CMOS: Confidence-Guided Multi-Scale Semi-Supervised Segmentation for Retinal Layers in OCT Images

Accurate segmentation of retinal layers in optical coherence tomography (OCT) images is crucial for the quantitative analysis and diagnosis of ophthalmic diseases. However, existing segmentation methods face two significant challenges. First, the scarcity of high-quality annotated data severely restricts the performance of supervised learning approaches. Second, structural alterations in retinal layers caused by lesions, combined with inconsistencies in imaging quality, further complicate the task of achieving precise segmentation. To address these challenges, we propose a novel Confidence-Guided Multi-Scale OCT Segmentation (CMOS) method, which leverages unlabeled data to significantly enhance the accuracy of OCT retinal layer segmentation while improving model robustness under complex pathological conditions. Specifically, our approach incorporates a bi-directional feature alignment strategy to refine the accuracy of pseudo-labels and a multi-scale aggregation (MSA) module to reduce feature variability and mitigate the impact of image quality fluctuations in lesion-affected regions. Experimental results demonstrate that our method surpasses existing semi-supervised learning approaches, achieving state-of-the-art performance.