We read with great interest the article on GastroNet-5M, a foundation model for endoscopic artificial intelligence (AI) applications.1 Although computer-aided detection (CADe) systems built on convolutional neural network (CNN) architecture have improved adenoma detection rates, they remain inadequate for the detection of sessile serrated lesions (SSLs). We believe that moving from task-specific CNNs to endoscopy-trained foundation models (including those built on GastroNet-5M and the Meta Segment-Anything framework [Meta, Menlo Park, CA]) can address the systematic under detection of SSLs.
Gastroenterology (AGA) published a clinical update in Research Highlights on 21 Feb 2026.
The item focuses on From Narrow Convolutional Neural Networks to Endoscopic Foundation Models: Implications for Sessile Serrated Lesion Detection.
Review the original article for the full source wording and details.