Purchase this article with an account.
Sapna Tibrewal, Peng Tian, Dharanish Kedarisetti, Jayashree Kalpathy-Cramer, Stratis Ioannidis, Deniz Erdogmus, J. Peter Campbell, Robison Vernon Paul Chan, Michael F Chiang; Evaluation of computer-based image analysis for retinopathy of prematurity screening. Invest. Ophthalmol. Vis. Sci. 2017;58(8):5539.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Retinopathy of prematurity (ROP) diagnosis may be highly subjective and qualitative, even by experts. Computer-based image analysis (CBIA) systems offer the potential to help clinicians diagnose ROP more accurately and consistently, and may have applications for ROP screening in areas of the world where accessibility to clinicians in limited. This project was designed to explore whether a CBIA system (the “i-ROP” system) can identify infants with clinically-significant ROP.
We developed the “i-ROP” system to calculate a ROP severity score using a computer based algorithm. The algorithm was developed using the image characteristics that provided the best correlation between the calculated severity scores and cumulative ranks obtained for each image after grading by 13 experts. On a dataset of 195 images, we measured the receiver operating characteristic curve, calculated the area under the curve (AUC), and identified the sensitivity and specificity of the i-ROP system for detecting pre-plus or worse disease. We also compared the i-ROP score to zone, stage, and overall early treatment for retinopathy of prematurity (ETROP) category to determine the sensitivity of the system for detecting clinically-significant disease.
Linear regression scores obtained using the acceleration and dilation indices of arteries and veins together correlated best with the cumulative ranks of the images. The AUC for the i-ROP system was 0.94. The system could detect presence of pre-plus or plus disease with a sensitivity of 95% and a specificity of 72%. In terms of the ETROP categories, i-ROP system could detect 93% of ETROP type 2 disease or worse (n=47). The sensitivity for detection of ETROP type 1 disease (n=27) was 100%.
This study shows that the i-ROP computer-based image analysis tool is able to reliably detect clinically-significant ROP with high accuracy. This system may have potential to assist ophthalmologists in making more accurate and consistent diagnoses. These systems could also have a significant impact in ROP telemedicine programs worldwide by optimizing the screening capacity of limited human resources and improving access to care.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.
This PDF is available to Subscribers Only