AI in Pancreatic Diseases (Cancer, Cysts, Diabetes and Pancreatitis)

Welcome to our research portal dedicated to advancing the diagnosis of pancreatic diseases through artificial intelligence (AI). Our team is pioneering novel computational approaches for the early detection and accurate characterization of pancreatic conditions, including Intraductal Papillary Mucinous Neoplasms (IPMNs), pancreatitis, diabetes, and pancreatic cancer.

Through innovative machine learning algorithms, deep learning architectures, and advanced image analysis techniques, we aim to transform the landscape of pancreatic disease diagnosis, leading to improved patient outcomes through earlier intervention and personalized treatment strategies.

About

Our Mission

Bagci Lab, in other words Hybrid & Machine Intelligence Lab, is dedicated to developing and validating artificial intelligence tools that enhance the diagnosis of pancreatic diseases. We believe that AI-powered diagnostic approaches have the potential to significantly improve detection rates, reduce diagnostic delays, and ultimately save lives.

Research Focus

Our work centers on four critical areas of pancreatic disease:

1. IPMN Cysts: Developing algorithms for accurate detection, characterization, and risk
   stratification of Intraductal Papillary Mucinous Neoplasms
2. Pancreatitis: Creating computational models to improve diagnosis and severity
   assessment of acute and chronic pancreatitis
3. Diabetes: Quantifying pancreatic tissue changes to identify changes related to diabetic
   conditions, and prognosis of diabetic patients towards various outcomes
4. Pancreatic Cancer: Advancing early detection methods and prognostic indicators for
   pancreatic adenocarcinoma and other pancreatic malignancies

Interdisciplinary Approach

Our team brings together expertise from multiple disciplines, including:

• Radiology and Medical Imaging
• Computer Science and Machine Learning
• Gastroenterology
• Pathology
• Statistics

This cross-disciplinary collaboration enables us to develop comprehensive AI solutions that
address the complex challenges of pancreatic disease diagnosis.

Research

IPMN Cyst Diagnosis

Intraductal Papillary Mucinous Neoplasms (IPMNs) represent one of the most common
pancreatic cystic lesions and pose a significant diagnostic challenge due to their malignant
potential. Our research focuses on developing AI tools that can:

• Accurately detect and diagnose IPMN cysts on MRI scans
• Distinguish IPMNs from other pancreatic cystic lesions
• Assess malignant potential and predict progression risk
• Guide clinical decision-making for surveillance versus surgical intervention

Key Publications:

• Adaptive Aggregation Weights for Federated Segmentation of Pancreas MRI
• IPMN Risk Assessment under Federated Learning Paradigm
• Large-Scale Multi-Center CT and MRI Segmentation of Pancreas with Deep Learning
  (PanSegNet)
• Bridging the Gap in Pancreatic Imaging: A New Era of AI-Powered Precision
• Advances for Managing Pancreatic Cystic Lesions: Integrating Imaging and AI
  Innovation

 

Pancreatitis Diagnosis

Our work in pancreatitis encompasses both acute and chronic forms of the disease, with AI applications designed to:

• Rapidly identify acute pancreatitis and predict severity
• Differentiate between interstitial and necrotizing forms
• Monitor disease progression and treatment response
• Diagnose chronic pancreatitis at earlier stages through subtle imaging biomarkers

Key Publications:

• Detection of Peri-Pancreatic Edema using Deep Learning and Radiomics Techniques
• DDW 2024 abstracts:
  • Deep Learning based automated pancreas segmentation and volumetry measurement in patients with acute pancreatitis
  • Patients with acute pancreatitis have significant decline in pancreas volume over time
  • The association of volumetric changes and disease severity with endocrine dysfunction following acute pancreatitis

 

 

Pancreatic Cancer Diagnosis

Early detection remains the greatest challenge in pancreatic cancer. Our research leverages AI to:

• Identify subtle imaging features associated with early pancreatic cancer
• Develop risk prediction models integrating imaging, clinical, and laboratory data
• Differentiate pancreatic cancer from benign conditions with similar presentations
• Assess treatment response and predict survival outcomes

