When a woman is treated for uterine cancer, the make-up of her tumour shapes the treatment plan, and increasingly that means asking a pathologist to run DNA analysis. One marker matters especially: a mutation in the POLE gene. Patients who carry it have an excellent prognosis and usually need no radiotherapy or chemotherapy after surgery. “For patients with a POLE mutation, that’s good news,” says project lead prof. Tjalling Bosse. “Which is exactly why it matters that we test for it and identify every one of them.”
The problem is that only a DNA test can confirm a POLE mutation, and that test demands capacity and expertise not available everywhere. As a result, in the Netherlands and abroad, it isn’t run on every patient, and some miss information that could directly change their treatment.
Researchers at LUMC think AI can close that gap. Their model, POLARIX, developed by the AIRMEC research team, reads digital images of tissue and predicts which patients are likely to carry a POLE mutation. Only those patients then go on for a confirming DNA test. The approach could cut the number of DNA tests needed sharply without weakening the diagnosis.
The team has now received more than €1 million from KWF Kankerbestrijding to find out whether POLARIX is safe and reliable enough to use as a pre-selection step in everyday care. It will be tested across several hospitals in the Netherlands and beyond, with the researchers checking not only the accuracy of its predictions but whether it holds up across different patient groups and clinical settings. The work fits LUMC’s Digitaal Dichtbij programme.
If POLARIX proves out, roughly three-quarters of current DNA tests could be avoided, with only patients flagged as higher-risk going on for confirmation. The implications reach past the Netherlands: in countries where DNA diagnostics are scarce, the model could help direct limited capacity where it counts. Through its LUMC Global platform, the hospital recently invested in rolling POLARIX out in South Africa.
Beyond uterine cancer, the project offers a template for putting AI into pathology responsibly. As Bosse puts it: “A biomarker only has real value if every patient who could benefit from it can actually access it.”
