AI FOR LITTLE EYES: A SYSTEMATIC REVIEW OF DEEP LEARNING IN EVALUATING PEDIATRIC CATARACT

Abstract

Background: Pediatric or congenital cataract (CC) is a leading cause of visual impairment and blindness in children worldwide. Deep learning (DL), a subfield of artificial intelligence, has the potential to enhance diagnosis, treatment, and outcomes in various medical fields.
 Research Objectives: summarize and evaluate the diagnostic and prediction capabilities of DL algorithms for CC.
 Methods: From 1st February to 25th March 2023, a literature search was conducted in databases such as PubMed, ScienceDirect, EMBASE, and EBSCO, as well as alternative sources such as Google Scholar. Search terms included “pediatric/congenital cataract”, “artificial intelligence", "deep learning", "convolutional neural network", “diagnosis”, "screening", "prediction" and other relevant synonyms. Quality assessment of studies were assessed based on CONSORT-AI and QUADAS-2. Outcomes extracted included accuracy, sensitivity, specificity, and area under the curve (AUC).
 Results: Out of 69 studies screened, five studies with different study designs, dataset sizes, and type of DL algorithms employed were included in the systematic review. Most studies employed DL to analyze slit-lamp images to diagnose CC, while one study utilized DL to predict existence of CC from several risk factors. In silico, most studies demonstrated high accuracy and validity of DL algorithms in detecting and predicting CC; however, DL algorithm is not as accurate in diagnosing CC when compared to human counterparts. These studies had limited generalizability given the homogenous population.
 Conclusion: DL shows potential as an adjunct tool for ophthalmologists to improve diagnosis and, therefore, treatment decisions for CC, particularly in remote and underdeveloped regions with limited medical resources.