FunduSegmenter Adapts RETFound for Optic Disc and Cup Segmentation
FunduSegmenter is a newly developed model that modifies the RETFound foundation model to perform joint segmentation of the optic disc and optic cup in retinal fundus images. Originally designed for fundus camera and optical coherence tomography images, RETFound has demonstrated potential in diagnosing diseases. FunduSegmenter incorporates innovative components, such as a Pre-adapter, Decoder, Post-adapter, skip connections with a Convolutional Block Attention Module, and a Vision Transformer block adapter. The model was tested on a proprietary dataset (GoDARTS) and four public datasets (IDRiD, Drishti-GS, RIM-ONE-r3, REFUGE) through various verification methods. It recorded an average Dice similarity coefficient of 90.51% in internal verification, surpassing baseline models like nnU-Net (82.91%) and DUNet (89.17%). This marks the first use of RETFound for this specific segmentation task.
Key facts
- FunduSegmenter is the first adaptation of RETFound for joint OD and OC segmentation.
- RETFound is a foundation model for fundus camera and OCT images.
- FunduSegmenter integrates Pre-adapter, Decoder, Post-adapter, CBAM, and ViT block adapter.
- Evaluated on GoDARTS, IDRiD, Drishti-GS, RIM-ONE-r3, and REFUGE datasets.
- Achieved 90.51% average Dice coefficient in internal verification.
- Outperformed nnU-Net (82.91%) and DUNet (89.17%).
- Experiments included internal, external verification, and domain generalization.
Entities
—