Automated Diagnosis of Plus Disease in Retinopathy of Prematurity Using Deep Convolutional Neural Networks

Abstract

Importance: Retinopathy of prematurity (ROP) is a leading cause of childhood blindness worldwide. The decision to treat is primarily based on the presence of plus disease, defined as dilation and tortuosity of retinal vessels. However, clinical diagnosis of plus disease is highly subjective and variable.

Objective: To implement and validate an algorithm based on deep learning to automatically diagnose plus disease from retinal photographs.

Design, setting, and participants: A deep convolutional neural network was trained using a data set of 5511 retinal photographs. Each image was previously assigned a reference standard diagnosis (RSD) based on consensus of image grading by 3 experts and clinical diagnosis by 1 expert (ie, normal, pre-plus disease, or plus disease). The algorithm was evaluated by 5-fold cross-validation and tested on an independent set of 100 images. Images were collected from 8 academic institutions participating in the Imaging and Informatics in ROP (i-ROP) cohort study. The deep learning algorithm was tested against 8 ROP experts, each of whom had more than 10 years of clinical experience and more than 5 peer-reviewed publications about ROP. Data were collected from July 2011 to December 2016. Data were analyzed from December 2016 to September 2017.

Exposures: A deep learning algorithm trained on retinal photographs.

Main outcomes and measures: Receiver operating characteristic analysis was performed to evaluate performance of the algorithm against the RSD. Quadratic-weighted κ coefficients were calculated for ternary classification (ie, normal, pre-plus disease, and plus disease) to measure agreement with the RSD and 8 independent experts.

Results: Of the 5511 included retinal photographs, 4535 (82.3%) were graded as normal, 805 (14.6%) as pre-plus disease, and 172 (3.1%) as plus disease, based on the RSD. Mean (SD) area under the receiver operating characteristic curve statistics were 0.94 (0.01) for the diagnosis of normal (vs pre-plus disease or plus disease) and 0.98 (0.01) for the diagnosis of plus disease (vs normal or pre-plus disease). For diagnosis of plus disease in an independent test set of 100 retinal images, the algorithm achieved a sensitivity of 93% with 94% specificity. For detection of pre-plus disease or worse, the sensitivity and specificity were 100% and 94%, respectively. On the same test set, the algorithm achieved a quadratic-weighted κ coefficient of 0.92 compared with the RSD, outperforming 6 of 8 ROP experts.

Conclusions and relevance: This fully automated algorithm diagnosed plus disease in ROP with comparable or better accuracy than human experts. This has potential applications in disease detection, monitoring, and prognosis in infants at risk of ROP.

Figure 1.. Receiver Operating Characteristic (ROC) Curves for Diagnosis of Plus Disease in Retinopathy of Prematurity

Data were analyzed from 5-fold cross-validation of 5511 retinal images. Mean areas under the ROC curves (AUCs) for the 5 sets were 0.94 for identifying normal images (vs pre–plus disease or plus disease; A) and 0.98 for identifying plus disease images (vs normal or pre–plus disease; B).

Figure 2.. Diagnostic Performance of the Imaging and Informatics in Retinopathy of Prematurity (i-ROP) Deep Learning (DL) Algorithm and 8 ROP Experts Compared With the Reference Standard Diagnosis (RSD) on a Data Set of 100 Images

A, Confusion matrix for the DL algorithm, with numbers of correctly and incorrectly classified images in each class. B, Interrater heat map, with quadratic-weighted κ scores comparing 8 independent experts, the DL algorithm, and the RSD. The consensus diagnosis among 8 experts is also shown, calculated as the most frequent (mode) diagnosis. C, Receiver operating characteristic (ROC) curve for the DL algorithm and performance of 8 experts in terms of true-positive rates (ie, sensitivity) and false-positive rates (ie, 1 − specificity).

Figure 3.. t-Distributed Stochastic Neighbor Embedding Visualization of Features Extracted From an Intermediate Layer of a Trained Convolutional Neural Network for Plus Disease Diagnosis in Retinopathy of Prematurity

This visualization demonstrates that the convolutional neural network is able to automatically generate features that roughly separate the 3 diagnoses and that they appear to run along a continuum of disease severity.