An Efficient Honeycomb Lung Segmentation Network Combining Multi-Paradigms Representation and Cascade Attention

• DOI: 10.14569/ijacsa.2023.0141256
• Publication Date: 2023-12-29T14:08:21Z
• Authors: Bingqian Yang; Xiufang Feng; Yunyun Dong
• Keywords: Representation; Market Segmentation

Honeycomb lung is a pulmonary manifestation that occurs in the terminal stage of various diseases, which greatly threatens patients. Due to different locations and irregular shapes lesions, accurate segmentation honeycomb region an essential challenging problem. However, most deep learning methods struggle effectively utilize both global local information from lesion images, resulting cannot accurately segment lesion. In addition, these often ignore some semantic necessary for location shape decoding stage. To alleviate challenges, this paper, we propose dual-branch encoder cascaded decoder network (DECDNet) segmenting honeycombs lesions. First, design consisting ResNet34 Swin-Transformer with paradigm representations extract features long-range dependencies respectively. Next, further combine features, develop feature fusion module obtain richer representation information. Finally, considering problem loss during decoder, attention constructed aggregate multi-stage get final result. Experimental results demonstrate our method outperforms other on in-house dataset. Notably, compared nine universal methods, proposed DECDNet obtains highest IoU (86.34%), Dice (92.66%), Precision (93.21%), Recall (92.13%), F1-Score achieves lowest HD95 (7.33) ASD (2.30). particular, enables precisely lesions under clinical scenarios as well. Our code dataset are available at https://github.com/ybq17/DECDNet.