AUTHOR=Shi Xu , Wang Long , Li Yu , Wu Jian , Huang Hong 

TITLE=GCLDNet: Gastric cancer lesion detection network combining level feature aggregation and attention feature fusion

JOURNAL=Frontiers in Oncology

VOLUME=Volume 12 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2022.901475

DOI=10.3389/fonc.2022.901475

ISSN=2234-943X

ABSTRACT=(1) Background: Histopathological slices of gastric is the gold standard for diagnosing gastric cancer. While manual identification is time-consuming and highly relies on experience of pathologists. Artificial intelligence methods, particularly deep learning, can assist pathologists in finding cancerous tissues and realize automated detection. However, due to different shape and size of gastric cancer lesions and many interference factors, GCHIs has the problems of high complexity and difficult in accurately locating lesion area. Traditional deep learning methods cannot effectively extract discriminative features because of their simple decoding method so that they cannot detect lesions accurately, and there are fewer research dedicated to detecting gastric cancer lesions. (2) Methods: We propose a gastric cancer lesion detection network (GCLDNet). At first, GCLDNet designs a level feature aggregation structure in decoder, which can effectively fuse deep and shallow features of GCHIs. Second, an attention feature fusion module is introduced to accurately locate lesion area, which merges attention feature of different scales and obtains rich discriminative information focusing on lesion. Finally, focal tversky loss is employed as loss function to depress false negative predictions and mine difficult samples. (3) Results: Experimental results on two GCHI datasets of SEED and BOT show that DSCs of GCLDNet are 0.8265 and 0.8991, ACCs are 0.8827 and 0.8949, JIs are 0.7092 and 0.8182, and PREs are 0.7820 and 0.8763, respectively. (4) Conclusions: Experimental results demonstrates the effectiveness of GCLDNet in detection of gastric cancer lesions. Compared with other State-of-the-Art (SOTA) detection methods, GCLDNet obtains more satisfactory performance. This research can provide good auxiliary support for pathologists in clinical diagnosis.