Predicting progression of myopia using AI

Researchers have developed an algorithm that could help clinicians to determine the best strategy for managing myopia in a child

A young child wearing a blue polka dot puffer vest stands in on a path through the forest blowing bubbles
Pixabay/Daniela Dimitrova

Writing in BMJ Open Ophthalmology, researchers from China and Germany have described their efforts to develop a machine learning-based algorithm for the prediction of refractive error and its progression.

The scientists highlighted that while several management solutions exist to prevent the onset of myopia and slow its progression, the application of these techniques depends on an assessment of the risk of developing myopia.

“The developed algorithm uses accessible inputs [age, gender and spherical power] to provide an estimate of refractive development and may serve as guide for the eye care professional to help determine the individual best strategy for management of myopia,” the researchers highlighted.

The algorithm was developed using cross-sectional data from 12,780 Chinese children between the ages of five and 16 and longitudinal data from 226 children. Information collected included age, gender, biometry and refractive parameters.

The authors highlighted that the purpose of the study was to develop an algorithm for the prediction of spherical power progression using the least possible number of inputs – without hindering an acceptable performance.

“The resulting algorithm with the best performance allowed the use of the age, sphere and gender as inputs. This arrangement may lead to a flexible application, without the need to collect all available biometrical parameters of the eye,” the authors shared.

The performance of this algorithm showed a correlation between prediction and measured true data of 0.77.

Limitations of the study include that the validation of the algorithm was performed using longitudinal data from 81 children.

There are also other factors separate to the algorithm inputs – such as parental myopia, near-work time and time spent outdoors – that affect refractive power progression.

“Thus, algorithms that do not consider these variables must be taken with specific care… They must be used only as a reference and not as a unique tool for risk assessment,” the researchers shared.