Related Experiment Video
Updated: May 8, 2026
10:10
Ultrahigh Resolution Mouse Optical Coherence Tomography to Aid Intraocular Injection in Retinal Gene Therapy Research
Published on: November 2, 2018
9.3K
FNeXter: A Multi-Scale Feature Fusion Network Based on ConvNeXt and Transformer for Retinal OCT Fluid Segmentation.
Zhiyuan Niu1, Zhuo Deng1, Weihao Gao1
1Tsinghua Shenzhen International Graduate School, Tsinghua University, Shenzhen 518055, China.
Sensors (Basel, Switzerland)
|April 27, 2024
Summary
Accurate segmentation of retinal fluid in Optical Coherence Tomography (OCT) images is vital for diagnosing eye diseases. A new FNeXter network improves OCT fluid segmentation by effectively capturing multi-scale features and lesion regions.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Accurate segmentation of retinal fluid in Optical Coherence Tomography (OCT) images is critical for diagnosing and treating ophthalmic conditions like age-related macular degeneration.
- Challenges in segmentation include variations in fluid size, shape, position, and complex boundaries.
Purpose of the Study:
- To propose a novel multi-scale feature fusion attention network (FNeXter) for improved OCT fluid segmentation.
- To enhance the model's ability to capture both long-range dependencies and local features, with region-aware capabilities.
Main Methods:
- Developed FNeXter, a network integrating ConvNeXt, Transformer, and region-aware spatial attention in a global multi-scale hybrid encoder.
- Introduced a self-adaptive multi-scale feature fusion attention module to enhance encoder-decoder skip connections.
Main Results:
- The FNeXter model demonstrated superior performance in OCT fluid segmentation tasks.
- Comprehensive experiments confirmed the effectiveness of the proposed approach compared to state-of-the-art methods.
Conclusions:
- The FNeXter network effectively addresses the challenges of retinal fluid segmentation in OCT images.
- The proposed architecture enhances the learning of global features and multi-scale contextual information, leading to improved diagnostic capabilities.
