Automated detection of diabetic retinopathy: barriers to translation into clinical practice

Abstract

Automated identification of diabetic retinopathy (DR), the primary cause of blindness and visual loss for those aged 18-65 years, from color images of the retina has enormous potential to increase the quality, cost-effectiveness and accessibility of preventative care for people with diabetes. Through advanced image analysis techniques, retinal images are analyzed for abnormalities that define and correlate with the severity of DR. Translating automated DR detection into clinical practice will require surmounting scientific and nonscientific barriers. Scientific concerns, such as DR detection limits compared with human experts, can be studied and measured. Ethical, legal and political issues can be addressed, but are difficult or impossible to measure. The primary objective of this review is to survey the methods, potential benefits and limitations of automated detection in order to better manage translation into clinical practice, based on extensive experience with the systems we have developed.

Figure 1. The general process for automated detection of diabetic retinopathy lesions from color images

The system estimated this image to be of good quality and the image quality step is not shown. (A) Image with diabetic retinopathy, (B) vessel segmentation, and optic disc and fovea localization, (C) resulting detected lesions. (D) Detail where lesions are visible. Suspect pixels indicates thresholded pixel classification step for red respectively bright lesions.

Figure 2. Computer-assisted detection of diabetic retinopathy presupposes automated detection of diabetic retinopathy lesions

(A) Image of a patient with diabetes classified as having cotton-wool spots, exudates and hemorrhages caused by diabetic retinopathy. (B) Visualization of system detected ‘bright’ lesions (purple) and ‘red’ lesions (yellow–green), with the saturation of the color indicating the level of confidence the system has that the lesion is a true lesion. Automated detection resulted in a p(DR) of 0.87 – the system-estimated likelihood of this patient having diabetic retinopathy.

Figure 3. Receiver operator characteristics for different versions of the Iowa system, as well as the performance of retinal specialists, on a random sample of this data

ARVO 2008 is the version of the system as presented at the ARVO Annual Meeting in 2008. Diabetes Care 2008 is the version of the system as published in the journal Diabetes Care [12]. Retinal specialists 1, 2 and 3 are the sensitivity and specificity of three retinal specialists on a subset of 500 exams from the dataset used in the Diabetes Care paper. ARVO: Association for Research in Vision and Ophthalmology.