Summaries of Must-Read Clinical Literature, Guidelines, and FDA Actions
How Well Can This System Detect Diabetic Retinopathy?
JAMA; ePub 2016 Nov 29; Gulshan, Peng, et al
A machine learning-based algorithm was able to accurately detect diabetic retinopathy in a study involving nearly 6,000 individuals.
Investigators looked at how specific and sensitive the algorithm performed on more than 11,700 images from 2 different collections of retinal images: EyePACS-1 and Messidor-2. Among the results:
- The algorithm had an area under the receiver operating curve of 0.991 for EyePACS-1 and 0.990 for Messidor-2.
- Using the cut point with high specificity, sensitivity was 90% and specificity was 98% in the EyePACS-1 collection; it was 87% and 99%, respectively, in the Messidor-2 set.
- Using the cut point with high sensitivity, sensitivity was 98% and specificity was 93% in the EyePACS-1 collection; it was 96% and 94%, respectively, in the Messidor-2 set.
Citation:
Gulshan V, Peng L, Coram M, et al. Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs. [Published online ahead of print November 29, 2016]. JAMA. doi:10.1001/jama.2016.17216.