[A multi-scale supervision and residual feedback optimization algorithm for improving optic chiasm and optic nerve segmentation accuracy in nasopharyngeal carcinoma CT images].

• DOI: 10.12122/j.issn.1673-4254.2025.03.21
• Publication Date: 2025-03-20
• Authors: Jinyu Liu; Shujun Liang; Yu Zhang
• Keywords: Optic chiasm; Optimization algorithm

We propose a novel deep learning segmentation algorithm (DSRF) based on multi-scale supervision and residual feedback strategy for precise of the optic chiasm nerves in CT images nasopharyngeal carcinoma (NPC) patients. collected 212 NPC their ground truth labels from SegRap2023, StructSeg2019 HaN-Seg2023 datasets. Based hybrid pooling strategy, we designed decoder (HPS) to reduce small organ feature loss during convolutional neural networks. This uses adaptive average refine high-level semantic features, which are integrated with primary features enable network finer details. employed layers learn rich multi-level under supervision, thereby enhancing boundary identification nerves. A module that enables multiple iterations was contrast enhancement by utilizing information fuzzy boundaries easily confused regions iteratively results supervision. The entire framework optimized each iteration enhance accuracy clarity. Ablation experiments comparative were conducted evaluate effectiveness component performance proposed model. DSRF could effectively representation organs achieve accurate an DSC 0.837 ASSD 0.351. further verified contributions method. can NPC.