Glioblastoma, a brain tumor, is a tricky foe: in some patients, the drug temozolomide works, while in others it barely helps. The difference often lies in a small stretch of DNA — the MGMT gene. If this region is chemically switched off (methylated), the tumor is worse at repairing its cells, and the drug hits harder. Previously, to find out the gene's status, a surgeon would take a piece of the tumor with a needle — a risky procedure, and the result could be inaccurate, since the tumor is as patchy as a quilt.
This new approach looks at the tumor from the outside — through routine MRI scans. The computer extracts hundreds of numerical features from four types of images: how bright the tumor is, what its shape is, how it takes up contrast. Another algorithm learns from these images on its own, without guidance, like a child looking at pictures and finding patterns. Both streams of information are combined, and the model gives a prediction: whether the MGMT gene is methylated or not.
Such a "transparent" artificial intelligence helps doctors trust the prediction. The method was tested on scans from hundreds of patients in two independent databases, and it consistently predicted MGMT status. This is a step toward choosing treatment without opening the skull — based on images alone.
The MGMT gene encodes a protein that repairs DNA in tumor cells. If the gene is methylated, there's little protein, and tumor cells die from chemotherapy, like a house with no repair crew.