A Novel Time-Aware Deep Learning Model Predicting Myopia in Children and Adolescents

Abstract

Listen

Translate article

ObjectiveTo quantitatively predict children’s and adolescents’ spherical equivalent by leveraging their variable-length historical vision records. DesignRetrospective analysis. ParticipantsEight hundred ninety-five myopic children and adolescents aged 4-18 years, with a complete ophthalmic examination and retinoscopy in cycloplegia prior to spectacle correction were enrolled in the period from January 1st 2008 till July 1st, 2023 at the University Hospital “Sveti Duh”, Zagreb, Croatia. MethodsA novel modification of Time-Aware Long Short-Term Memory (T-LSTM) was used to quantitatively predict children’s and adolescents’ spherical equivalent within 7 years after diagnosis. Main Outcome MeasuresThe utilization of Extended Gate Time-Aware Long Short-Term Memory (egT-LSTM) involved capturing temporal features within irregularly sampled time series data. This approach aligned more closely with the characteristics of real-world data, increasing its applicability and contributing to the early identification of myopia progression. ResultsThe testing set exhibited a mean absolute prediction error (MAE) of 0.10±0.15 D for spherical equivalent. Lower MAE values were associated with longer sequence lengths, shorter prediction durations, older age groups, and low myopia, while higher MAE values were observed with shorter sequence lengths, longer prediction durations, younger age groups, and in premyopic or high myopic individuals, ranging from as low as 0.03±0.04 D to as high as 0.45±0.24 D. ConclusionExtended Gate Time-Aware Long Short-Term Memory capturing temporal features in irregularly sampled time series data can be used to quantitatively predict children’s and adolescents’ SE within seven years with an overall error of 0.10±0.15 D. This value is substantially lower than the threshold for prediction to be considered clinically acceptable, such as a criterion of 0.75 D.