Validated 3D Deep Learning Method for Vascular Invasion in PDAC

Overview

PAN-VIQ is an automated deep learning framework that quantifies three-dimensional tumor–vessel interactions in pancreatic ductal adenocarcinoma (PDAC) using contrast-enhanced CT scans. It segments pancreatic tumors and five critical vessels, achieving over 90% accuracy in external validation and outperforming junior radiologists in prospective testing.

Background

Pancreatic ductal adenocarcinoma is a highly lethal cancer with limited curative options, where surgical resection depends heavily on accurate assessment of vascular invasion. Current CT-based evaluations rely on subjective 2D interpretations, leading to significant interobserver variability and incomplete assessment of multiple vessel involvement. Precise preoperative knowledge of tumor encasement around key arteries and veins is essential for surgical planning and prognosis. Advances in artificial intelligence offer opportunities to improve the objectivity and reproducibility of vascular invasion assessment.

Data Highlights

Key Findings

• PAN-VIQ segments pancreatic tumors and five major vessels (CA, CHA, SMA, SMV, PV) automatically from CT scans.
• Quantifies vascular involvement using continuous 3D encasement angles rather than binary or ordinal categories.
• Achieved over 90% accuracy in external validation datasets.
• Outperformed junior radiologists and matched senior radiologists in prospective evaluation of vascular invasion.
• Addresses limitations of prior models by assessing multiple vessels simultaneously and providing anatomically precise quantification.

Clinical Implications

PAN-VIQ offers a standardized, objective tool for preoperative vascular invasion assessment in PDAC, potentially reducing interobserver variability and improving surgical planning. Its multi-vessel, continuous quantification approach may better capture the complexity of tumor–vessel relationships, aiding decisions on resectability and vascular reconstruction needs.

Conclusion

The PAN-VIQ deep learning framework represents a significant advancement in automated, three-dimensional quantification of vascular invasion in PDAC, demonstrating high accuracy and clinical utility. Its adoption could enhance consistency and precision in preoperative evaluation, ultimately improving patient management.

References

1. Original Article -- A validated three-dimensional deep learning method for measuring vascular invasion in pancreatic ductal adenocarcinoma