The system, called REDMOD (Radiomics-based Early Detection MODel), analyzes fine textural patterns in pancreatic tissue, known as radiomics, that are not readily visible on conventional CT scans, and automatically outlines the pancreas, removing the need for specialists to trace its boundaries by hand. The researchers behind the study say the approach could help detect pancreatic ductal adenocarcinoma, the most common type of pancreatic cancer, at stage 0, when treatment is more likely to succeed.
The researchers examined abdominal CT scans from 219 patients treated at multiple hospitals. At the time of the scans, experienced radiologists had found no evidence of pancreatic cancer, yet every patient in the group was subsequently diagnosed with the disease. In 40% of cases, the images were obtained 3 to 12 months before diagnosis; in 35%, 12 to 24 months before; and in 25%, more than 24 months before, up to around three years. In nearly two-thirds (64%) of patients, the cancer developed in the head of the pancreas.
The images were compared with scans from 1,243 people who remained free of pancreatic cancer for up to three years afterward, matched by age, sex and scan dates. Patients who eventually developed cancer averaged 69 years old (range 34 to 88); those in the comparison group averaged 64, with the same age range.
REDMOD identified imaging signatures linked to the cancer an average of 475 days before clinical diagnosis, signs that had gone undetected during routine radiological assessments. It achieved a sensitivity of 73%, compared with 39% for experienced radiologists. On scans taken more than two years before diagnosis, REDMOD's detection rate was 68%, nearly three times the 23% recorded by radiologists. In an independent group of 539 patients, the system correctly classified just over 81% of scans as cancer-free, and 87.5% on a publicly available US National Institutes of Health dataset of 80 patients. When the researchers analyzed scans of the same patients taken months apart, the system produced matching results in 90 to 92% of cases.
"This temporal window holds profound significance, as attaining such early detection would substantially augment the probability of cure and improved survival," the researchers said. "In fact, modelling studies indicate that increasing the proportion of localized [pancreatic ductal carcinomas] from 10% to 50% would more than double survival rates, thereby underscoring that the timing of diagnosis is the single most critical determinant of survival outcomes."
The researchers caution that several questions remain before the system can enter routine care. The study population did not include a sufficiently diverse range of ethnic backgrounds, which could affect performance across different populations, and the model must undergo prospective testing, particularly in people at increased risk, including those with unexplained weight loss or recently diagnosed diabetes. "This study validates REDMOD as a fully automated AI framework capable of identifying the imaging signatures of stage 0 [pancreatic ductal adenocarcinoma] in normal pancreas, achieving this with substantial lead times and performance superior to expert radiologists," they concluded.