Early Detection of Pancreatic Cancer

Sushil Kumar Garg; Suresh T. Chari


Curr Opin Gastroenterol. 2020;36(5):456-461. 

In This Article

Finding Early Lesion in Pancreatic Cyst

Currently, there is no perfect way to differentiate between high risks versus low-risk lesions. Unnecessary surgery has significant operative mortality (2-4%) and morbidity (40–50%).[49] Several guidelines have been proposed by different medical societies[50–53] for surveillance of pancreatic cyst, but the management of pancreatic cyst remains challenging. EUS with FNA and pancreatic cyst fluid (PCF) analysis is the current modality to assess worrisome features including mural nodules, thick internal septations, mural nodularity, solid masses, and vascular invasion.[54] EUS-guided confocal laser endomicroscopy has shown great potential in differentiating mucinous from the nonmucinous pancreatic cyst with 98% sensitivity, 94% specificity, and 97% accuracy.[55] Another study using DNA-based testing of PCF was able to differentiate mucinous cyst with 89% sensitivity and 100% specificity by identifications of mutations in KRAS and GNAS.[56] The presence of mutations in PIK3CA, SMAD4, and TP53 can differentiate low-grade dysplasia versus high-grade dysplasia and cancer in 80% of patients of IPMN.[56] Patients with IPMN with low-grade dysplasia tend to a high prevalence of heterogeneity mutations in KRAS and GNAS compared with patients with high-grade dysplasia.[57]

There is increasing use of machine learning and artificial intelligence in differentiating low-risk versus high-risk pancreatic cyst lesions. Springer et al.[58] constructed a diagnostic algorithm with the help of machine learning using clinical features, imaging characteristics, and cyst fluid genetic and biochemical markers. This machine learning-based algorithm has sensitivity, specificity, and accuracy of 95.7, 91.9, and 92.9% and would have decreased the number of unnecessary surgeries by 60–74%. In another study using deep learning algorithm and convolutional neural network on EUS images which were taken before resection of pancreases for IPMN was able to predict high-grade dysplasia and invasive carcinoma with sensitivity, specificity, and accuracy of 95.7, 92.6, and 94.0% and AUC was 0.99.[59]