New AI Helps Doctors Spot Brain Scan Mistakes in Babies
This could help save newborns' lives by making brain scans faster and cheaper.
Ultra-low-field MRI can make newborn brain imaging more accessible in low-resource settings, but it is prone to image artifacts. To support automated quality control, researchers tackled a challenge involving severity grading of seven common artifacts in ULF T2-weighted brain scans. They found no single backbone worked best for all artifacts, so they tested routing multiple complementary models through a per-artifact gate and distilling them into one compact in-domain model called LoFi RADIO. Both approaches improved performance, and the distilled model matched or beat the gated approach while needing only one model at inference.
- AI can now check newborn brain scans for mistakes, making it easier to use in poor or rural hospitals.
- The AI spots 7 common problems in scans, like blurry images, and grades how bad they are.
- It could help doctors diagnose brain issues in babies faster, even where big, expensive machines aren’t available.
Why It Matters
This could help doctors catch brain problems in newborns earlier, saving lives and making healthcare fairer worldwide.