AI Model Outperforms Clinicians in Tumor Differentiation

An artificial intelligence model employing endoscopic ultrasonography (EUS) has demonstrated effective differentiation between small gastric stromal tumors (GSTs) and gastric leiomyomas (GLs). Utilizing a dataset of 358 EUS images, the ResNet50 architecture achieved impressive area under the curve (AUC) values of 0.994 and 0.911 for micro and small tumors respectively. In comparison, clinicians and clinical signatures yielded significantly lower AUCs, highlighting the model’s superior diagnostic capabilities and its potential as a pre-surgical decision-making tool.