Purchase this article with an account.
Catarina A. Neves, Carlos M. Oliveira, Maria Luisa Ribeiro, Helder J. Ferreira, Jose G. Cunha-Vaz; Prospective Evaluation of an Automated System to reduce Human Grading Burden in a Diabetic Retinopathy Screening Program. Invest. Ophthalmol. Vis. Sci. 2012;53(14):2890.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To evaluate in a prospective study the efficacy of an Automated System to reduce the human grading burden in a Diabetic Retinopathy (DR) Screening Program.
The RetmarkerDR Screening software has been implemented in a DR Screening Programme in the central region of Portugal. A total of 2117 patient exams (4234 eyes) were included in September and October 2011. Non-mydriatic color fundus photography was performed at 3 different local health centres and graded at a centralized reading centre. The RetmarkerDR Screening analyses all image sets received and separates those that need human grading, and those for which no human grading is needed. Human Grading is thus reduced only to those image sets identified. For quality control and to test the safety of the system, in this study 35% of the image sets marked as not requiring human grading were sent for human grading. The human grader was blind to this process.
The objective of DR Screening is to identify eyes/patients that need to be referred for a specialist appointment, and the ones that do not need immediate referral, and therefore should return one year later for the annual screening. The analysis/grading is performed per eye, evaluating the presence of referable DR (maculopathy and/or proliferative diabetic retinopathy). From 4234 eyes, 2277 were sent to human grading by the automated system. The grader confirmed referable DR in 70 eyes. When testing the safety of the system, of a total of 695 eyes that were considered as not requiring human grading, only 1 was identified as having referable DR, and none had proliferative diabetic retinopathy. The automated system shows, therefore, a specificity of 46.96% and an estimated sensitivity of 96.13%.
Prospective analysis of the efficacy of the RetmarkerDR Screening, an automated system to analyse retinal photographs in the context of a DR Screening Programme, shows that it has a potential human grader burden reduction of 46%.
This PDF is available to Subscribers Only