A Novel Local-Global Spatial Attention Network for Cortical Cataract Classification in AS-OCT

Xiao，Zunjie1; Zhang，Xiaoqing1; Sun，Qingyang1; Wei，Zhuofei1; Xu，Gelei1; Jin，Yuan3; Higashita，Risa1,2; Liu，Jiang1,4

10.1007/978-3-031-18910-4_22

2022

0302-9743

1611-3349

LECT NOTES ARTIF INT

13535 LNCS

262-273

Cataracts are the leading causes of reversible blindness and visual impairment. Cortical cataract (CC) is one of the most common cataract types. Early intervention is a useful way to improve the vision of CC patients. Anterior segment optical coherence tomography (AS-OCT) is a new ophthalmic image type that can capture the cortex region clearly. To our best knowledge, no previous work has paid attention to AS-OCT image-based CC classification. Thus, this paper presents an effective local-global spatial fusion network (LGSF-Net) to automatically recognize the severity levels of CC. In the LGSF-Net, we construct a novel local-global spatial fusion (LGSF) attention block by incorporating the clinical prior knowledge of CC. The extensive experiments conduct on the CASIA2 AS-OCT dataset of CC, and the results show that our LGSF-Net achieves better performance than state-of-the-art attention-based convolutional neural networks (CNNs). Moreover, we also demonstrate the generalization of our method on the public ChestXRay2017 dataset through comparisons to existing methods.

AS-OCT Cortical cataract classification Local-global

第一 ; 通讯

英语

2-s2.0-85142746204

Scopus

1.Research Institute of Trustworthy Autonomous Systems and Department of Computer Science and Engineering,Southern University of Science and Technology,Shenzhen,518055,China
2.TOMEY Corporation,Nagoya,4510051,Japan
3.Zhongshan Ophthalmic Center,Sun Yat-sen University,Guangzhou,510060,China
4.Guangdong Provincial Key Laboratory of Brain-inspired Intelligent Computation,Southern University of Science and Technology,Shenzhen,518055,China

Xiao，Zunjie,Zhang，Xiaoqing,Sun，Qingyang,et al. A Novel Local-Global Spatial Attention Network for Cortical Cataract Classification in AS-OCT[C],2022:262-273.