AI ocular screenings facilitate early intervention for paediatric patients

The researchers' findings highlight the promise of AI in enhancing the diagnosis and management of paediatric ocular conditions. Image credit: ©Microgen – stock.adobe.com

A recent study that evaluated the usefulness of artificial intelligence (AI) to assess the ocular health of children in a region of Poland found the technology to be comprehensive and accurate in diagnosing ocular pathologies in this patient population,¹ according to first author Regulski Piotr, PhD. He is from the Laboratory of Digital Imaging and Virtual Reality, Department of Dental and Maxillofacial Radiology, Medical University of Warsaw, Poland.

He and his colleagues conducted this study to determine the prevalence rates of myopia, hyperopia and astigmatism in school-age children aged 7 to 9 years in Lublin, Poland. They also wanted to determine the ability of AI to detect severe ocular diseases such as glaucoma, diabetic macular oedema and macular degeneration.

The onset of the COVID-19 pandemic highlighted the increasing prevalence of visual impairment in children, particularly myopia, hyperopia and astigmatism. What became apparent during the pandemic, the investigators explained, was that “long-lasting distance learning and reduced physical activities have all contributed to the development of significant negative effects on the health of children, particularly vision impairment and eye disorders.”

“Not only can these conditions impede academic achievement, but they may also, in severe cases, lead to blindness, underscoring the urgency of early intervention,” the investigators emphasised.

Study methodology

A total of 1,049 children (533 boys and 516 girls) from 10 schools in the region participated in the study. This number of students represented about 1.7% of the children aged 7 to 9 years living in that region. Five of the schools were situated in rural areas and 5 in urban settings; 173 boys and 184 girls were from rural settings, and 360 boys and 332 girls were from urban areas.

The children were examined using standardised visual acuity tests, autorefraction, and assessment of fundus images using a convolutional neural network (CNN) model, which automatically analyses images captured by the fundus camera. The investigators conducted the study from September 2021 to May 2022.

Paediatric rates of common ocular disorders

The investigators reported the following prevalence rates of ocular disorders in the student population: myopia, 3.7%; hyperopia 16.9%; and astigmatism 7.8%. They explained that the prevalence rate of myopia increased with the increasing age of the children, that is, it increased from 1.9% in the first grade students to 5.8% in the children in the third grade, a difference that reached significance (χ2 = 7.2306, P = .0269).

They also reported that the prevalence rate of hyperopia decreased significantly with increasing age from 21.8% in the first graders to 11.9% in the third graders, which also reached significance (χ2 = 12.0886, P = .0024).

Finally, the data also showed that the prevalence rate of astigmatism was significantly higher in urban areas (8.9%) than in rural areas (5.1%) (χ2 = 5.8346, P = .0236).

The investigators did not identify any significant correlations between sex, age and school region and eyeglass use.

AI performance

Dr Piotr and colleagues reported, “The AI model predicted the presence of drusen and potential degeneration in 7 cases. These predictions were subsequently corroborated by ophthalmologists during clinical examinations of all 7 children. Therefore, the model was accurate in detecting macular degeneration.”

The model predicted diabetic retinopathy in one child; however, that diagnosis was not confirmed during subsequent clinical examination. A total of 109 participants were subsequently referred for further evaluation.

The authors explained that the Dice coefficient, a statistical validation metric that shows the predictive accuracy of the model, yielded a high value of 93.3%, which attested to the model’s efficiency and reliability in detecting ocular pathologies.

In their discussion, the authors pointed out the scarcity of reports on the use of AI for detecting visual impairments in children primarily because of the small number of extensive and age-specific screening protocols for this population.²

"This void in the literature highlights a pressing need to develop and test advanced diagnostic methods that are adapted to young populations.^3,4. The present study addresses this need by implementing a CNN algorithm, thereby presenting an innovative approach to paediatric ophthalmic screening. Despite the single instance of overprediction where the CNN model identified diabetic retinopathy that was not confirmed by ophthalmologists, this occurrence should not be disregarded. It could potentially indicate a prediabetic state or an early stage of the disease that has not yet manifested in noticeable ocular changes. Given that diabetic retinopathy is a known complication of diabetes, this case may underscore the predictive capabilities of the AI model and its potential usefulness in preventive medicine. Therefore, we suggest that this child be monitored more closely for signs of diabetes development,” the authors commented.

They also emphasised the importance of amassing an extensive body of paediatric ocular data that can serve as “foundational training material for developing AI algorithms.”

Dr Piotr and colleagues concluded, “This study provides a comprehensive evaluation of ocular health among children in the Lublin Voivodeship, underscoring the prevalence of common visual impairments and the potential role of factors such as age and geographical location. The findings also highlight the promise of innovative technologies, such as AI, in enhancing the diagnosis and management of paediatric ocular conditions. Future research should seek to replicate and extend these findings, exploring the underlying causes of the observed trends and examining the long-term impact of these conditions on children’s educational and health outcomes.”

References

1. Piotr R, Robert R, Marek N, Michal I. Artificial intelligence enhanced ophthalmological screening in children: insights from a cohort study in Lubelskie Voivodeship. Sci Rep. 2024;14(1): 254. Published online Jan 2, 2024. doi: 10.1038/s41598-023-50665-5

2. Wolf RM, Liu TY, Thomas C, et al. The SEE study: safety, efficacy, and equity of implementing autonomous artificial intelligence for diagnosing diabetic retinopathy in youth. Diabetes Care. 2021;44(3):781–787. doi:10.2337/dc20-1671

3. Varela MD, Sen S, Cabral de Guimaraes T, et al. Artificial intelligence in retinal disease: clinical application, challenges, and future directions. Graefes Arch Clin Exp Ophthalmol. 2023;261(11):3283-3297. doi:10.1007/s00417-023-06052-x

4. Demir F, Taşcı B. An effective and robust approach based on R-CNN+LSTM model and NCAR feature selection for ophthalmological disease detection from fundus images. J Pers Med. 2021; 11(12):1276. doi:10.3390/jpm11121276