Researchers at the Mayo Clinic have developed an artificial intelligence approach that scans standard pathology slides to predict pancreatic cancer recurrence by analyzing the microscopic geography of residual tumors. According to a study published by the institution, the tool measures tissue fragmentation and cell mixing following initial treatments to identify patients at higher risk of early relapse.
Pancreatic ductal adenocarcinoma remains one of the most difficult malignancies to manage, characterized by aggressive local recurrence and high treatment resistance. Traditional pathology assessments primarily quantify the sheer volume of residual tumor left behind after therapies. However, standard methods often fail to explain why patients with similar tumor burdens experience divergent outcomes.
Mapping the Pancreatic Ecosystem with Landscape Ecology
To uncover hidden prognostic biomarkers, a research team led by Mayo Clinic oncologist Ryan Carr utilized methods adapted from landscape ecology. Investigators analyzed tissue samples from 203 patients with pancreatic ductal adenocarcinoma who received treatment before undergoing surgery but showed only a limited pathologic response.
By combining an AI-enabled digital pathology platform with ecological analysis techniques, the team examined standard hematoxylin and eosin (H&E) slides generated during routine clinical care. The algorithm measured tissue shape, boundary complexity, and the degree to which cancer cells and surrounding connective tissue, or stroma, were intermixed. Patients with a fragmented, intermixed pattern experienced earlier recurrence.
Predicting Recurrence Risk Without Extra Tissue Tests
The study demonstrated that two distinct spatial models successfully predicted disease-free survival even after adjusting for traditional clinical confounders, including tumor stage and lymph node status. In one analytical model, high-risk spatial patterns correlated with a 71% higher adjusted risk of recurrence. In a secondary model, high-risk patients exhibited more than double the adjusted risk.
Because the algorithm evaluates pathology slides already generated as part of routine surgical workups, it provides clinicians with additional prognostic information without requiring invasive repeat biopsies or extra tissue tests. “What is exciting is that this information is already present in the tissue,” Ryan Carr stated, noting that AI-enabled analysis measures features that are difficult to capture by eye.
Tumor Microenvironment and Immune Cell Sequestration
The digital pathology platform also uncovered distinct immunological differences tied to spatial architecture. High-risk spatial patterns contained fewer immune cells directly inside the tumor core. Instead, immune cells tended to collect around the periphery of the tumor rather than infiltrating the malignant mass.
This finding underscores the critical role of the tumor microenvironment in disease behavior and treatment resistance. The research aligns with the Mayo Clinic’s Precure Research priority, which seeks to use data and technology to predict risk earlier and intercept serious disease before progression.
The researchers emphasize that while the findings are promising, the approach requires prospective validation before it can be integrated into routine clinical decision-making. Future applications aim to guide more individualized surveillance, adjuvant therapy, and clinical trial design for patients with pancreatic cancer. The study was supported by the Gerstner Family Foundation Career Development Award, the Grand Forks Career Development Award, the Mayo Clinic Center for Clinical and Translational Science, and the ARPA-H ADAPT program.
Worth a look