XAI Reduces NEXI Brain Scan Time on Connectome 2.0 from 27 to 14 Minutes
Researchers have developed a new protocol for Neurite Exchange Imaging (NEXI) using an Explainable AI (XAI) framework on the advanced Connectome 2.0 ultra-high gradient scanner. By combining XGBoost, SHAP, and Recursive Feature Elimination, they identified an optimal set of 8 features, which cut the scanning time from 27 minutes down to 14. In tests conducted on seven healthy participants, this XAI method reliably reproduced parameter estimates and showed impressive test-retest consistency, outperforming the traditional 15-feature method, a theoretical Cramér-Rao Lower Bound (CRLB), and two other heuristics, the "Mid-Range" and "Corner." This innovative approach addresses the lengthy multi-shell and multi-diffusion-time challenges of NEXI.
Key facts
- NEXI protocol scan time reduced from 27 to 14 minutes
- XAI framework used XGBoost, SHAP, and Recursive Feature Elimination
- Optimal 8-feature subset identified from synthetic signals
- Validated in vivo in seven healthy participants
- Benchmarked against full 15-feature acquisition, CRLB, and two heuristics
- Connectome 2.0 ultra-high gradient scanner employed
- NEXI estimates compartment diffusivities, neurite fraction, and exchange time
- Study published on arXiv (2509.09513)
Entities
Institutions
- arXiv