Scholarlink

This paper presents a deep ensemble framework for cyst segmentation in optical coherence tomography (OCT) images, incorporating a majority voting mechanism. The proposed method integrates five state-of-the-art segmentation models—UNet, Swin-UNet, DeepLabV3+, Attention UNet, and SegResNet—to enhance accuracy and robustness across diverse cyst appearances. Experimental evaluation on a curated dataset demonstrates that the ensemble approach achieves a Dice score of 0.7967 on the validation set, outperforming individual models and illustrating its effectiveness in clinical scenarios.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep Ensemble with Voting Mechanism for OCT Cyst Segmentation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Rong Jiao,
  • Sijie Niu,
  • Haoyu Ding,
  • Yuhan Li,
  • Yutong Liu

摘要

This paper presents a deep ensemble framework for cyst segmentation in optical coherence tomography (OCT) images, incorporating a majority voting mechanism. The proposed method integrates five state-of-the-art segmentation models—UNet, Swin-UNet, DeepLabV3+, Attention UNet, and SegResNet—to enhance accuracy and robustness across diverse cyst appearances. Experimental evaluation on a curated dataset demonstrates that the ensemble approach achieves a Dice score of 0.7967 on the validation set, outperforming individual models and illustrating its effectiveness in clinical scenarios.