Fig. 4: Performance of the Natural Progression Module (NPM) on the internal test set and external test set.
From: AI-guided personalized predictions on myopia progression and interventions
a–d internal test set. e–h external test set. a, b Scatter plots showing the comparisons between observed and predicted SE values in the internal test set for right eyes (a) and left eyes (b), respectively. c, d Scatter plots showing the comparisons between observed and predicted AL values in the internal test set for right eyes (c) and left eyes (d), respectively. e, f Scatter plots comparing observed and predicted SE values in the external test set for right eyes (e) and left eyes (f). g, h Scatter plots comparing observed and predicted AL values in the external test set for right eyes (g) and left eyes (h). NPM demonstrated high accuracy in predicting SE and AL for both the right and left eyes in both the internal validation set and the external test set, with all R² values exceeding 0.9. For SE prediction, over 90% of the predictions had an absolute error <0.75D, which is considered a clinically acceptable error. For AL prediction, >71% of the predictions had an absolute error below 0.25 mm, which represents a very small margin of error. SE spherical equivalent, AL axial length. od right eyes, os left eyes, R² coefficient of determination, MAE mean absolute error. P(AE < 0.75D): proportion of predictions with absolute error less than 0.75 diopters. P(AE < 0.25 mm): proportion of predictions with absolute error <0.25 mm.
