A new AI-model is showing promising potential to detect pancreatic cancer at its earliest, often undetectable stage. Researchers report that the Radiomics-based Early Detection Model (REDMOD) can identify subtle tissue changes linked to pancreatic ductal adenocarcinoma (PDAC), the most common and deadliest form of pancreatic cancer, long before conventional imaging or human interpretation can do so.
The findings, published in Gut, suggest that this technology could fundamentally shift how pancreatic cancer is diagnosed. Instead of being detected at an advanced, often terminal stage, REDMOD opens the possibility of identifying the disease at stage 0, when treatment options and survival chances are significantly better.
Detecting the ‘invisible’
Pancreatic cancer is notoriously difficult to diagnose early. In its initial stages, patients typically show no symptoms, and standard CT scans fail to reveal clear abnormalities. As a result, most diagnoses occur late, contributing to poor survival rates.
REDMOD addresses this challenge by analyzing radiomics, complex patterns in medical imaging data that are invisible to the human eye. The AI framework automatically segments the pancreas from surrounding tissues and detects subtle textural changes that may indicate early malignancy. This automation also reduces variability associated with manual image interpretation.
By focusing on these micro-level imaging signatures, REDMOD can identify what researchers describe as “pre-clinical” disease, changes that precede visible tumors or symptoms.
Strong performance
To validate the model, researchers applied REDMOD to CT scans from 219 patients across several hospitals. At the time of scanning, none of these patients showed signs of pancreatic cancer according to radiologists. However, all were diagnosed with the disease later, ranging from three months to more than two years after the scans were taken.
The results were striking. REDMOD detected early disease signatures on average 475 days, well over a year, before clinical diagnosis. In comparative analyses, the model significantly outperformed experienced radiologists. It demonstrated nearly double the sensitivity (73% versus 39%) in identifying early malignant changes, and was almost three times as accurate in cases detected more than two years before diagnosis.
Further validation on independent datasets confirmed the model’s robustness. REDMOD correctly identified over 81% of cancer-free scans in a separate cohort of 539 patients and achieved an accuracy of 87.5% on a dataset from the National Institutes of Health.
Clinical implications
The potential clinical impact of earlier detection is substantial. According to the researchers, shifting diagnosis to earlier stages could dramatically improve survival outcomes. Modeling studies suggest that increasing the proportion of localized pancreatic cancers from 10% to 50% could more than double survival rates.
Importantly, REDMOD also demonstrated consistency in its predictions. When patients underwent multiple scans over time, the model produced the same results in 90–92% of cases, reinforcing confidence in its reliability as a diagnostic tool.
However, researchers caution that further validation is required, particularly in high-risk populations such as patients with unexplained weight loss or newly diagnosed diabetes, groups known to have elevated pancreatic cancer risk.
Data-driven oncology
While limitations remain, such as the lack of ethnic diversity in the study population, the development of REDMOD represents a significant step toward proactive cancer detection. Instead of reacting to symptoms, clinicians may soon be able to identify and intervene in cancer development before it becomes clinically apparent.
If validated in prospective clinical trials, REDMOD could become a key component in future screening strategies, particularly for high-risk patients. Ultimately, it signals a shift toward predictive, data-driven healthcare, where early detection is not the exception, but the standard.
Early cancer diagnosis
Earlier this month a large-scale Swedish study showed that AI can predict melanoma risk up to five years before diagnosis. Using healthcare data from over six million individuals, researchers identified subtle risk patterns based on factors such as age, medical history, medication use, and socioeconomic status.
The most advanced AI model achieved a predictive accuracy of 73 percent, outperforming simpler models based only on age and gender. It also identified high-risk subgroups with up to a 33 percent chance of developing melanoma within five years.
