Pancreatic cancer has a five-year survival rate of just 11%, the lowest of any major cancer. The disease is almost always diagnosed late, when surgery is no longer an option and chemotherapy has limited effectiveness. Improving outcomes requires breakthroughs in early detection and treatment personalization.
Artificial intelligence, particularly machine learning and deep learning, has emerged as a powerful tool across medicine. In pancreatic cancer, AI is being applied to radiology images, pathology slides, blood biomarkers, surgical planning, and treatment outcome prediction. This comprehensive review from Peking Union Medical College surveys the full landscape of these applications.
The authors highlight that while AI cannot yet replace expert clinicians, it can augment their capabilities—finding patterns invisible to the human eye, processing data at superhuman speed, and generating predictions that inform more targeted treatments.
One of the most promising applications of AI is in early detection from medical images. Deep learning models applied to CT scans have demonstrated the ability to detect pancreatic tumors that are smaller than 2 centimeters—tumors that are surgically curable but often missed by radiologists on routine scans.
AI has also been applied to identify high-risk precancerous lesions, including intraductal papillary mucinous neoplasms (IPMNs). These fluid-filled cysts sometimes transform into cancer, and AI models trained on imaging data can predict which cysts are most likely to become malignant.
Beyond imaging, AI models have analyzed patterns in electronic health records—changes in blood sugar, new-onset diabetes, subtle shifts in pancreatic enzyme levels—to flag patients at elevated cancer risk months or years before traditional diagnosis. This EHR-based surveillance approach could integrate seamlessly into routine primary care.
Digital pathology is another major area where AI is making inroads. Deep learning models trained on stained tissue slides can identify cancer cells, grade tumor aggressiveness, and detect molecular features—such as KRAS mutations—directly from images without additional genetic testing.
AI is also improving the value of blood-based biomarkers. While CA19-9, the standard pancreatic cancer blood marker, has poor accuracy on its own, machine learning models combining CA19-9 with other proteins, metabolites, and clinical data have achieved significantly higher diagnostic accuracy.
Prognosis prediction—forecasting how long a patient is likely to survive and how well they will respond to treatment—is perhaps the most clinically urgent AI application. Models incorporating tumor genetics, imaging features, and clinical variables can now stratify patients into risk groups that guide decisions about aggressive surgery versus palliative care.
Surgical planning is being transformed by AI-powered 3D reconstruction of pancreatic anatomy from CT scans. These models map the tumor's relationship to critical blood vessels—information essential for surgeons deciding whether a tumor is resectable. AI can perform this analysis faster and sometimes more accurately than manual methods.
In radiation therapy, AI is automating the contouring of tumors and organs at risk on treatment planning images, a step that typically takes experienced physicians hours. AI-assisted contouring reduces variability between treatment centers and frees radiation oncologists to focus on plan optimization.
Drug discovery for pancreatic cancer is also benefiting from AI. Machine learning models are predicting which existing drugs might be repurposed against pancreatic cancer, identifying synergistic drug combinations, and designing novel compounds targeting KRAS—the most common mutation in this disease, previously considered undruggable.
Despite impressive results, the review identifies important barriers to clinical adoption. Most AI models have only been tested retrospectively on single-institution datasets, raising concerns about whether they will perform as well when applied in real-world settings across different hospitals with different imaging equipment and patient populations.
Data access and standardization remain critical obstacles. Pancreatic cancer is relatively rare, meaning no single hospital accumulates enough cases to train robust AI models alone. International data-sharing consortia and federated learning approaches—where AI trains across institutions without sharing raw patient data—are essential next steps.
The authors call for a collaborative effort among clinicians, data scientists, biostatisticians, and engineers to move AI applications from research papers into clinical trials and eventually routine practice. Regulatory pathways for AI diagnostic tools are also still evolving, and real-world evidence standards need to be established.