Key Publications:

• Optimizing synthetic data for enhanced pancreatic tumor segmentation
• Harmonized spatial and spectral learning for robust and generalized medical image segmentation (BIRPA Prize ICPR 2024)
• SynergyNet: Bridging the gap between discrete and continuous representations for precise medical image segmentation
• Beyond self-attention: deformable large kernel attention for medical image segmentation

 

Technical Approaches

Deep Learning Architectures

Our research employs various deep learning architectures optimized for medical imaging analysis including segmentation and diagnostic/patient outcome predictions:

• Generative AI: For synthetic data generation to augment limited training datasets, and for segmentation improvements too. Our conditional generative models create realistic pancreatic lesion images that enhance model training in scarce data scenarios. Additionally, we leverage these generative approaches to improve segmentation in cases with poor lesion boundaries or artifacts.
• Transformers: For integrating multimodal data and capturing long-range dependencies in volumetric pancreatic imaging. Our transformer-based models excel at contextualizing local features within global anatomical structures, enabling more accurate differentiation between benign and malignant pancreatic lesions. These architectures have proven particularly effective for characterizing the heterogeneous presentation of pancreatic diseases.
• U-Net style algorithms: For pancreatic segmentation and lesion delineation, offering precise boundary identification of both normal pancreatic tissue and pathological regions. Our enhanced U-Net variants incorporate attention mechanisms and deep supervision to improve segmentation accuracy in challenging cases with irregular lesion morphology.
• Convolutional Neural Networks (CNNs): For feature extraction and classification in pancreatic imaging, enabling automated detection of subtle imaging biomarkers associated with early pancreatic lesions. Our custom architectures incorporate domain-specific modifications to standard CNN designs to account for the unique challenges of pancreatic imaging.
• Federated Learning: For collaborative model training across multiple institutions without sharing sensitive patient data. This privacy-preserving approach enables our AI systems to learn from diverse patient populations while maintaining data security and regulatory compliance. Our federated learning framework has facilitated multi-center validation of our diagnostic algorithms, significantly enhancing their generalizability.

Radiomics and Feature Engineering

We developed comprehensive radiomic feature extraction pipelines that capture:

• Morphological characteristics of pancreatic lesions
• Textural features reflecting tissue heterogeneity
• First-order statistics describing intensity distributions
• Wavelet and frequency domain features
• Delta-radiomics for temporal evolution assessment

Multimodal Data Integration

Our AI systems integrate information from multiple sources:

• Cross-sectional imaging (CT, MRI)
• Laboratory biomarkers
• Clinical parameters
• Histopathological data (ongoing)
• Genomic and molecular profiles (ongoing)

Resources

Datasets

We are committed to advancing the field through data sharing and collaboration. Below are dataset developed by Bagci lab with several national/international partners (Northwestern University, NYU, Mayo Clinic, Erasmus MC, Columbia University, Istanbul University School of Medicine, Allegenhy Health ), and available to qualified researchers:

• Cyst-X: IPMN MultiContrast MRI Dataset (segmentation and diagnosis)
• Peri-Pancreatic Edema (CT)
• Pediatric Pancreatitis (MRI)

 

Software Tools

Open-source software tools developed by our team:

• PanSegNet: Automated pancreas segmentation tool (MRI, CT) GitHub
• Cyst-X IPMN-Classifier: Pre-trained models for IPMN classification (MRI) GitHub
• Per-Pancreatic Edema Classifier: Risk prediction toolkit (CT) GitHub
• Radiomics for IPMN Classification: MRI GitHub

 

Educational Materials

Resources for clinicians and researchers:

• AI in Pancreatic Imaging Tutorial: View Dr. Bagci’s one of recent talk at SAR 2025.
• AI for Gastroenterology: View Dr. Bagci’s one of recent talks
• Review paper on AI and Radiomics for Pancreatic Cancer Diagnosis
• Review paper on Pancreas Segmentation from Radiology Scans

 

Team

Collaborations and Our Team

Funding Organizations

Our work is supported by funding from:

