A computer program that spots brain tumors on MRI scans
Original title: An Explainable and Comparative Transfer Learning Framework for Brain Tumor Classification from MRI Images
How far along is this research?
This is a preprint. Other scientists have not checked it yet, so treat it as an early signal rather than an answer.
This was only tested on a small set of stored scan images, not with patients.
The short version
Researchers tested computer programs that look at MRI scans and try to tell if a brain tumor is there.
What was studied. Researchers compared several computer programs that read MRI images. They used a public set of 253 MRI images, and they tested the programs on 38 images that were held back.
What they found. One program, called MobileNetV2, did best. It got 94.74% accuracy on the held back images. It was also small and fast, taking 5.9 ms per image. Some of the bigger programs failed in a simple way. They called every scan positive, which shows the risk of training on a small set of images.
What this means, and what it doesn't
What it could mean: This is early computer work. Tools like this may one day help doctors read scans faster and more evenly. The team shared their code so other researchers can check it.
What it doesn't mean: This is not a treatment. It does not change your care today. The program was only tested on a small set of stored images, not on patients in a real clinic. It is not approved for use anywhere, and it is not a cure.
Source: medRxiv (preprint), August 10, 2026 · Read the original
This plain-language summary was written by AI and published automatically after passing our automatic safety checks. How we write.