Endoscopic ultrasound-based artificial intelligence for gastrointestinal subepithelial lesions

Gastrointestinal subepithelial lesions (SELs), including gastrointestinal stromal tumors (GISTs), leiomyomas, and neuroendocrine tumors (NETs), are intramural lesions found beneath the mucosa of the gastrointestinal tract. These lesions can be benign, malignant, or potentially malignant. Therefore, accurately distinguishing between different types and assessing malignant potential is essential for establishing appropriate treatment plans and predicting patient outcomes. Endoscopic ultrasound (EUS) is recognized as the preferred method for diagnosing SELs. While certain lesions can be identified by their echogenic features alone, a histological examination is often necessary, potentially increasing patient risk. Artificial intelligence (AI) has demonstrated impressive capabilities in medical image recognition and offers significant potential for the non-invasive assessment of SELs. It achieves this by extracting key features from EUS images and applying them to classify or segment the images. This paper reviews recent advances in the application of AI to assist EUS in diagnosing SELs.