Deep learning network with differentiable dynamic programming for retina OCT surface segmentation

• DOI: 10.1364/boe.492670
• Publication Date: 2023-05-29T02:00:08Z
• Authors: Hui Xie; Weiyu Xu; Ya Xing Wang; Xiaodong Wu
• Keywords: Feature (linguistics)

Multiple-surface segmentation in optical coherence tomography (OCT) images is a challenging problem, further complicated by the frequent presence of weak image boundaries. Recently, many deep learning-based methods have been developed for this task and yield remarkable performance. Unfortunately, due to scarcity training data medical imaging, it learning networks learn global structure target surfaces, including surface smoothness. To bridge gap, study proposes seamlessly unify U-Net feature with constrained differentiable dynamic programming module achieve end-to-end retina OCT explicitly enforce It effectively utilizes feedback from downstream model optimization guide learning, yielding better enforcement structures surfaces. Experiments on Duke AMD (age-related macular degeneration) JHU MS (multiple sclerosis) sets retinal layer demonstrated that proposed method was able subvoxel accuracy both datasets, mean absolute distance (MASD) errors 1.88 ± 1.96 μm 2.75 0.94 , respectively, over all segmented