We can’t show the full text here under this license.
Using explainable AI to improve patient selection in a phase II depression trial with the NetraAI platform
Updated
Abstract
Essence
NetraAI identified treatment-response subgroups in a small ketamine trial for treatment-resistant depression.
Evidence
A platform demonstration in a Phase II trial of 63 patients used psychiatric scale and MRI-derived features, reporting about 25-30% better predictive accuracy, a 0.32 AUC gain for a 10-variable clinical model, and 95% accuracy with 100% specificity for an 8-feature MRI model.
Caveat
The evidence comes from a small trial dataset, and prospective enrichment performance across other trials or disorders was not tested.
Simplified
Key numbers
0.32
Increase in Predictive
Comparison of 's 10-variable model vs. standard machine learning models.
81%
True Positive Rate
Among 63 trial participants, 43 were classified as PNR/TR.
100%
100% Accuracy
Achieved by the MRI-based model for identifying a subgroup of 15 patients.
Psychedelic Science