Scholarlink

Background <p>Peripapillary hyperreflective ovoid mass-like structures (PHOMS) are emerging optical coherence tomography (OCT) markers for axoplasmic stasis. Accurate volumetric quantification is critical for monitoring their progression but remains challenging due to the complex 3D geometry of these structures.</p> Methods <p>We developed a dual-branch deep convolutional neural network to automatically segment and quantify PHOMS volumes using swept-source OCT data from 70 adolescent eyes. The architecture integrates a segmentation branch and a classification branch to resolve boundary ambiguities. Performance was validated against manual ground truth from expert graders.</p> Results <p>Our automated segmentation achieved a human-equivalent accuracy level in identifying PHOMS, achieving a mean Dice coefficient of 0.810 and an intraclass correlation coefficient of 0.990 in comparison with expert graders’ assessments. Volumetric measurements demonstrated a robust correlation (<i>r</i> = 0.992) between the automated and manual methods.</p> Conclusions <p>The proposed deep learning framework enables rapid, reliable, and automated 3D quantification of PHOMS in adolescents. This tool holds significant potential for large-scale clinical screening and the longitudinal monitoring of PHOMS.</p>

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

Automated segmentation and quantification of peripapillary hyperreflective ovoid mass-like structures using swept-source optical coherence tomography

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月28日

查看全文
  • Zhuo Huang,
  • Di Xiao,
  • Fangyuan Zhou,
  • Changzheng Chen,
  • Yishuang Xu,
  • Dihao Hua

摘要

Background

Peripapillary hyperreflective ovoid mass-like structures (PHOMS) are emerging optical coherence tomography (OCT) markers for axoplasmic stasis. Accurate volumetric quantification is critical for monitoring their progression but remains challenging due to the complex 3D geometry of these structures.

Methods

We developed a dual-branch deep convolutional neural network to automatically segment and quantify PHOMS volumes using swept-source OCT data from 70 adolescent eyes. The architecture integrates a segmentation branch and a classification branch to resolve boundary ambiguities. Performance was validated against manual ground truth from expert graders.

Results

Our automated segmentation achieved a human-equivalent accuracy level in identifying PHOMS, achieving a mean Dice coefficient of 0.810 and an intraclass correlation coefficient of 0.990 in comparison with expert graders’ assessments. Volumetric measurements demonstrated a robust correlation (r = 0.992) between the automated and manual methods.

Conclusions

The proposed deep learning framework enables rapid, reliable, and automated 3D quantification of PHOMS in adolescents. This tool holds significant potential for large-scale clinical screening and the longitudinal monitoring of PHOMS.