Explainable ML for Predicting Vision Loss in Pediatric NF-1 Patients Using OCT Data

We applied explainable machine learning to optical coherence tomography (OCT) data from 168 pediatric NF-1 patients to predict vision abnormalities. The Balanced Random Forest model achieved an AUC-ROC of 0.82. SHapley Additive exPlanations (SHAP) analysis identified data-driven cut-offs providing actionable clinical thresholds. Combining multiple abnormal OCT features further improved specificity. This approach demonstrates the value of interpretable AI for early risk detection and individualized care in pediatric NF-1.