AI Matches Doctors on Cancer Scans, But Close Calls Still Need a Human

Doctors use a scoring system called the Peritoneal Cancer Index (PCI) to decide how far cancer has spread in the abdomen. This number, from 0 to 39, helps determine whether a patient is a good candidate for aggressive surgery. But scoring requires carefully examining CT scans region by region — a slow, subjective process where even experts sometimes disagree.

A new study tested whether an AI model could do this scoring as well as humans. The AI was trained to outline 13 anatomical regions on CT scans and calculate the cancer score. Four expert radiologists also scored the same scans. The result: the AI matched human performance in most areas, and overall score differences were small — typically less than one point.

The catch? In certain regions — especially the small bowel — the AI strayed further from human judgment. More importantly, when a patient's score was near the 20-point threshold that determines whether surgery is recommended, even tiny errors could flip the decision. The researchers call these "borderline cases" the main place where expert review remains essential.

What does this mean for patients? In the near future, AI could pre-score scans and flag the easy cases, letting radiologists focus their energy on the ambiguous ones. That could speed up diagnosis and reduce burnout. But it also means AI won't replace the specialist — it will make them more efficient, while the final call stays human.