Artificial Intelligence Versus Human Grading for Diabetic Retinopathy

The use of automated retinal image analysis software using true-color, wide-field confocal scanning images and standard fundus images in the English National Diabetic Eye Screening Programme (NDESP) was compared with human grading to identify diabetic retinopathy in a cross-sectional study.

Patients attending annual diabetic eye screening underwent imaging with mydriasis using the EIDON platform and standard NDESP cameras. Human grading was performed per NDESP protocol. The human grade of standard NDESP images was used as the reference standard.

Final analysis included 1,257 patients. For EIDON images, the sensitivity estimates for retinopathy grades were: for any retinopathy, 92.27% (95% confidence interval [CI], 88.43-94.69); for vision-threatening retinopathy, 99.00% (95% CI, 95.35-100.00); and for proliferative retinopathy, 100.00% (95% CI, 61.00-100.00). For NDESP images, the sensitivity estimates for retinopathy grades were: for any retinopathy, 92.26% (95% CI, 88.37-94.69); for vision-threatening retinopathy, 100.00% (95% CI, 99.53-100.00); and for proliferative retinopathy, 100.00% (95% CI, 61.00-100.00). When analyzing the EIDON images, the EyeArt missed one case of vision-threatening retinopathy, which was identified by the human graders. In the standard images, the EyeArt identified all vision-threatening retinopathy cases.

“EyeArt identified diabetic retinopathy in EIDON images with similar sensitivity to standard images in a large-scale screening [program], exceeding the sensitivity threshold recommended for a screening test. Further work to [optimize] the identification of ‘no retinopathy’ and to understand the differential lesion detection in the two imaging systems would enhance the use of these two innovative technologies in a diabetic retinopathy screening setting,” the researchers concluded.