AE-UNet: a composite lung CT image segmentation framework using attention mechanism and edge detection

The primary impediments in lung CT image segmentation stem from the ambiguity in edge definition and the inadequate segmentation accuracy. Addressing these issues, this paper introduces a novel composite lung CT image segmentation framework that integrates an attention mechanism with an edge detection operator. We utilize residual dynamic convolutions as the encoder to augment the network's capability for extracting and representing nuanced lesion features. Sobel edge detection is integrated into the skip connections to facilitate the transmission and utilization of edge information. In particular, we introduce an information fusion attention module for deeper layers, optimizing feature reorganization and utilization by attention mechanisms and dilated convolution. Experimental evaluations on two lung CT datasets reveal that our proposed AE-UNet achieves outstanding segmentation performance, surpassing the best baseline network by an average of 0.93%.