Abstract
Purpose :
Infantile retinopathy is the leading cause of irreversible blindness in infants, and early detection is crucial for their management. However, many regions suffer from a severe shortage of pediatric ophthalmologists, resulting in insufficient coverage for screening of infantile fundus diseases and causing numerous patients to miss the optimal window for treatment. Therefore, the objective of this study is to develop an intelligent screening system that encompasses common infantile fundus diseases, thus promoting the scalability of infantile retinopathy screening.
Methods :
During the screening of infantile retinopathy, doctors need to review fundus color photographs from many fields of view to make accurate diagnoses. Considering this characteristic, we adopted a multi-instance learning framework to train our model. We collected a total of 11,160 cases of infantile fundus photography from two centers, comprising 128,090 fundus photographs. Subsequently, we developed and validated a multi-class intelligent system capable of identifying fundus hemorrhage, retinopathy of prematurity, and retinoblastoma in infants.
Results :
In the validation set, the intelligent system achieved F1 scores of 0.87 (95% confidence interval (CI):0.85-0.89), 0.81 (95%CI:0.79-0.83), and 0.96 (95%CI:0.95-0.97) for identifying fundus hemorrhage, retinopathy of prematurity, and retinoblastoma, respectively. The average accuracy for identifying the three diseases was 0.85 (95%CI:0.83-0.87). Additionally, the intelligent system was able to identify which field of the retina with abnormalities.
Conclusions :
Our intelligent system has achieved high accuracy in detecting common infantile retinopathy, and it can also provide cues to doctors regarding the regions of fundus images where potential abnormalities are suspected, thus prioritizing attention. Therefore, our system has the potential to enhance the efficiency of pediatric ophthalmologists in conducting infantile retinopathy screening.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.