FNeXter: A Multi-Scale Feature Fusion Network Based on ConvNeXt and Transformer for Retinal OCT Fluid Segmentation

• DOI: 10.3390/s24082425
• Publication Date: 2024-04-10T14:55:28Z
• Authors: Zhiyuan Niu; Zhuo Deng; Weihao Gao; Shurui Bai; Zheng Gong; Chucheng Chen; Fuju Rong; Fang Li; Lan Ma
• Keywords: 0301 basic medicine; Transformer; 0303 health sciences; optical coherence tomography; retinal fluid segmentation; Chemical technology; TP1-1185; Article; Retina; attention; Macular Degeneration; Image Processing, Computer-Assisted; Humans; Neural Networks, Computer; Tomography, Optical Coherence; Algorithms

The accurate segmentation and quantification of retinal fluid in Optical Coherence Tomography (OCT) images are crucial for the diagnosis treatment ophthalmic diseases such as age-related macular degeneration. However, is challenging due to significant variations size, position, shape fluid, well their complex, curved boundaries. To address these challenges, we propose a novel multi-scale feature fusion attention network (FNeXter), based on ConvNeXt Transformer, OCT segmentation. In FNeXter, introduce global hybrid encoder module that integrates ConvNeXt, region-aware spatial attention. This can capture long-range dependencies non-local similarities while also focusing local features. Moreover, this possesses capabilities, enabling it adaptively focus lesions regions. Additionally, self-adaptive enhance skip connections between decoder. inclusion elevates model’s capacity learn features contextual information effectively. Finally, conduct comprehensive experiments evaluate performance proposed FNeXter. Experimental results demonstrate our approach outperforms other state-of-the-art methods task