AI Spots How Aggressive Kidney Cancer Is Without a Biopsy

When doctors find a tumor on a kidney, one key question decides the treatment: how aggressive is it? Today the answer usually comes from a biopsy — a needle inserted into the kidney to grab tissue, which is uncomfortable and can miss the most aggressive parts of a tumor. A team of US researchers, led by Dartmouth and Memorial Sloan Kettering, asked whether a plain CT scan could answer that question instead.

Their AI, called RCC-Align, learns by studying matched pairs: a microscope slide of a patient's tumor tissue placed beside that same patient's CT scan. During training, the software figures out which blurry patterns on a scan tend to go with aggressive-looking tissue under the microscope. Once trained, it only needs the CT scan — no needle, no lab work.

How well does it work? The team uses a score called AUC, where 0.5 is a coin flip and 1.0 is perfect. RCC-Align scored about 0.60, beating an AI that looked at CT scans alone (about 0.55). But an AI reading actual biopsy tissue scored 0.72 — noticeably better. So the new method is a genuine step forward, just not a replacement for a biopsy.

The realistic promise is narrower and still valuable: helping when a biopsy is unsafe, impossible, or unreliable because a tumor is patchy. The researchers are clear that the tool needs testing on larger, more varied groups of patients from multiple hospitals before any doctor could use it. For now, it is promising research — not a diagnosis you can ask for at your next scan.