Purchase this article with an account.
Nisha Donthi, Osamah Saeedi, Abhishek Rege, M. Jason Brooke, Janet Alexander; Comparison of tortuosity indices across software for retinopathy of prematurity (ROP). Invest. Ophthalmol. Vis. Sci. 2016;57(12):6304.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Current methods used in screening for Retinopathy of Prematurity (ROP) are based on guidelines set forth by the International Committee for the Classification of Retinopathy of Prematurity in 2005. However, even amongst experienced pediatric ophthalmologists there is disagreement in the diagnosis of ROP suggesting subjectivity in these guidelines. ROP is a leading cause of childhood blindness and is rising in incidence in regions without readily available pediatric ophthalmologists thus establishing the need for more objective measures to evaluate ROP. Multiple software packages have been developed to automatically segment vessels and analyze tortuosity from fundus images. To our knowledge, their automated segmentation algorithms have not been compared on the same set of images. We performed a cross-sectional study comparing the tortuosity indices of three software packages with automated segmentation to manual segmentation.
16 retinal vessels from five premature infants were analyzed from photos by the RetCam3 (Clarity Medical Systems, Inc.). Visible vessels were divided into quadrants and compared. Automated Retinal Image Analyzer (ARIA) (Peter Bankhead), Computer Aided Image Analysis of the Retina (CAIAR) (University College London) and the investigational Automated Vessel Analysis Suite (AVAS) (Vasoptic Medical Inc.), were used to analyze the same images and the tortuosity indices were calculated for each of four quadrants. We used a linear regression analysis to compare the algorithms of software to manually segmented images.
We found no significant linear relationship between tortuosity indices determined by manual segmentation and any of the software packages: ARIA (R2=0.051, P=0.15), CAIAR (R2=0.0873, P=0.13), and AVAS (R2=0.0172, P=0.64). When comparing each software package to each other, there was no relationship between ARIA and CAIAR (R2=0.0096, P=0.89), AVAS and ARIA (R2=0.1531, P=0.79), or AVAS and CAIAR (R2=0.084, P=0.56).
The weak correlations among the different software underline the need for more consistent, sensitive methods to identify the retinal vessels and to calculate tortuosity. Poor image quality may also play a role in these weak correlations. More significantly, this data emphasizes the need for additional quantitative measures aside from vessel tortuosity for assessing disease stage in ROP.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.
a. manual segmentation b. CAIAR c. AVAS d. ARIA
This PDF is available to Subscribers Only