文献速递眼底病与葡萄膜炎

基于深度学习的病理性近视后巩膜收缩术后视网膜劈裂与脱离体积自动测量

Automated deep learning-based retinoschisis and detachment volume measurement in pathological myopia with posterior scleral contraction.
PubMed 文献 · Br J Ophthalmol · 2026-06-30

本研究开发了一种基于深度学习的自动系统,用于量化接受后巩膜收缩术的病理性近视患者的视网膜劈裂与脱离体积(RDV)。研究纳入51例病理性近视患者(36例手术,15例对照),使用Mask R-CNN模型分割并计算RDV。结果显示模型交并比为80.9%,RDV随近视牵引性黄斑病变分期加重而增加,术后RDV显著下降,黄斑和中心凹恢复率分别为80.90%和99.50%,眼轴缩短与中心凹RDV变化相关性最强(r=0.479,p=0.003)。

要点

  • Mask R-CNN模型自动测量RDV的交并比达80.9%。
  • RDV随近视牵引性黄斑病变分期加重而增加,分布从中心凹旁向中心凹转移。
  • 术后眼轴显著缩短、BCVA显著改善,黄斑和中心凹RDV恢复率分别为80.90%和99.50%。

核心结论

  • 基于Mask R-CNN的RDV测量可作为评估后巩膜收缩术疗效和监测近视牵引性黄斑病变进展的敏感三维生物标志物。

研究设计与样本

  • 前瞻性研究纳入51例病理性近视患者,其中36例接受后巩膜收缩术,15例为对照。
  • 手术组按近视牵引性黄斑病变分期分为视网膜劈裂组(Stage 2,n=21)和视网膜脱离组(Stages 3-4,n=15),对照组为Stage 1(n=15)。
  • 使用扫频源OCT和血管成像获取视网膜数据,训练Mask R-CNN模型分割并计算RDV。
  • 主要结局指标为术后12个月区域黄斑恢复率,即RDV减少百分比。

主要发现

  • 模型交并比为80.9%。
  • 黄斑RDV随MTM分期增加而增加:对照组0.94 mm3,RS组5.24 mm3,RD组9.50 mm3(p<0.001)。
  • MTM分期越晚,RDV分布从中心凹旁/中心凹周围为主转向中心凹受累为主。
  • 术后眼轴显著缩短、BCVA显著改善(均p<0.001)。
  • 术后RDV显著下降,黄斑和中心凹恢复率分别为80.90%和99.50%;眼轴缩短与中心凹RDV变化相关性最强(r=0.479,p=0.003)。

临床怎么用

  • 该自动RDV测量系统可提供客观、高精度的视网膜结构变化量化,弥补手动OCT分析的局限性。
  • RDV可作为评估后巩膜收缩术疗效和监测MTM进展的敏感三维生物标志物。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

原文

PURPOSE: This study developed an automated deep learning-based system to quantify retinoschisis and detachment volume (RDV) in pathological myopia (PM) patients undergoing posterior scleral contraction (PSC). METHODS: A prospective study included 51 PM patients (36 PSC-treated and 15 controls). Based on the myopic traction maculopathy staging system classification, PSC-treated patients were divided into retinoschisis (RS, Stage 2, n=21) and retinal detachment (RD, Stages 3-4, n=15), while controls were classified as Stage 1 (n=15). Swept-source optical coherence tomography (OCT) and angiography imaging was used to acquire retinal data, and a Mask R-CNN model was trained to segment and calculate RDV. The primary outcome measures were regional macular recovery rate, defined as the percentage reduction in RDV at 12 months postoperatively. Clinical parameters such as axial length (AL) and best-corrected visual acuity (BCVA) were analysed. RESULT: The model achieved an Intersection over Union of 80.9%. RDV increased with more advanced stages of myopic traction maculopathy (MTM). As MTM stage advanced, RDV distribution shifted from predominantly parafoveal/perifoveal involvement to foveal involvement. Macular RDV increased progressively with MTM stage (control: 0.94 mm3, RS: 5.24 mm3, RD: 9.50 mm3; p<0.001). Postoperative AL decreased significantly and BCVA improved significantly (all p<0.001). Postoperative RDV decreased significantly, achieving macular and foveal recovery rates as 80.90% and 99.50%, respectively. AL reduction demonstrated the strongest correlation with foveal RDV changes (r=0.479, p=0.003). CONCLUSION: The Mask R-CNN-based RDV measurement provides a sensitive 3D biomarker for assessing PSC efficacy and monitoring MTM progression. The system offers objective, high-precision quantification of retinal structural changes, addressing limitations of manual OCT analysis. TRIAL REGISTRATION NUMBER: ChiCTR2200065561.

来源:Br J Ophthalmol,查看原文。版权归原作者所有。

引用

  1. Chai M, Rong H, Yu D, et al. Automated deep learning-based retinoschisis and detachment volume measurement in pathological myopia with posterior scleral contraction. Br J Ophthalmol. 2026. PMID: 42379831. doi:10.1136/bjo-2025-328592

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。