Chinese Optics Letters, Volume. 17, Issue 1, 011701(2019)

Automated segmentation of optical coherence tomography images

C. Kharmyssov1, M. W. L. Ko2、*, and J. R. Kim1
Author Affiliations
  • 1School of Engineering, Nazarbayev University, Astana 010000, Kazakhstan
  • 2The University of Hong Kong, Pokfulam, Hong Kong, China
  • show less
    Figures & Tables(9)
    SDOCT image with retinal pigment epithelium layers (RPE, solid red line), internal limiting membrane (ILM, solid green line), and choroid representation.
    Overview of automatic segmentation of the ILM and RPE layers in SDOCT images.
    Example of the ILM layer (blue line) and RPE layer (red line) segmentation with the solely intensity-based algorithm.
    Segmentation result of RPE (red line) and ILM (blue line) of the hybrid algorithm, which is solely a combination of intensity and graph-based algorithms after filling RPE gaps with the graph-based approach.
    Depiction of the ONHSD, which represents an average of perpendicular distances from a red line joining two BMO points to the ILM layer.
    Bland–Altman plot of the ONHSD measurement between the proposed hybrid algorithm and manual segmentation.
    Illustration of the (a) raw image of ONH from SDOCT of a primary open-angle glaucoma patient and (b) automatically segmented image with the hybrid algorithm of ILM (blue line) and RPE (red line) layers.
    • Table 1. Summary of the Processing Time of One Image, Mean and Standard Deviation of ILM and RPE Comparison of Relative Difference with Manually Segmented Image, and Dice’s Coefficients of 120 Images

      View table
      View in Article

      Table 1. Summary of the Processing Time of One Image, Mean and Standard Deviation of ILM and RPE Comparison of Relative Difference with Manually Segmented Image, and Dice’s Coefficients of 120 Images

      AlgorithmProcessing time (s)AMRSD (µm)Dice’s coefficient (%)
      Intensity thresholding algorithm3.75.42±0.0396.8±1.7
      Graph-based algorithm34.324.7±0.2374.1±14.8
      Proposed hybrid algorithm9.35.73±0.0396.6±1.6
    • Table 2. Summary of RSPP Comparison for RPE and ILM Layers of 120 Images

      View table
      View in Article

      Table 2. Summary of RSPP Comparison for RPE and ILM Layers of 120 Images

      AlgorithmRPEaILMa
      Intensity thresholding algorithm52.2±18.584.6±12.8
      Graph-based algorithm76.9±15.976.9±12.4
      Proposed hybrid algorithm88.3±9.588.5±9.8
    Tools

    Get Citation

    Copy Citation Text

    C. Kharmyssov, M. W. L. Ko, J. R. Kim, "Automated segmentation of optical coherence tomography images," Chin.Opt.Lett. 17, 011701 (2019)

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Medical optics and biotechnology

    Received: Aug. 10, 2018

    Accepted: Nov. 22, 2018

    Posted: Nov. 22, 2018

    Published Online: Jan. 17, 2019

    The Author Email: M. W. L. Ko (matchkoust@gmail.com)

    DOI:10.3788/COL201917.011701

    Topics