• National Institutes of Health (NIH): R01 CA246704, U01 DK 127384-02S1

 

Academic Partners

• We collaborate with leading institutions worldwide:

Principal Investigators

• Ulas Bagci, PhD
  • Research focus: AI applications in healthcare
  • Northwestern University
  • ulas.bagci@northwestern.edu| GoogleScholar
• Michael B. Wallace, MD
  • Research focus: Gastroenterology
  • Mayo Clinic Jacksonville
  • Wallace.michael@northwestern.edu | GoogleScholar

Co-Investigators & Collaborators

• Rajesh N. Keswani, MD
  • Research focus: Gastroenterology
  • Northwestern University
• Frank Miller, MD
  • Research focus: Radiology
  • Northwestern University
• Tamas Gonda, MD
  • Research focus: Gastroenterology
  • New York University
• Candice Bolan, MD
  • Research focus: Radiology
  • Mayo Clinic Jacksonville
• Chenghang Huang, PhD
  • Research focus: Radiology
  • New York University
• Marco Bruno, MD, PhD
  • Research focus: Gastroenterology
  • Erasmus MC
• Ivo G. Schoots, MD
  • Research focus: Radiology
  • Erasmus MC
• Sachin Jambawalikar, PhD
  • Research focus: Radiology, AI, Imaging
  • Columbia University
• Ziyue Xu, PhD
  • Research focus: AI in Healthcare
  • NVIDIA
• Pallavi Tiwari, PhD
  • Research focus: AI in Healthcare
  • University of Wisconsin-Madison
• Concetto Spampinato, PhD
  • Research focus: Computer Vision
  • University of Catania
• Yury Velichko, PhD
  • Research focus: AI in Healthcare
  • Northwestern University
• Lili Zhao, PhD
  • Research focus: Statistics
  • Northwestern University
• Alpay Alibeyoglu, MD
  • Research focus: Internal Medicine, Gastroenterology
  • Istanbul University School of Medicine

 

Mayo-Northwestern Clinical Fellow and Coordinator

Gorkem Durak, MD

• • Research focus: Radiology
  • Northwestern University

 

Graduate Students & Fellows

• Elif Keles, MD, PhD
  • Senior Clinical Fellow
  • Northwestern University
• Hongyi Pan, PhD
  • Senior Research Fellow
  • Northwestern University
• Zheyuan Zhang, PhD
  • Research Fellow
  • Northwestern University
• Ziliang Hong
  • PhD Student
  • Northwestern University
• Andrea Mia Bejar
  • MD Student
  • Northwestern University
• Halil Ertugrul Aktas
  • MD Student
  • Northwestern University
• Deniz Seyithanoglu
  • MD Collaborator
  • Istanbul University School of Medicine
• Maria Jaramillo Gonzalez
  • PhD Student
  • University of Wisconsin-Madison
• Linkai Peng
  • PhD Student
  • Northwestern University

 

For Patients

While our research is primarily aimed at the scientific and clinical communities, we understand that patients with pancreatic conditions may visit this site seeking information.

Understanding AI in Pancreatic Disease Diagnosis

• AI systems analyze medical images and data much like radiologists do, but can detect patterns that might be invisible to the human eye
• These tools aim to assist doctors, not replace them
• Early research shows AI may help detect pancreatic diseases earlier and more accurately

Participating in Research

If you’re interested in contributing to our research as a donor, please contact PI: Dr. Bagci.

Resources for Patients

• Pancreatic Cancer Action Network
• National Pancreas Foundation
• American Gastroenterological Association Patient Center
• Mayo Clinic Pancreatic Cancer Care
• Lurie Cancer Center – Pancreatic Cancer Treatment

 

Contact, Media Inquiries and Research Collaborations

• Email: ulas.bagci@northwestern.edu

 

Location

Machine & Hybrid Intelligence Lab,
Department of Radiology,
737 N Michigan Ave,
Chicago, IL 60611, USA

Follow Our Work

• Twitter
• LinkedIn
• GitHub
• GoogleScholar