Non-physician grader reliability in measuring morphological features of the optic nerve head in stereo digital images

Eye (Lond). 2019 May;33(5):838-844. doi: 10.1038/s41433-018-0332-8. Epub 2019 Jan 11.

Abstract

Objective: To introduce a new method of grading optic nerve stereo disc photographs and evaluate reproducibility of assessments by non-physician graders in a reading center.

Methods: Three non-physician graders, experienced in grading features of the retina but not the optic nerve head (ONH), were trained by glaucoma specialists to assess digital stereo color images of the ONH. These graders assessed a total of 2554 digital stereo disc images from glaucoma cases and controls participating in the Primary Open-Angle African American Glaucoma Genetics (POAAGG) study by outlining the optic cup and disc. Inter-grader reproducibility of area, height, and width measurements was analyzed.

Results: Among all images, the intraclass correlation (95% confidence interval) was 0.90 (0.89, 0.90) for the cup area using only color cues; 0.92 (0.91, 0.92) for the cup area using contour and vascular cues; and 0.99 (0.99, 0.99) for the optic disc area. The intraclass correlation for cup-to-disc ratio (CDR) was 0.61 (0.58, 0.63), as determined by the ratio of optic cup area to optic disc area (using contour and vascular cues). The CDR difference by graders for area was ≤ 0.1 in 65% of images using color/vascular cues and ≤0.1 in 71% of images using color cues.

Conclusions: After adequate training, non-physician graders were able to measure the optic nerve CDR with high inter-grader reliability.

MeSH terms

  • Adult
  • Allied Health Personnel / statistics & numerical data*
  • Black or African American / genetics
  • Female
  • Glaucoma, Open-Angle / diagnosis*
  • Glaucoma, Open-Angle / genetics
  • Humans
  • Image Processing, Computer-Assisted / methods*
  • Male
  • Middle Aged
  • Optic Disk / pathology*
  • Optic Nerve Diseases / diagnosis*
  • Optic Nerve Diseases / genetics
  • Photography / methods*
  • Reproducibility of Results