Automated Algorithm Accurately IDs Plus Disease in ROP
TUESDAY, May 8, 2018 -- A fully automated algorithm can accurately diagnose plus disease in retinopathy of prematurity (ROP), according to a study published online May 2 in JAMA Ophthalmology.
James M. Brown, Ph.D., from Massachusetts General Hospital in Charlestown, and colleagues trained a deep convolutional neural network using a data set of 5,511 retinal photographs. Based on consensus of image grading by three experts and clinical diagnosis by one expert, each image was assigned a reference standard diagnosis (RSD). The algorithm was assessed by five-fold cross-validation and tested on 100 independent images. The deep learning algorithm was then tested against eight ROP experts.
The researchers found that the mean area under the receiver operating characteristic curve statistics were 0.94 and 0.98, respectively, for the diagnosis of normal versus pre-plus or plus disease and for the diagnosis of plus disease versus normal or pre-plus disease. In an independent set of 100 retinal images, the algorithm achieved sensitivity and specificity of 93 and 94 percent, respectively, for diagnosis of plus disease. For detection of pre-plus disease or worse, there was 100 and 94 percent sensitivity and specificity, respectively. The algorithm achieved a quadratic-weighted κ coefficient of 0.92 compared with the RSD, outperforming six experts.
"This fully automated algorithm diagnosed plus disease in ROP with comparable or better accuracy than human experts," the authors write.
Several authors disclosed financial ties to the pharmaceutical and medical device industries.
© 2019 HealthDay. All rights reserved.
Posted: May 2018