上海交通大学学报(医学版) ›› 2019, Vol. 39 ›› Issue (6): 613-.doi: 10.3969/j.issn.1674-8115.2019.06.009

• 论著·临床研究 • 上一篇    下一篇

视网膜血管性疾病光学相干断层成像图像的自动分层研究

许毓鹏 1*,杜宇辰 1, 2*,陈凤娥 1   

  1. 1. 上海交通大学附属第一人民医院眼科,上海市眼底病重点实验室,上海市眼视觉及光医学工程研究中心,上海 200080 ;2. 上海交通大学电子信息与电气工程学院自动化系,上海 200240
  • 出版日期:2019-06-28 发布日期:2019-07-26
  • 通讯作者: 陈凤娥,电子信箱:chen_feng_e@hotmail.com。
  • 作者简介:许毓鹏(1991—),男,住院医师,博士;电子信箱: kevinxyp@163.com。杜宇辰(1991—),男,博士;电子信箱: yuchendu@sjtu.edu.cn。*为共同第一作者。
  • 基金资助:
    国家重点研发计划精准研究专项(2016YFC0904800);上海市青年科技英才扬帆计划(19YF1439300);上海交通大学“转化医学交叉研究基金”项目(ZH2018QNA24)

Automatic layer segmentation of optical coherence tomography images in retinal vascular diseases

XU Yu-peng1*, DU Yu-chen1, 2*, CHEN Feng-e1   

  1. 1. Shanghai Engineering Center for Visual Science and Photomedicine, Shanghai Key Laboratory of Fundus Disease, Department of Ophthalmology, Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai 200080, China; 2. Department of Automation, School of Electronic Information and Electrical Engineering, Shanghai Jiao Tong University, Shanghai 200240, China
  • Online:2019-06-28 Published:2019-07-26
  • Supported by:
    National Key R&D Plan on Precision Medicine, 2016YFC0904800; Shanghai Sailing Program, 19YF1439300; “Translation Medicine Cross Research Fund” Project of Shanghai Jiao Tong University, ZH2018QNA24)。

摘要: 目的 ·使用无监督机器学习的方法对视网膜血管性疾病的光学相干断层成像(optical coherence tomography,OCT)图像进行分层研究,并与机器内置分层方法比较。方法 ·标准化采集 50例病理性近视脉络膜新生血管(myopic choroidal neovascularization, mCNV)患者以及 20例糖尿病性黄斑水肿(diabetic macular edema,DME)患者的 OCT图像资料。由专业医师手工标记分层信息建立标准。提出一种基于层间能量最小化的视网膜多层分割方法,将其结果与 OCT机器内置分层方法的结果进行比较。利用无标记边界位置误差验证分层方法的准确性。结果 ·基于层间能量最小化的分层方法将每位患者的视网膜分为 5层:内界膜层、神经纤维层下层、外核层上层、椭圆体带上层和 Bruchs层。该方法在整体数据集的平均无标记边界位置误差为(4.831±7.015)μm。在 mCNV组视网膜各层的平均无标记边界位置误差为(4.839±16.819)μm,而在 DME组视网膜各层的平均无标记边界位置误差为(5.048±9.986)μm,均远低于 OCT机器内置分层方法的结果 [mCNV组为(13.638±58.024)μm,DME组为(14.796±45.342)μm]。该方法在视网膜各层分层的准确性均高于 OCT机器内置方法。结论 ·基于层间能量最小化的分层方法可用于不同类型视网膜血管性疾病 OCT图像的分层,结果明显优于机器内置分层方法,可拓展用于其他类型视网膜血管性疾病的分层。

关键词: 视网膜血管性疾病, 光学相干断层成像, 自动分析算法, 机器学习, 计算机视觉

Abstract:

Objective · To explore the layer segmentation method of optical coherence tomography (OCT) images of retinal vascular diseases using an unsupervised learning method, and compare it with the built-in layering method of OCT machine. Methods · Standardized image acquisition was performed on OCT images 50 patients with myopic choroidal neovascularization (mCNV) and 20 patients with diabetic macular edema (DME). Standards were establishedmanual marking of hierarchical informationprofessional physicians. A retinal multi-layer segmentation method based on the minimization of interlayer energy was proposed, and the results were compared with those obtainedthe built-in layering method of OCT machine. The layering accuracy was verifiedthe unmarked boundary position error. Results · This segmentation method divided the retina of each patient into five layers: internal limiting membrane, lower layer of nerve fiber layer, upper layer of outer nuclear layer, upper layer of ellipsoid zone and Bruchs membrane. The average segmentation error in the overall data set was (4.831±7.015) μm. The error of mCNV group and DME group were (4.839±16.819) μm and (5.048±9.986) μm, respectively, both of which were lower than the automatic measurement results of OCT machine [(13.638±58.024) μm and (14.796±45.342) μm, respectively]. The accuracy of this method at each layer was higher than that of the automatic measurement. Conclusion · This multi-layer segmentation method can be used for segmentation of different types of retinal vascular diseases, and the results are significantly better than those obtainedthe built-in method in OCT machine. It can be extended for layer segmentation of other retinal vascular diseases.

Key words: retinal vascular diseases, optical coherence tomography (OCT), automatic analysis algorithm, machine learning, computer vision

中图分类号: