why the researchers stress: you need to tune both hardware and software together, and before using it in hospitals, carefully verify that the cleaned images truly help doctors make diagnoses.