Pioneering a New Frontier: Detecting early stage pancreatic cancer with AI

A new artificial intelligence (AI) tool has shown potential for detecting early-stage pancreatic cancer in asymptomatic individuals. Researchers from the Mayo Clinic Comprehensive Cancer Center have developed an AI model, using an extensive imaging dataset, for automated detection of pancreatic cancer on standard CT scans.

Pancreatic cancer is the 3rd leading cause of cancer death in Canada. It presents a grim prognosis, with ~70% of patients facing mortality within the first year of diagnosis. More than 60% of cases are only diagnosed when the cancer has progressed to an incurable stage, due to a lack of screening tests.

The new AI model addresses this diagnostic gap. The model utilized a diverse dataset (3,014 scans) of diagnostic CT scans with pancreatic ductal adenocarcinoma (PDA) and controls for training and testing. Of the testing dataset, the AI model correctly classified 88% of PDA scans and 94% of controls, with a mean accuracy of 0.92. Model performance remained reliable across patient demographics, tumour stage, CT slice thickness and scanning equipment.

Most importantly, the model identified PDA on pre-diagnostic CT scans a median of 475 days before clinical diagnosis, underscoring its value in early identification of occult cancer. The model is currently undergoing clinical validation and regulatory processes.

Written by Zaraa Malvat