Researchers at the Mayo Clinic, led by Panagiotis Korfiatis, PhD, have developed an artificial intelligence (AI) model that shows promise in detecting early-stage, “hidden” pancreatic cancer in asymptomatic individuals. This could pave the way for timely surgical interventions and potential cures. Here are the main findings:
- Study Methodology: The study utilized a dataset of 3014 CT scans, with 1105 diagnostic CT scans showing pancreatic ductal adenocarcinoma (PDA) and 1909 control CT scans. The AI model was trained on a subset of these scans and then tested on various other subsets, including prediagnostic CT scans taken before PDA diagnosis.
- Performance Metrics: The AI model demonstrated high accuracy in detecting PDA. In the intramural test subset, it correctly classified 88% of CT scans with PDA and 94% of control CT scans. The model’s sensitivity was high across different tumor stages, and its performance was consistent across various patient demographics, CT slice thicknesses, and vendors.
- Early Detection: The AI model was able to detect occult PDA on prediagnostic CT scans approximately 475 days before clinical diagnosis.
- Potential Applications: The AI model could address the challenges of imaging and diagnostic errors that often lead to delayed pancreatic cancer diagnoses. When combined with emerging blood-based biomarkers, this model could be used in screening trials for high-risk cohorts.
- Limitations: The study’s retrospective design might introduce selection bias. The results are preliminary, and further evaluation through prospective clinical trials is needed.
- Funding and Disclosures: The research received support from various institutions, including the National Cancer Institute of the National Institutes of Health. Some authors disclosed affiliations with healthcare and biotech companies.
In essence, the AI model developed by Mayo Clinic researchers offers a promising tool for early detection of pancreatic cancer, potentially leading to timely interventions and improved patient outcomes.