AI Model Detects Pancreatic Cancer Before Visible on CT Scans
An artificial intelligence (AI) model detected pancreatic cancer on routine CT scans more than a year before clinical diagnosis and substantially outperformed expert radiologists in identifying early cancers that were not yet visually apparent, according to a study published in BMJ’s Gut.
Researchers led by Mayo Clinic developed a fully automated radiomics-based early detection model (REDMOD) designed to identify subtle imaging signatures of pancreatic ductal adenocarcinoma (PDAC) before a tumor can be recognized by conventional CT interpretation.
The AI model identified 73% of pre-diagnostic cancers at a median of about 16 months before diagnosis, compared with 39% detected by expert radiologists reviewing the same scans. For scans obtained more than two years before diagnosis, REDMOD detected 68% of cancers, compared with 23% for radiologists.
The researchers said the findings suggest AI analysis could potentially identify imaging signatures of very early, or stage 0, pancreatic cancer that are not visible to radiologists, creating an opportunity to intervene before patients develop clinically apparent disease.
However, the retrospective study does not establish that using REDMOD in clinical practice would improve patient outcomes. The authors said prospective validation will be necessary to determine its clinical utility and noted limitations, including a lack of ethnic diversity in the study population.