Fig. 1: Parallel machine learning (ML) pipeline comparing traditional classifiers to the NetraAI relabeling approach. | npj Digital Medicine

Fig. 1: Parallel machine learning (ML) pipeline comparing traditional classifiers to the NetraAI relabeling approach.

From: Explainable AI-driven precision clinical trial enrichment: demonstration of the NetraAI platform with a phase II depression trial

Fig. 1

Both workflows are applied to the same trial dataset (n = 63; endpoint ≥40% MADRS improvement at Day 7; 175 baseline clinical scales variable per patient). For the traditional pipeline, four classifiers (Naïve Bayes, Random Forest, Gradient Boosting, Deep Neural Network) are used to identify models to characterize ketamine response. Performance is assessed by accuracy, AUC, sensitivity, and specificity. For the NetraAI Relabeling Pipeline, NetraAI relabels patients, excludes unexplainable patients, and derives a 10-variable model. The same four traditional classifiers are retrained on the NetraAI-relabeled cohort and reduced feature set, and their post-relabeling metrics are compared to the traditional baseline to quantify performance lift.

Back to article page