Computer-based image analysis for ROP: development of a quantitative index based on vascular tortuosity

Abstract

Purpose: Plus disease is a critical marker of treatment-requiring retinopathy of prematurity (ROP), but clinical diagnosis has been shown to be subjective and qualitative. Computer-based image analysis of vascular tortuosity has potential to provide diagnostic assistance based on objective and quantitative principles. This pilot study proposes and evaluates a methodology for creating a quantitative “plus index” for diagnostic assistance.

Methods: A set of wide-angle retinal images (RetCam; Clarity Medical Systems, Pleasanton, CA) from infants with ROP was collected and diagnosed by experts as plus or not plus. 41 images were reviewed by 23 experts. 384 retinal vessels (both arterioles and venules) from all images were manually segmented by author consensus. Tortuosity, defined as vessel length from starting point to end point divided by the straight line distance between those points, was calculated using a computer-based image analysis system developed by the authors. Mean tortuosity of all vessels was calculated for each image at various distances from the optic disc. For each image, correlations between mean tortuosity and proportion of experts who diagnosed it as plus disease were calculated at each distance to determine which distance resulted in strongest correlation. The full range of tortuosity at that distance was divided into a scale from 1 to 7 to create a "plus index". Sensitivity and specificity for detecting plus disease were calculated at each "plus index" value using the expert majority diagnosis as the reference standard.

Results: The strongest correlation between mean tortuosity and proportion of experts who diagnosed plus disease was found at 0-2 disc diameter (DD) from the optic disc (R^2=0.636). At 0-2 DD, the range of tortuosity was 1.04-1.45 and was divided into the 7 point scale to create the "plus index" values. Sensitivity/specificity for diagnosing plus disease were 0.929/0.556 (using “plus index” ≥3 as cutoff) and 0.571/1.00 (using “plus index” ≥4 as cutoff).

Conclusions: This study outlines a methodology for creating a quantitative plus disease index utilizing computer-generated retinal vessel tortuosity measurements from wide-angle retinal images. With a large cohort of retinal images from infants with ROP, this methodology could generate an index to aid in the identification of plus disease with potential applications in ROP telemedicine systems.

Keywords: 706 retinopathy of prematurity • 550 imaging/image analysis: clinical