FunduSAM: A Specialized Deep Learning Model for Enhanced Optic Disc and Cup Segmentation in Fundus Images

• DOI: 10.48550/arxiv.2502.06220
• Publication Date: 2025-02-10
• Authors: J. M. Yu; Yongwei Nie; Fei Qi; Wenxiong Liao; Hongmin Cai
• Keywords: Fundus (uterus); Optic cup (embryology); Optic disc

The Segment Anything Model (SAM) has gained popularity as a versatile image segmentation method, thanks to its strong generalization capabilities across various domains. However, when applied optic disc (OD) and cup (OC) tasks, SAM encounters challenges due the complex structures, low contrast, blurred boundaries typical of fundus images, leading suboptimal performance. To overcome these challenges, we introduce novel model, FunduSAM, which incorporates several Adapters into create deep network specifically designed for OD OC segmentation. FunduSAM utilizes Adapter each transformer block after encoder parameter fine-tuning (PEFT). It enhances SAM's feature extraction by designing Convolutional Block Attention Module (CBAM), addressing issues related contrast. Given unique requirements segmentation, polar transformation is used convert original images format better suited training evaluating FunduSAM. A joint loss achieve structure preservation between OC, while accurate Extensive experiments on REFUGE dataset, comprising 1,200 demonstrate superior performance compared five mainstream approaches.