A computer tool learns to outline brain tumors with less hand labeling
Original title: Gradient-Based Self-supervised Multi-task Optimization for Weakly Supervised Brain Tumor Segmentation.
How far along is this research?
- Lab cells
- Animals
- Review
- Tested in people
This was done on cells in a lab, not in people. It is a very early step.
This was only tested on stored scan files on a computer, not with patients.
The short version
Researchers built software that traces brain tumors on scans with much less hand labeling.
What was studied. Scientists tested a new computer program that outlines tumors on brain scans. They ran it on two stored sets of scans, called BraTS2020 and BraTS2023.
What they found. The program picks on its own which practice tasks help it learn best. It also drops the practice tasks that hurt. On the BraTS2020 scans, it matched the whole tumor outline with a score of 90.7%. On the BraTS2023 scans, the score was 90.3%. The authors say this came close to programs trained with full hand labeling.
What this means, and what it doesn't
What it could mean: Outlining a tumor on a scan by hand is slow work. Tools like this may cut that work down. Over time, better tools could help doctors measure tumors more easily.
What it doesn't mean: This is not a treatment, and it is not a cure. It is software for reading scans, not a drug or a surgery. It was only tested on stored scan files, not with patients in a clinic. Nothing here changes your care today. This is early lab work, far from everyday use.
Source: PubMed, October 5, 2026 · Read the original
This plain-language summary was written by AI and published automatically after passing our automatic safety checks. How we write.