MLDA-Unet:Multi Level Dual Attention Unet for Polyp Segmentation

In recent years, the attention mechanism has received much attention and application in medical image segmentation. Moreover, many studies have obtained specific results by applying different attention mechanisms in various medical image segmentation tasks. However, many studies have applied the attention mechanism to the segmentation model without further discussion. In this paper, we focus on applying the dual attention mechanism and propose a model called MLDA-Unet, which applies the dual attention mechanism block to multiple layers of Unet. Its performance is 9.15%, 6.09%, and 9.47% higher than the normal DA-Unet, Unet, and Unet++, where DA Unet means that a dual attention mechanism block is used at the bottom of Unet. We first review some studies that apply attention to the medical field, and then we discuss dual attention by setting up rich control group experiments. Finally, we analyze and discuss the results, proving that the multi-level DA-Unet can better focus on target features, which could improve the model's performance.