Skip to main content

Table 3 Performance of the classification models for optimal ICL size prediction

From: Predicting post-operative vault and optimal implantable collamer lens size using machine learning based on various ophthalmic device combinations

Device (s)

Top 3 algorithms

Validation set

Test set

ACC (95% CI)

AUC (95% CI)

ACC (95% CI)

AUC (95% CI)

UBM

et

0.834 (0.828 to 0.840)

0.905 (0.898 to 0.912)

0.837 (0.830 to 0.843)

0.906 (0.897 to 0.914)

rf

cbt

Pentacam

et

0.682 (0.667 to 0.698)

0.803 (0.789 to 0.817)

0.682 (0.667 to 0.697)

0.802 (0.789 to 0.817)

lgb

rf

Sirius

et

0.701 (0.689 to 0.713)

0.809 (0.797 to 0.823)

0.695 (0.682 to 0.708)

0.808 (0.795 to 0.821)

cbt

lgb

UBM and Pentacam

et

0.851 (0.842 to 0.859)

0.908 (0.898 to 0.918)

0.855 (0.847 to 0.864)

0.911 (0.901 to 0.921)

rf

cbt

UBM and Sirius

lgb

0.863 (0.855 to 0.872)

0.915 (0.903 to 0.927)

0.862 (0.853 to 0.870)

0.913 (0.901 to 0.926)

cbt

xbt

Pentacam and Sirius

et

0.728 (0.714 to 0.742)

0.816 (0.812 to 0.819)

0.734 (0.720 to 0.749)

0.818 (0.814 to 0.821)

rf

cbt

UBM and Pentacam and Sirius

cbt

0.891 (0.879 to 0.904)

0.926 (0.913 to 0.939)

0.895 (0.883 to 0.907)

0.928 (0.916 to 0.941)

rf

lgb

Only other devices

et

0.544 (0.530 to 0.558)

0.636 (0.627 to 0.646)

0.543 (0.539 to 0.547)

0.634 (0.630 to 0.638)

rf

cbt

  1. et Extra trees classifier; rf Random forest classifier; lgb Light gradient boosting machine; cbt CatBoost classifier; xbt Extreme gradient boosting; ACC accuracy; AUC the area under the curve; CI confidence interval