.png){kind=icon}
.png){kind=icon}
Return
Optimizing optic cup and optic disc delineation: Introducing the efficient feature preservation segmentation network
.png){kind=icon}
PRE
AI
DOI：10.1016/j.engappai.2025.110038.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Recently, computer vision and healthcare had significantly enhanced glaucoma diagnosis, necessitating automated solutions due to the rising prevalence and subjective nature of current diagnostic methods. Optic cup (OC) and optic disc (OD) segmentation in retinal fundus imaging is crucial, yet challenging due to non-distinctive OC boundaries and image property variations. Ongoing efforts to develop automated diagnostic systems aim to enhance glaucoma detection. Accurate OC and OD segmentation has great potential for providing valuable clinical insights and validating glaucoma screening effectiveness. Retinal fundus images exhibit wide variations in dimensions, shape, color, and background, posing segmentation challenges. Consequently, we propose the efficient feature preservation segmentation network (EFPS-Net) to accurately segment OC and OD. EFPS-Net incorporates an atrous residual block for capturing features and facilitating computational efficiency on the encoder side, along with a recurrent residual block on the decoder side to prevent information loss. Additionally, intrinsic feature fusion and extrinsic feature aggregation were incorporated to improve the model's learning ability and efficient feature preservation. The proposed network achieves high segmentation performance even with only 2.6 million parameters. Our model was evaluated on five publicly available retinal fundus image datasets and outperformed state-of-the-art models. These results highlight EFPS-Net's advancements in empowering healthcare professionals in glaucoma diagnosis and treatment by leveraging advanced artificial intelligence techniques with engineering innovations.
Keywords:
Deep learning
Optic cup and optic disc segmentation
Retinal fundus images
Glaucoma screening
Medical image analysis
Journal
IF:
8
Papers:
5.3K
Citations:
3.5W