Pancreatic ductal adenocarcinoma (PDAC) is the deadliest common cancer, with an overall five-year survival rate of only 8.5%. Early-stage diagnosis can raise survival to 20%, but the vast majority of patients are diagnosed too late for curative surgery — less than 15% have resectable disease at diagnosis.
The explosion of biomedical data — from imaging, genomics, pathology slides, and electronic health records — has created both an opportunity and a challenge. Artificial intelligence (AI) can analyze this data at a scale and speed far beyond human capacity, potentially unlocking earlier detection, better treatment matching, and survival prediction.
This review from Kumamoto University in Japan surveys the current state of AI applications across the full spectrum of PDAC care, providing a practical reference for clinicians wanting to understand where the technology stands.
Several AI models have been developed to identify individuals at high risk of developing PDAC before they have symptoms. Because about 50% of PDAC patients develop diabetes before their cancer diagnosis, AI models targeting new-onset diabetic patients have shown particular promise.
One logistic regression model analyzed 109,385 new-onset diabetic patients and identified PDAC risk using clinical variables including age, smoking, BMI change, medication use, and blood test results. By flagging only 6.2% of new-onset diabetics for intensive screening, the model could detect 44.7% of PDAC cases at 94% specificity — a practical trade-off for population screening.
Other AI approaches have analyzed electronic health records using ICD diagnosis codes, personal health data with neural networks, and laboratory values to build composite risk scores, with area-under-the-curve values ranging from 0.71 to 0.85.
Multiple AI algorithms have been applied to CT imaging to detect and locate PDAC tumors automatically. Deep convolutional neural networks (DCNNs) have achieved sensitivity of 83–98% for tumor detection on CT, with area-under-the-curve values exceeding 0.92 in several studies.
Endoscopic ultrasound (EUS) is another key diagnostic modality where AI has shown strong performance. Studies using artificial neural networks and support vector machines have achieved sensitivity and specificity both above 90% for distinguishing PDAC from benign pancreatic tissue on EUS images — comparable to or exceeding expert endoscopists.
AI models analyzing liquid biopsy markers (circulating microRNAs, bile juice) have also been tested, with one microRNA signature model achieving 93% sensitivity and 92% specificity for PDAC detection.
Beyond diagnosis, AI has been applied to predict surgical outcomes and complications. AI models using preoperative CT features can predict pancreatic fistula after pancreaticoduodenectomy (Whipple procedure), one of the most feared surgical complications, helping surgeons plan accordingly.
For molecular subtyping — determining which biological subtype of PDAC a patient has — AI models using CT and MRI imaging features achieved 84–90% sensitivity and 92% specificity for classifying patients as quasi-mesenchymal or non-quasi-mesenchymal subtypes, which have different treatment responses.
AI also shows promise for predicting microsatellite instability (MSI) status from imaging alone, which is important because MSI-high tumors respond to immunotherapy — a finding that could spare patients invasive biopsies.
The review highlights that AI-based omics analysis — integrating genomic, transcriptomic, and proteomic data — represents the next frontier for improving PDAC outcomes. By discovering biomarkers for early detection, molecular subtyping, and treatment guidance, AI could transform PDAC from a disease caught too late to one diagnosed early enough to cure.
Major barriers remain, including the need for large, well-annotated multicenter datasets, standardization of AI model development and reporting, and prospective validation in clinical trials. Most AI tools reviewed are still in research phases and have not yet been implemented in routine clinical care.
The authors advocate for collaborative data sharing between institutions and the development of AI tools with clear clinical workflows, so that the promising research findings can actually benefit patients.