Accurate diagnosis of early gastric cancer is valuable for asymptomatic populations, while current endoscopic examination combined with pathological tissue biopsy often encounters bottlenecks for early-stage cancer and causes pain to patients. Liquid biopsy shows promise for noninvasive diagnosis of early gastric cancer
however, it remains a challenge to achieve accurate diagnosis due to the lack of highly sensitive and specific biomarkers. Herein, we propose a protocol combining metabolomics profiling from plasma extracellular vesicles (EVs) and machine learning to identify the metabolomics discrepancies of early gastric cancer individuals from other populations. Efficient metabolomics profiling is achieved by efficient, high-purity, and damage-free plasma EVs separation using elaborately designed nanotrap-structured microparticles (NanoFisher) by taking advantage of stereoscopic interaction and affinity interaction. Significant metabolomics discrepancies are obtained from 150 early gastric cancer (50), benign gastric disease (50), and non-disease control (50) plasma samples. Machine learning enables ideal distinction between early gastric cancer and non-disease control samples with an area under the curve (AUC) of 1.000, achieves an AUC of 0.875-0.975 for differentiating early gastric cancer from benign gastric diseases, and demonstrates an overall accuracy of 92% in directly classifying these three categories. The plasma EV metabolomics profiling enabled by NanoFisher materials, integrated with machine learning, holds considerable promise for broad clinical acceptance, enhancing gastric cancer screening outcomes.