Deep learning models may be helpful in identifying risk of retinopathy of prematurity


New research suggests that deep learning approaches could be useful in detecting the development of retinopathy of prematurity (ROP) in high-risk infants and reducing blindness in the population.

The results of the retrospective study show that the deep learning system provided an accurate prediction of the occurrence and severity of ROP before 45 weeks postmenstrual age based on retinal photographs and features clinics before or during the first screening.

“With the help of our deep learning system, ophthalmologists and parents could be alerted to ensure that regular screening is done before the onset of ROP,” the Honghua study authors wrote. Yu, PhD, Guangdong Eye Institute and Songfu Feng, PhD, Department of Ophthalmology, Zhujiang Hospital of Southern Medical University. “This deep learning system can also be applied to reduce the workload of pediatric ophthalmologists and to increase the treatment rate of severe ROP.”

As the leading cause of childhood blindness, early detection of ROP can reduce healthcare rates and costs and lead to multigenerational benefits for individuals, families and society itself. As a result, early detection, regular monitoring and prompt treatment are crucial, according to the investigators.

However, because of the considerable effort required for ROP screening and the low rate of ROP requiring treatment, cost-effective programs are needed to identify infants at high risk for severe ROP.

Investigators in the present study collected data on 988 infants born between June 2017 and August 2019 who underwent ROP examination at 2 sites in China to develop the deep learning system. Ultimately, the study included 7033 retinal photographs of 725 infants in the training set and 763 retinal photographs of 90 infants in the external validation set, along with 46 features for each infant.

All images of each infant’s two eyes taken during the first screening were labeled according to the final diagnosis between screening and 45 weeks post-menstrual age, the investigators said. In addition, they used 2 models specifically designed for occurrence predictions (occurrence network [OC-Net]) and gravity (gravity network [SE-Net]) of the ROP.

Key results considered the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity to assess performance in predicting ROP.

Of the 815 infants, 450 (55.2%) were boys, mean gestational age was 33.1 weeks (95% confidence interval [CI], 32.9 – 33.3 weeks; median, 33 weeks; interval, 25 – 40 weeks) and mean birth weight was 1.91 kg (95% CI 1.87 – 1.95 kg).

In internal validation, using the threshold of 0.10, researchers found the mean AUC, accuracy, sensitivity, and specificity of OCT-NET to predict the occurrence of ROP to be 0, 90 (95% CI, 0.88 – 0.92), 52.8% (95% CI, 49.2% – 56.4%), 100% (95% CI, 97.4% – 100%) and 37.8% (95% CI, 33.7% – 42.1%), respectively.

Then, applying the threshold of 0.26, the mean AUC, precision, sensitivity and specificity of SE-Net for predicting severe ROP was 0.87 (95% CI, 0.82 – 0, 91), 68.0% (95% CI, 61.2% – 74.8% ), 100% (95% CI, 93.2% – 100%) and 46.6% (95% CI , 37.3%-56.0%).

Using external validation, the data show that the AUC, precision, sensitivity and specificity were 0.94, 33.3%, 100% and 75%, respectively for OC-Net and 0.88, 56.0%, 100% and 35.3%, respectively, for SE -Report.

The study, “Development and validation of a deep learning model to predict the occurrence and severity of retinopathy of prematurity“, was published in Open JAMA Network.


Comments are closed.