A Myopia Degree Prediction Model for Children and Adolescents Based on Follow-up Data with Different Time Intervals

• DOI: 10.62051/ijcsit.v5n1.14
• Publication Date: 2025-02-21T02:16:02Z
• Authors: Pengyuan Ma; Shukui Ma; Guowei Zhang; Fei Ma; Guangping Zhuo

Myopia has become a global public health issue, and existing research primarily focuses on predicting the future spherical equivalent (SE) of children and adolescents based on ocular optical data. However, electronic medical record (EMR) data often exhibit characteristics such as multi-dimensionality, irregular time series, missing values, and non-uniform time intervals, which present significant challenges for prediction. To address these issues, this paper proposes a prediction model named IST-xLSTM, which utilizes a temporal dense prediction module to handle missing data and overcome long-range dependencies and irregular time intervals. Additionally, a hierarchical temporal attention mechanism is introduced to extract both local and global features along the irregular time dimension, enhancing prediction accuracy. Experimental results show that the mean absolute error on the validation set is 0.216 ± 0.176 (D), which is significantly lower than the clinically acceptable prediction threshold of 0.50 (D).