A new AI model, REDMOD, can identify extremely subtle tissue changes linked to Pancreatic ductal adenocarcinoma—the most common and deadliest form of pancreatic cancer. According to research published in Gut, the model detects patterns that conventional imaging and even experienced radiologists often miss. As a result, it opens the possibility of diagnosing the disease at stage 0, when treatment is far more effective.
How REDMOD Enhances Imaging Precision
To overcome the limitations of standard CT scans, researchers developed the Radiomics-based Early Detection Model (REDMOD). It analyses subtle tissue texture patterns—known as radiomics—that remain invisible to the human eye. Moreover, the system uses automated pancreatic segmentation, which accurately outlines the pancreas without manual input, thereby improving consistency and reducing error.
Robust Study Design and Findings
As reported by medicalxpress, researchers tested REDMOD on CT scans from 219 patients who initially showed no signs of disease but were later diagnosed with pancreatic cancer. They compared these with scans from 1,243 matched individuals who remained cancer-free. Notably, REDMOD detected early “invisible” disease signals an average of 475 days before clinical diagnosis, offering a crucial window for intervention.
Outperforming Radiologists
Importantly, REDMOD demonstrated superior performance compared to expert radiologists. It achieved a sensitivity of 73%, nearly double that of radiologists at 39%. Furthermore, in cases identified more than two years before diagnosis, its accuracy reached 68%, compared to just 23% for human experts. It also reliably identified cancer-free scans in independent datasets, reinforcing its diagnostic strength.
Implications for Survival and Future Use
Early detection significantly improves survival outcomes. Researchers note that increasing early-stage diagnoses could more than double survival rates. However, they emphasise the need for further validation, especially in high-risk and diverse populations. Nevertheless, REDMOD represents a promising shift toward proactive, pre-clinical detection, offering new hope in tackling this highly aggressive disease.