New AI Spots Brain Tumors Even When MRI Scans Are Missing

When doctors look for brain tumors using MRI, they usually take four different types of scans, each highlighting different tissue. But in busy hospitals, sometimes one or more scans are missing—maybe due to time, equipment, or patient movement. Until now, AI tools for tumor detection assumed all four were available, so they didn't work well with incomplete data. A new study presents an AI that can handle any combination of missing scans. It uses a technique called latent diffusion, which is like the AI imagining different plausible tumor shapes and then picking the most likely one. This makes it flexible and accurate, even with just one scan type. In tests on thousands of scans, it performed very well and was often the best among compared methods. The AI also provides a map of uncertainty, showing where it's less sure—a helpful signal for doctors.

For patients, this could mean that even if a scan is incomplete, the AI can still provide useful information, potentially speeding up diagnosis and reducing the need for repeat scans. It might also help in settings where advanced imaging is limited. The uncertainty map acts like a second opinion, highlighting areas that might need a closer look. While the AI isn't perfect, it represents a step toward more robust medical tools that work with real-world data, which is often messy. The researchers tested it on over 2,600 studies, and it achieved high accuracy, especially with a common scan type called FLAIR. This suggests it could be a valuable aid for radiologists, helping them focus on likely tumor regions and flagging uncertain areas for review.