Pancreatic cancer remains the deadliest form of cancer globally, largely due to late diagnosis. Symptoms are often vague and easily mistaken for other conditions, and tumors are difficult to detect in early stages using standard abdominal CT scans. By the time the disease is identified, curative treatment is typically no longer an option. Only about 10% of patients survive five years after diagnosis. To address this challenge, AI researcher Henkjan Huisman and radiologist John Hermans developed a reliable benchmark to evaluate artificial intelligence tools for detecting pancreatic cancer. They created a confidential dataset comprising CT scans from nearly 400 patients across Western countries, assessed by a large panel of international imaging experts. This dataset was then used to invite AI developers worldwide to submit models capable of identifying pancreatic tumors. More than 250 AI models were submitted, evaluated, and compared against the performance of expert radiologists. When tested on the confidential dataset, the top-performing AI systems outperformed the average radiologist. The best AI models reduced false positives by 38% compared to the expert group and correctly identified cancer in 92% of scans—compared to 88% accuracy among radiologists. These findings suggest that AI can serve as a valuable tool to support radiologists, potentially improving diagnostic accuracy and reducing workload. However, the researchers emphasize that the AI systems are not yet ready for clinical use and require further validation before being deployed in patient care. Lead researcher Henkjan Huisman highlights the significance of the benchmark: “Because we’ve established a reliable standard, we now know that AI systems surpassing clinicians are genuinely effective.” The study’s results were published in The Lancet Oncology. There is also promising potential for earlier diagnosis. Radiologist John Hermans notes, “Our findings indicate that the best AI model could help detect pancreatic cancer sooner, leading to faster treatment—a crucial step forward for a disease where early intervention could dramatically improve outcomes.” Still, Hermans cautions against premature implementation: “We must avoid false alarms, as they cause unnecessary stress for patients and strain healthcare systems.” Moving forward, the team is training enhanced AI models using broader abdominal scans to improve detection across a wider range of cases.