AUTHOR=Peng Yuanyuan , Zhu Weifang , Chen Zhongyue , Shi Fei , Wang Meng , Zhou Yi , Wang Lianyu , Shen Yuhe , Xiang Daoman , Chen Feng , Chen Xinjian 

TITLE=AFENet: Attention Fusion Enhancement Network for Optic Disc Segmentation of Premature Infants

JOURNAL=Frontiers in Neuroscience

VOLUME=Volume 16 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2022.836327

DOI=10.3389/fnins.2022.836327

ISSN=1662-453X

ABSTRACT=Retinopathy of prematurity (ROP) and ischemic brain injury resulting in periventricular white matter (PVWM) damage are the main causes of visual impairment in preterm infants. Accurate optic disc (OD) segmentation has important prognostic significance for the auxiliary diagnosis of the above two diseases of preterm infants. Due to the complexity and non-uniform illumination of the fundus images, the segmentation of OD for infants is challenging and rarely reported in the literature. In this paper, we propose a novel attention fusion enhancement Network (AFENet) for the accurate segmentation of OD in the fundus images of premature infants by fusing adjacent high-level semantic information and multi-scale low-level detailed information from different levels based on encoder-decoder network. Specifically, we firstly design a dual-scale semantic enhancement (DsSE) module between the encoder and the decoder based on self-attention mechanism, which can enhance the semantic contextual information for the decoder by reconstructing skip-connection. Then, to reduce the semantic gaps between the high-level and low-level features, a multi-scale features fusion (MsFF) module is developed to fuse multiple features of different levels at the top of encoder by using attention mechanism. Finally, the proposed AFENet was evaluated on the fundus images of premature infants for OD segmentation, which shows that the proposed two modules are both promising. Based on the baseline (Res34UNet), using DsSE or MsFF module alone can increase Dice similarity coefficients (Dsc) by 1.51% and 1.70%, respectively. While the integration of two modules together can rise 2.11%. Compared with other state-of-the-art segmentation methods, the proposed AFENet achieves high segmentation performance.