Cyst-Suite: A Federated Foundational Model for Pancreatic Cyst Management

Abstract public domain (NIH RePORTER)

SUMMARY/ABSTRACT Pancreatic cancer is one of the deadliest cancers, with only 10% of patients surviving 5 years after diagnosis. One way to catch this cancer early is by finding and monitoring fluid-filled sacs called pancreatic cysts that form in the pancreas. Some types of these cysts can eventually become cancer, while others are harmless. Unfortunately, doctors struggle to tell these different types apart from using available guidelines and imaging. This research aims to develop a new computational tool called "Cyst-Suite" that uses artificial intelligence to analyze medical images of the pancreas. This tool will help doctors better identify which pancreatic cysts are dangerous and which ones are harmless. When doctors find pancreatic cysts today, they face a difficult choice: perform major surgery to remove what might be a harmless cyst, or risk missing early signs of deadly cancer. Cyst-Suite will work by analyzing CT and MRI scans of the pancreas using sophisticated computer algorithms. First, it will precisely outline the pancreas and any cysts within it, called segmentation. Then, it will identify what type of cyst is present from six common varieties (classification/diagnosis). Finally, it will estimate how likely each cyst is to become cancerous (diagnosis/prognosis), helping doctors make better decisions about whether to perform surgery or continue monitoring. An additional angle that makes this approach special is that Cyst-Suite can learn from thousands of patient scans across multiple hospitals without sharing sensitive patient information. This means the tool will work well for patients regardless of which hospital they visit or what type of scanner was used for their images. By developing this technology, I hope to save lives through earlier detection of dangerous pancreatic cysts while avoiding unnecessary surgeries for patients with harmless cysts. Cyst-Suite project via this K99/R00 grant will be instrumental in my transition to becoming an independent investigator at the intersection of artificial intelligence and medical imaging. Through this project, I will develop specialized expertise in pancreatic imaging analysis while building essential skills in multi-institutional collaboration, clinical validation, and regulatory navigation. The mentorship team assembled for this project— spanning radiology, gastroenterology, AI, and biostatistics—will provide the interdisciplinary training foundation necessary for my future research program focused on AI-driven early cancer detection systems. Once complete, Cyst-Suite could transform how doctors’ approach pancreatic cancer prevention and serve as a model for similar tools targeting other cancers. Additionally, the proposed work will generate preliminary data and technological infrastructure that will serve as a springboard for future R01 applications, helping establish my independent research career in developing innovative AI solutions for challenging clinical problems.

Public health relevance public domain (NIH RePORTER)

NARRATIVE Pancreatic cystic lesions present a critical diagnostic dilemma where current guidelines fail to accurately distinguish benign cysts from precancerous lesions, resulting in both unnecessary surgeries and missed opportunities for early intervention. Cyst-Suite will transform clinical practice through an AI-powered platform that precisely segments pancreatic anatomy, differentiates between cyst subtypes, and provides accurate risk stratification across diverse imaging conditions. Through innovative federated learning, this tool will undergo rigorous validation across major medical centers without compromising patient privacy, ensuring robust performance for clinical deployment. This K99/R00 research will establish both a powerful clinical tool for personalized pancreatic cancer surveillance and my independent research career at the intersection of artificial intelligence and precision oncology.