EVAD-YOLO: An endoscopic video anomaly detection based on improved YOLOV11.

In recent years, automated analysis of gastrointestinal endoscopy videos has become increasingly important for early clinical screening as the incidence of gastrointestinal diseases continues to increase. However, the complex characteristics of gastrointestinal lesions pose significant challenges for accurate identification and diagnosis. This paper proposes an Endoscopic Video Anomaly Detection based on improved YOLOV11(EVAD-YOLO) for detecting typical lesions such as gastric ulcers and gastric cancer. Specifically, we construct a Residual global expansion attention (RGEA) module to enhance global contextual perception and improve sensitivity to lesions with complex shapes and color variations. In addition, we design an Enhanced multi-scale fusion (EMSF) module to effectively integrate lesion features across different spatial scales, thereby improving detection robustness for lesions of varying sizes. Furthermore, we built a mixed endoscopic dataset containing polyps, gastric ulcers, and early gastric cancers to comprehensively evaluate the proposed method. Experimental results demonstrate that EVAD-YOLO achieves superior performance, with 90.4% precision, 84.3% recall, and 90.4 % mAP50, indicating its strong robustness and potential for reliable clinical-assisted endoscopic diagnosis.