Software for Reading and Grading Diabetic Retinopathy

Abstract

OBJECTIVE—To evaluate the validity and reproducibility of software for reading digital images and grading diabetic retinopathy.

RESEARCH DESIGN AND METHODS—A prospective, comparative observational study was conducted on a series of patients with type 2 diabetes who presented at the retina clinic of a tertiary care center in India. A total of 210 eyes of 105 patients were allocated to one of three ophthalmologists, who performed dilated indirect and direct ophthalmoscopy and subsequently assessed the digital images of the same group of patients who were masked to the patient's identity. The interobserver and intertest agreement between clinical assessments and grading of diabetic retinopathy using the software was estimated.

RESULTS—Moderate nonproliferative diabetic retinopathy (NPDR) was most frequently diagnosed, both clinically and on evaluating digital images. The overall agreement between the clinical grading of diabetic retinopathy and the grading of images was 81.3% (κ = 0.69, SE 0.04, P < 0.0001); there was good agreement (81.3%) for NPDR (κ = 0.61, SE 0.05, P < 0.0001), but agreement was not as good (54.6%) for proliferative diabetic retinopathy (κ = 0.29, SE 0.11, P = 0.005). Clinically significant macular edema was diagnosed in 33.3% (70 of 210) of eyes clinically and in 40.2% (84 of 209) of eyes by grading images, and there was good agreement (89.5%) between the two (κ = 0.77, SE 0.07, P < 0.0001).

CONCLUSIONS—Aravind Diabetic Retinopathy Screening 3.0 is a simple and valid tool to assist in the detection of sight-threatening retinopathy and could supplement dilated fundus examinations by ophthalmologists on patients to detect diabetic retinopathy.