AIMA 2026 projects ✦ Project VI ✦ AI in gastroenterology

Catching pancreatic cancer early, on MRI & CT

TranslationalHuman-Centered

Pancreatic cancer is the third leading cause of cancer death in the US, and most cases are found late1. Two chances to find it earlier hide in routine scans: cysts that may or may not turn malignant, and small tumours visible on CT months before diagnosis. Building on years of joint work between Northwestern and Mayo Clinic, we will make AI that grades cyst risk honestly and flags the scans that cannot wait.

In partnership with

  • Northwestern University
  • Mayo Clinic

Two chances to find it earlier

    MRI: branch-duct IPMN, T2 axial, Hellerhoff, Wikimedia Commons, CC BY-SA 4.0. CT: carcinoma of the pancreatic head, MBq, Wikimedia Commons, public domain; the pancreas outline is the uploader’s own, and the crosshair was removed. The AI steps are illustrative.

    What the collaboration has built

    • It began with 139 MRIs: deep learning matched the AGA and Fukuoka guidelines at grading IPMN risk (AUC 0.78 against 0.76 and 0.77)5.
    • PanSegNet segments the pancreas on 767 MRIs from five centers and 1,350 CTs, and the data are public6.
    • Cyst-X: 1,461 MRIs from 764 patients at seven centers, AUC 0.85 for risk, with sensitivity that matched or beat three radiologists (preprint)7.

    What is still hard

    • High-risk cysts: one seven-center model was right on 81.3% of low-risk but only 42.9% of high-risk cases (preprint)8.
    • Guidelines catch cancers but over-treat: 100% sensitivity, 19.8% specificity in 106 resected IPMNs2.
    • On CT, AI beat 68 radiologists in a 1,130-patient international study9, kept 92.9% sensitivity in 20,530 consecutive patients10, and has found cancers 3–36 months before diagnosis3. Prospective use is the next step. In colonoscopy, AI has already changed practice: in a tandem trial it halved the adenoma miss rate (15.5% against 32.4%)11.
    Aim 1 · Months 1–2

    Reproduce on open data

    Reproduce PanSegNet and the Cyst-X baselines, then move from whole-pancreas to cyst-level models aimed at the high-risk cases.

    Aim 2 · Months 3–4

    Guideline-aware and calibrated

    Measure the Kyoto worrisome features12 (cyst size, duct diameter) from the masks, combine them with deep features, and abstain when unsure.

    Aim 3 · Months 5–6

    Across centers

    Leave-one-center-out and federated tests across the seven centers, compared with radiologists and with the guidelines.

    Data

    • Cyst-X MRI benchmark, seven centers (OSF)7
    • PanSegNet MRI masks (OSF)6
    • PANORAMA CT training set and MSD Pancreas9,13

    You will learn

    • 3D segmentation on MRI and CT
    • Risk models with calibration and abstention
    • Federated and multi-center evaluation

    Good fit if you

    • Enjoy 3D medical imaging
    • Care about decisions, not only scores
    • Have an interest in GI or oncology
    1
    Data & masks

    Cyst-X, PanSegNet, center splits.

    2
    Risk model

    Cyst-level and multi-contrast models.

    3
    Features & uncertainty

    Guideline measurements, calibration, abstention.

    4
    Evaluation

    Center shift, readers, guideline comparison.

    ✦

    The plan · 6 months

    From first paper to submission.

    Six months, from first paper to submission, with weekly group meetings and a monthly review with mentors.

    Month 1
    Month 2
    Month 3
    Month 4
    Month 5
    Month 6
    VI · Gastroenterology
    Data & readingReproduce Cyst-XCyst-level, calibrated riskMulti-center testsWrite & submit
    Onboarding Baselines Method Experiments & studies Writing
    M1
    Onboard

    Read 10–15 key papers, set up data access and compute, write a one-page proposal.

    M2
    Reproduce

    Match a published baseline. Share a reproducibility report.

    M3
    Propose

    First version of the method, defended in a mentor review.

    M4
    Iterate

    Ablations, failure analysis, fix what the review found.

    M5
    Validate

    External data or pilot study; freeze results.

    M6
    Submit

    Paper, code release and a talk for the AIMA community.

    Project VI · AI in gastroenterology

    Want to work on
    this question?

    Apply through the form by November 7, 2026, choosing Project VI, with an academic CV and a statement of purpose of up to 300 words. Interviews run November 9 to 13, and groups start around November 18. Mentors from Northwestern University · Mayo Clinic.

    Apply by November 7 All nine projects

    References

    1. Siegel et al. Cancer statistics, 2026. CA Cancer J Clin 2026. DOI
    2. Borges et al. Assessing the Accuracy of the International Evidence-Based Kyoto Guidelines for Detecting Malignancy in Intraductal Papillary Mucinous Neoplasms of the Pancreas. GE Portuguese Journal of Gastroenterology 2025. DOI
    3. Korfiatis et al. Automated Artificial Intelligence Model Trained on a Large Data Set Can Detect Pancreas Cancer on Diagnostic Computed Tomography Scans As Well As Visually Occult Preinvasive Cancer on Prediagnostic Computed Tomography Scans. Gastroenterology 2023. DOI
    4. American Cancer Society. Survival rates for pancreatic cancer (SEER, diagnosed 2015–2021). Accessed October 2026. link
    5. Corral et al. Deep Learning to Classify Intraductal Papillary Mucinous Neoplasms Using Magnetic Resonance Imaging. Pancreas 2019. DOI
    6. Zhang et al. Large-scale multi-center CT and MRI segmentation of pancreas with deep learning. Medical Image Analysis 2025. DOI
    7. Pan et al. Cyst-X: A Multi-Center MRI Benchmark and Federated Learning Framework for Malignancy-Risk Stratification of Pancreatic Cystic Neoplasm. arXiv preprint 2025. arXiv
    8. Aktas et al. Uncertainty-Aware Explainable AI for Pancreatic Cysts: Identifying Deep Learning Vulnerabilities and Ensuring Safe Clinical Triage in IPMN Management. Research Square (preprint) 2026. DOI
    9. Alves et al. Artificial intelligence and radiologists in pancreatic cancer detection using standard of care CT scans (PANORAMA): an international, paired, non-inferiority, confirmatory, observational study. Lancet Oncology 2026. DOI
    10. Cao et al. Large-scale pancreatic cancer detection via non-contrast CT and deep learning. Nature Medicine 2023. DOI
    11. Wallace et al. Impact of Artificial Intelligence on Miss Rate of Colorectal Neoplasia. Gastroenterology 2022. DOI
    12. Ohtsuka et al. International evidence-based Kyoto guidelines for the management of intraductal papillary mucinous neoplasm of the pancreas. Pancreatology 2024. DOI
    13. Antonelli et al. The Medical Segmentation Decathlon. Nature Communications 2022. DOI