Several attention modules—such as SENet, CBAM, and SimAM—have been successfully applied in image classification tasks and could be integrated into object detection frameworks like YOLOv5, YOLOv7, and YOLOv9. However, the optimal insertion point within these detection architectures—whether in the backbone, neck, or head—remains an open question. In this study, we systematically investigate the effects of incorporating attention modules at various network locations. Experiments conducted on a regurgitation dataset of echocardiography images demonstrate that strategically inserting attention modules significantly improves performance, as measured by the mAP50 metric. Notably, the CBAM module proves particularly effective for the task at hand.
