Dual-Curriculum Model Improves Cross-Device Retinal Image Segmentation

A dual-curriculum deep learning model outperforms state-of-the-art approaches in segmenting retinal OCT and fundus images across unseen imaging devices, potentially reducing annotation costs.

• The model uses dual-curriculum learning and orthogonal domain prototypes to handle domain shifts from varying imaging devices.
• It was evaluated on two multi-domain OCT datasets and one color fundus photography dataset, outperforming existing methods in all cases.
• By generalizing across devices without device-specific annotations, the approach could reduce manual labeling costs and aid clinical deployment.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.