Study Design Governance Sites Team Contact ↗ Patient Information
predict-panc.org For Healthcare Professionals
Clinical & Research Information

PREDICT-PanC

Protocol summary, algorithm performance data, and site participation information for clinicians, researchers, and site coordinators

Royal Free London NHS Foundation Trust · University College London · University of Basel

Ethics submission pending
Algorithm accepted for MIDL 2026
0.86
AUC — retrospective validation
0.79
Macro-F1 score
159
Validation cohort (patients)
2 phases
Prospective UK study design
Study protocol

Protocol overview

PREDICT-PanC is a prospective multicentre UK study evaluating a validated deep learning algorithm that analyses pre-MDT CT scans to classify pancreatic cancer resectability according to NCCN criteria (resectable / borderline resectable / locally advanced).

Phase A

Prospective Blinded Diagnostic Accuracy

The AI algorithm runs silently alongside standard MDT assessment. Clinical teams are blinded to AI outputs. Predictions are compared against MDT decisions and intraoperative / histopathological findings to establish diagnostic accuracy in a prospective real-world setting.

Phase B

Pragmatic Implementation Study

AI outputs are made available to MDTs as decision-support information prior to resectability discussion. The study evaluates whether AI-augmented MDT decision-making improves correct treatment allocation, reduces futile laparotomy rates, and affects time-to-treatment decision.

Primary Endpoint
Correct classification of resectability status against intraoperative findings and histopathology (diagnostic accuracy of AI vs standard MDT assessment)
Key Secondary Endpoints
Futile laparotomy rate (exploratory laparotomy without resection)
R0 resection rate in patients proceeding to surgery
90-day morbidity assessed by Clavien-Dindo classification and Comprehensive Complication Index (CCI)
Time-to-treatment decision from MDT discussion
Health-economic analysis: cost per futile laparotomy avoided

Deep learning model performance

The algorithm was developed and validated on a retrospective cohort using stratified nested 5-fold cross-validation. Results have been accepted for presentation at the Medical Imaging with Deep Learning (MIDL) conference, Taipei, Summer 2026.

AUC
0.86
Area under ROC curve — multiclass resectability classification
Macro-F1
0.79
Balanced F1 score across all three resectability classes
Accuracy
0.85
Overall classification accuracy on held-out validation folds
Classification Resectable · Borderline resectable · Locally advanced (NCCN criteria)
Input Routine pre-MDT contrast-enhanced CT (portal venous phase)
Cohort 159 patients — retrospective multicentre development and validation
Validation Stratified nested 5-fold cross-validation
Development University of Basel — Dept. Biomedical Engineering (Prof. Philippe Cattin)
Output Resectability classification + confidence scores per class
Accepted — MIDL 2026 Medical Imaging with Deep Learning · Taipei · Summer 2026

Full methodology and supplementary performance data available on request. Please contact the study team directly for access to the technical documentation or pre-publication manuscript.

Ethics, approvals & data governance

The prospective study has not yet commenced. All required approvals will be in place before data collection begins at any site.

NHS REC Ethics

Full NHS Research Ethics Committee submission in preparation. HRA approval required prior to site initiation.

Data Governance

All CT scans and clinical data fully anonymised. No personal identifiers in the research dataset. UK GDPR and NHS Code of Confidentiality compliant.

Portfolio Manager

Initial contact established with the NIHR CRN portfolio manager at the lead site. Portfolio registration planned prior to Phase A initiation.

NHS R&D

Trust R&D office engagement ongoing at the lead site. Site-specific assessments will be required at each participating centre.

Swiss Ethics (Retrospective)

Approved by the cantonal ethics committee. BASEC-ID: 2021-00457. Covers the retrospective algorithm development and validation cohort.

Sponsor

Sponsorship arrangements under discussion with the Trust Research Office at the lead site.

Study sites & participation criteria

PREDICT-PanC will recruit from established UK HPB centres managing a sufficient volume of pancreatic cancer MDT cases. We are actively seeking expressions of interest from sites wishing to participate.

Royal Free London NHS Foundation Trust

Lead site · London · HPB & Liver Transplant Surgery

Lead Site

Additional UK HPB Centres

Multicentre recruitment — expressions of interest welcome

Sites TBC

If your centre manages pancreatic cancer MDT discussions and you are interested in site participation, please contact the study team. No commitment required at this stage.

Site participation criteria

Designated HPB or upper GI cancer centre with a functioning pancreatic cancer MDT
Minimum case volume: approximately 20+ pancreatic cancer MDT discussions per year
Access to PACS system and ability to export anonymised DICOM CT data
Named site principal investigator with HPB or oncology background
Local R&D office capacity to support site set-up and regulatory approvals
Commitment to prospective data capture and follow-up at defined timepoints

Research team

The study brings together expertise in HPB surgery, clinical AI, biomedical engineering, radiology, anaesthetics, and biostatistics across Royal Free London, UCL, and the University of Basel.

SS

Mr Sebastian Staubli, MD, Dr. habil.

Principal Investigator
Senior Clinical Fellow, HPB & Liver Transplant Surgery · Royal Free London NHS FT · Honorary Research Associate, UCL
JP

Prof Joerg-Matthias Pollok, MD, PhD, FRCS

HPB & Liver Transplant Surgery
Consultant HPB & Liver Transplant Surgeon · Clinical Lead for HPB Surgery & Liver Transplantation · Royal Free London NHS FT · Professor of Surgery, UCL
PC

Prof Philippe C. Cattin, PhD

Medical Image Analysis & Navigation
Head, Center for medical Image Analysis & Navigation (CIAN) · Head, Dept. Biomedical Engineering · University of Basel
VO

Vincent Ochs

Algorithm Development
PhD Candidate · Center for medical Image Analysis & Navigation (CIAN) · Dept. Biomedical Engineering · University of Basel

Site enquiries & collaboration

For site participation enquiries, co-investigator discussions, access to technical documentation, or questions about the study design, please get in touch.

For patient and public information, visit predict-panc.org

Mr Sebastian Staubli, MD, Dr. habil.
Principal Investigator, PREDICT-PanC
Senior Clinical Fellow, HPB & Liver Transplant Surgery
Royal Free London NHS Foundation Trust
Honorary Research Associate, UCL
Data privacy notice

By submitting this form, your message is sent via Web3Forms, a third-party form processing service. Your name, email, and message are forwarded to the research team and are not stored by Web3Forms beyond delivery.

Study status: This website provides information about a research study currently under development. The study is being prepared for funding applications and ethics review, and has not yet begun recruiting participants. Information on this page is provided for transparency and to support patient and public involvement.