Clinical Scorecard: A validated three-dimensional deep learning method for measuring vascular invasion in pancreatic ductal adenocarcinoma

At a Glance

Key Highlights

• PAN-VIQ provides automated segmentation of pancreatic tumors and five critical vessels (CA, CHA, SMA, SMV, PV) from CT scans.
• The model quantifies vascular involvement through continuous 3D encasement angles, improving accuracy over subjective 2D assessments.
• Prospective and external validations demonstrated >90% accuracy, outperforming junior radiologists and matching senior radiologists.

Guideline-Based Recommendations

Diagnosis

• Use contrast-enhanced CT imaging for preoperative evaluation of vascular involvement in PDAC.
• Incorporate objective, quantitative assessment of tumor–vessel relationships rather than relying solely on subjective 2D interpretations.
• Assess multiple vessels simultaneously due to frequent multi-vessel involvement in PDAC.

Management

• Utilize precise vascular invasion quantification to guide surgical planning, anticipate vascular reconstruction needs, and reduce positive margin risk.
• Consider vessel-specific clinical implications, recognizing arterial involvement (e.g., SMA) may increase operative complexity even at lower encasement degrees.

Monitoring & Follow-up

• Apply standardized, reproducible imaging assessments to reduce interobserver variability, especially among less experienced radiologists.
• Monitor continuous encasement angles rather than fixed categorical cutoffs to better reflect biological and surgical relevance.

Risks

• Be aware that subjective and categorical vascular invasion assessments may lead to inconsistent surgical decision-making.
• Recognize that incomplete evaluation of multiple vessels may underestimate vascular involvement and affect treatment outcomes.

Patient & Prescribing Data

Patients diagnosed with pancreatic ductal adenocarcinoma undergoing preoperative imaging

Automated 3D quantification of vascular invasion supports individualized surgical planning and may improve resectability assessment accuracy.

Clinical Best Practices

• Adopt automated deep learning tools like PAN-VIQ for objective, continuous measurement of tumor encasement around key vessels.
• Evaluate all major peripancreatic vessels (CA, CHA, SMA, SMV, PV) to capture comprehensive vascular involvement.
• Integrate quantitative vascular invasion data into multidisciplinary treatment planning to optimize surgical outcomes.

References

• Pancreatic ductal adenocarcinoma prognosis and surgical resection statistics
• Interobserver variability in vascular invasion assessment in PDAC
• Deep learning applications in PDAC imaging and vascular invasion quantification