Towards Standardization of Retinal Vascular Measurements: On the Effect of Image Centering

Emanuele Trucco, Frsa, Fiapr, Fbmva, Faaia, Muthu Rama Krishnan Mookiah, Sarah McGrory, Stephen Hogg, Jackie Price, Rachel Forster, Thomas J. MacGillivray, Emanuele TruccoCOMPAY/OMIA @ MICCAI 2018 · DOI

TL;DR

This paper investigates how centering fundus images on either the optic disc (OD) or the macula affects quantitative measurements of retinal vasculature. The authors found significant variations in key parameters like vessel width, tortuosity, and fractal dimension depending on the centering method, highlighting a critical need for standardization in retinal biomarker studies to ensure reproducible and reliable results.

What Problem Does This Paper Address?

The human eye offers a unique, non-invasive window into the microvasculature. Fundus cameras are widely used to capture images of the retina, and various software packages exist to quantify vascular features. These quantitative measurements are increasingly used as biomarkers for a range of diseases, including diabetes, stroke, and cardiovascular conditions. However, the reliability of these biomarkers hinges on the accuracy and consistency of the measurements. This paper addresses a potential source of variability: the way fundus images are centered. Clinical protocols often require images to be centered on either the optic disc (OD) or the macula. The authors hypothesize that this difference in centering can lead to variations in the measured vascular parameters, potentially impacting the validity of biomarker studies. Specifically, macula-centered images may miss parts of the vasculature in nasal quadrants, which would be captured in OD-centered images, leading to different quantitative outcomes.

Methods

The study employed a quantitative analysis of retinal vascular measurements using the VAMPIRE software. The core of their method involved comparing measurements derived from two types of centered images: macula-centered (Type I) and optic disc-centered (Type II). They analyzed 149 different morphometric parameters, including vessel widths (like CRAE, CRVE, AVR), tortuosity measures, and fractal dimension (FD). To evaluate the agreement between measurements from the two centering types for the same parameter and eye, they used two-way mixed model intra-class coefficients (ICC). ICCs were chosen over simple correlation coefficients (like Pearson's r) because they are considered more appropriate for assessing agreement, as a perfect linear relationship doesn't necessarily imply good agreement. They also computed Pearson's correlation coefficients. Tortuosity measures were log-transformed to improve their distributions.

Datasets

The study utilized a dataset of 80 retinal fundus images from 20 subjects. For each subject, four images were acquired: one macula-centered and one optic disc-centered image for each eye (right and left). These images were sourced from the Edinburgh Type 2 Diabetes Study (ET2DS), a population-based cohort study. The images were acquired using a TOPCON TRC-50FX digital fundus camera with a 35° field-of-view. The authors noted that the images did not present diabetic lesions that would interfere with vasculature quantification, making them suitable for their analysis.

Key Insights and Ideas

  • Centering Matters: The most significant insight is that the choice of image centering (optic disc vs. macula) demonstrably impacts quantitative retinal vascular measurements. This is not a trivial difference; it can lead to substantial variations in key parameters.
  • Quadrant Exclusion: Macula-centered images often exclude nasal quadrants, which are visible in OD-centered images. This exclusion directly affects the completeness of the vascular network captured and, consequently, the measurements derived from it.
  • Variability Across Parameters: The impact of centering is not uniform across all parameters. Some width-related measures showed moderate agreement, while others, like fractal dimension, exhibited marked discrepancies.
  • Right-Left Symmetry Concerns: The study also touched upon right-left symmetry, suggesting that even with consistent centering, symmetry levels cannot be taken for granted, further complicating the interpretation of measurements.
  • Need for Standardization: The findings strongly underscore the necessity for international standardization in retinal imaging protocols and measurement techniques to ensure the reliability and reproducibility of biomarker studies.

Key Results

  • In the right eye, 5 out of 39 width-related measurements showed at least moderate correlation and association (ICC > 0.6) between OD- and macula-centered images. These included CRAE, CRVE, and AVR.
  • In the left eye, only 3 width-related measures met these criteria.
  • For tortuosity measures, only 17 out of 104 showed at least moderate correlation and ICC in the right eye, and 20 in the left. Only 10 satisfied these conditions in both eyes.
  • Fractal dimension (FD) measures showed a marked discrepancy, with venular measures missing altogether in the right eye due to the exclusion of macula-centered images.
  • For symmetry analysis, OD-centered images showed strong correlation and excellent agreement for CRAE and CRVE between right and left eyes, while macula-centered images showed poor correlation.

Limitations and Future Work

The authors acknowledge that the main limitation of their study is the modest number of images and subjects (80 images, 20 subjects). They plan to extend their analysis to larger samples from independent populations to better understand the effects of centering on morphometric vascular measurements. They also suggest that further research is needed to explore the statistical analysis methods that are most resilient to uncertainty and errors. Ultimately, they advocate for an international collaborative standardization effort for protocols used in reliable biomarker studies, with image centering being a crucial aspect.

Summary updated

BibTeX
@inproceedings{faaia2018towards,
  author = {Emanuele Trucco, FRSA, FIAPR, FBMVA, FAAIA and Muthu Rama Krishnan Mookiah and Sarah McGrory and Stephen Hogg and Jackie Price and Rachel Forster and Thomas J. MacGillivray and Emanuele Trucco},
  title = {{Towards Standardization of Retinal Vascular Measurements: On the Effect of Image Centering}},
  booktitle = {COMPAY/OMIA @ MICCAI},
  year = {2018},
  doi = {10.1007/978-3-030-00949-6\_35},
}
Figure