Project Summary
Pancreatic ductal adenocarcinoma (PDAC) is projected to become the second leading cause of cancer-related
death by 2030, yet there are no accurate diagnostic tests for early diagnosis. Among pancreatic cystic lesions
(PCLs), branch duct (BD) intraductal papillary mucinous neoplasm (IPMN) is the most common precursor
lesion for pancreatic cancer. Nearly 50% of all prevalent cysts are BD-IPMNs. Endoscopic ultrasound (EUS)-
guided fine needle aspiration (FNA) of PCLs and cyst fluid analysis are standard-of-care (SOC) diagnostic
modalities. Unfortunately, the current SOC is suboptimal (65-75% accuracy) for the detection and risk
stratification [high-grade dysplasia or adenocarcinoma (HGD-Ca) vs. low-grade dysplasia (LGD)] of BD-IPMNs.
The goal of surgery in BD-IPMNs is to resect lesions with HGD-Ca. However, multiple surgical series over the
last 5 years have revealed that nearly half to two-thirds of resected BD-IPMNs had only LGD, often
representing overtreatment. In these instances, the morbidity (30%) and mortality (2%) from surgical resection
of PCLs are not justified. On the other hand, several series reports missed (mean 13%) invasive cancers in
BD-IPMNs during follow-up. There are currently no accurate tests for detecting HGD-Ca in BD-IPMNs. We
have utilized a novel diagnostic modality of EUS-guided needle-based confocal laser endomicroscopy (nCLE),
a technology that provides in vivo, real-time, optical biopsies of PCLs. In a landmark study, we demonstrated a
high accuracy (97%) for nCLE-guided diagnosis of precancerous (includes mucinous BD-IPMNs) PCLs. We
have derived nCLE features of HGD-Ca that can be qualitatively assessed and quantitatively analyzed in BD-
IPMNs. We also have designed a pilot CLE-based convolutional neural network (CNN)-artificial intelligence
(AI) algorithm to risk-stratify BD-IPMNs (HGD-Ca vs. LGD). We have also pioneered cyst fluid Next-Generation
Sequencing (NGS) analysis, augmenting the diagnosis and risk-stratification of BD-IPMNs. The primary
objective of the proposed study is to accurately risk-stratify (HGD-Ca vs. LGD) BD-IPMNs to detect early-stage
PDAC and avoid unjustified pancreatic surgery. Supported by preliminary data, our central hypothesis is that
EUS-nCLE (manual and CNN-AI algorithm) and a combination of EUS-nCLE with NGS and SOC variables will
accurately risk-stratify BD-IPMNs. Specific aims – (1) Evaluate the accuracy and interobserver agreement of
EUS-nCLE differentiation (HGD-Ca vs. LGD) of BD-IPMNs among independent observers. (2) Improve and
prospectively evaluate an accurate nCLE-based CNN-AI algorithm for presurgical risk stratification (HGD-Ca
vs. LGD) of BD-IPMNs. (3) Evaluate an integrative diagnostic approach including nCLE, NGS, and SOC to
improve the accuracy of risk stratification (HGD-Ca vs. LGD) of BD-IPMNs. Successful completion of this
project and application in clinical practice will provide a method for early detection of PDAC arising from PCLs,
guiding surgical decision-making to help avoid unwarranted resections or delayed treatment.
