JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Validation of a Pretrained Artificial Intelligence Model for Pancreatic Cancer Detection on Diagnosis and Prediagnosis Computed Tomography Scans

  • 0Novo Nordisk Foundation Center for Protein Research, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark (L.D., D.P., P.C., S.B.); Guerbet Research, Villepinte, France (C.A.N., A.B., R.V., M.M.R.); Department of Health Technology, Technical University of Denmark (DTU), Kongens Lyngby, Denmark (F.D.M.).
Investigative Radiology +

|

June 24, 2025

|

June 24, 2025

View abstract on PubMed

Summary

This summary is machine-generated.

The AI model PANCANAI demonstrated high sensitivity in detecting pancreatic cancer (PC) from CT scans, even over a year before diagnosis. This tool shows promise for early pancreatic cancer detection in clinical practice.

Area Of Science

  • Medical Imaging
  • Artificial Intelligence in Oncology
  • Radiology

Background

  • Pancreatic cancer (PC) detection remains a challenge, often diagnosed at late stages.
  • Early detection is crucial for improving patient outcomes.
  • AI-powered tools offer potential for enhancing diagnostic accuracy.

Purpose Of The Study

  • To evaluate the performance of the PANCANAI AI model for pancreatic cancer detection.
  • To assess the model's ability to detect PC on CT scans acquired prior to histopathologic diagnosis.
  • To validate PANCANAI on a longitudinal cohort of patients.

Main Methods

  • The PANCANAI model, previously trained on 2134 portal venous CTs, was applied to a retrospective cohort of 1083 Danish patients with biopsy-confirmed PC.
  • CT scans were categorized into concurrent diagnosis (within 2 months of diagnosis) and prediagnosis groups (scans acquired before diagnosis).
  • Sensitivity was measured, with bootstrapping used to calculate median and 95% confidence intervals (CI).

Main Results

  • The study included 1083 PC patients (mean age 69 years).
  • Sensitivity was 91.8% for concurrent diagnosis scans and 68.7% for prediagnosis scans (median 7 months prior).
  • Sensitivity remained at 53.9% for scans acquired one year or more before diagnosis, and 82.9% for stage I PC.

Conclusions

  • PANCANAI demonstrated high sensitivity for automated pancreatic cancer detection in a large retrospective cohort.
  • The AI model successfully identified PC suspicion in over half of CT scans obtained at least one year before histopathologic diagnosis.
  • PANCANAI shows potential as a valuable tool for early pancreatic cancer detection.

Keywords: