Abstract:

Gastric carcinoma （GC）， a highly prevalent malignant tumor globally， often progresses to advanced stages by the time of diagnosis due to its insidious clinical presentation， thereby significantly reducing therapeutic effectiveness and patient quality of life. Accurate screening and histopathological characterization of early gastric cancer （EGC） are essential for developing individualized treatment approaches. Although endoscopic techniques remain the gold standard for early GC detection， their diagnostic accuracy is largely dependent on the operator′s skill， a challenge that current artificial intelligence （AI）-assisted innovations aim to address by standardizing diagnostic procedures. Deep learning （DL）-based computer vision systems have demonstrated remarkable performance in identifying subtle EGC features， not only improving lesion detection sensitivity but also enabling automated assessment of key pathological indicators. These technological advances offer objective， visualized diagnostic support for clinical decision-making. This review provides a systematic overview of recent developments in DL applications for endoscopic image analysis of EGC and evaluates their potential for clinical integration.

Key words: early gastric cancer, deep learning, endoscopic images