Objective:To systematically evaluate the intelligence-assisted endoscopic diagnosis system based on deep learning (DL-IEDS) for early cancer of the upper digestive tract.Methods:Literature on the value of DL-IEDS for diagnosis of early cancer of the upper digestive tract was searched in English (PubMed, Embase, Web of Science and Cochrane Library)and Chinese databases (Sinomed, CNKI, Wanfang and VIP). The quality of literatures was evaluated according to Quality Assessment of Diagnostic Accuracy Studies-2. The Rev Man 5.3, Meta-Disc 1.4 and Stata 15.1 were used for the meta-analysis.Results:Eight studies were included with a total of 9 675 images (including 2 748 images of early cancer). Meta-analysis results showed that the pooled sensitivity, specificity, positive likelihood ratio, negative likelihood ratio and comprehensive diagnostic ratio of DL-IEDS in the diagnosis of early cancer of the upper digestive tract were 0.920, 0.874, 6.824, 0.103 and 71.109, respectively. The area under the curve (AUC) of summary receiver operating characteristics was 0.958 7. Five studies reported the results of DL-IEDS in the diagnosis of early gastric cancer, and the combined analysis showed that the pooled sensitivity and specificity were 0.840 and 0.845 respectively, and the AUC was 0.919. Four studies reported the accuracy rate of endoscopic experts and endoscopic novices in diagnosing early upper gastrointestinal cancer, and results showed that the pooled sensitivity, specificity and AUC were 0.693, 0.892 and 0.892 3, and 0.586, 0.860 and 0.754 5, respectively. Compared with endoscopy experts, the AUC of DL-IEDS in diagnosis of early upper gastrointestinal cancer showed no statistically significant difference ( Z=1.510, P=0.131), while compared with endoscopy novices, the difference was statistically significant ( Z=6.841, P<0.001). Conclusion:The DL-IEDS has high diagnostic accuracy for early upper digestive tract cancer, and can significantly improve the diagnostic ability of endoscopy novices.