Artificial intelligence for automatic diagnosis of mucinous pancreatic cystic lesions in endoscopic ultrasound

Filipe Vilas-Boas, Tiago Ribeiro, Miguel Mascarenhas, João Afonso, Pedro Moutinho-Ribeiro, Susana Lopes, Guilherme Macedo, João Ferreira, Marco Parente, Renato Natal

Department of Gastroenterology – Centro Hospitalar de São João, Porto
Faculdade de Medicina do Porto
Department of Mechanical Engineering – Faculty of Engineering of the University of Porto
INEGI – Institute of Science and Innovation in Mechanical and Industrial Engineering

Patients with mucinous pancreatic cystic lesions (PCLs) have an increased risk of pancreatic malignancy compared with the general population. Endoscopic ultrasound is very useful for PCLs characterization but its accuracy to differentiate mucinous from non-mucinous lesions based on morphology is imperfect. Deep learning is an advanced form of artificial intelligence using convolutional neural networks (CNN) that can process complex images and provide high-performance predictions.
We aim to develop a CNN algorithm for the detection of mucinous pancreatic cysts using EUS images and to compare its diagnostic performance with the current standard of care based on clinical presentation, imaging, and cyst fluid analysis.