Electromagnetic neural source imaging under sparsity constraints with SURE-based hyperparameter tuning

TL;DR

This paper introduces a SURE-based method for automatic hyperparameter tuning of sparsity-promoting estimators in M/EEG source localization, demonstrating superior performance over traditional methods on large-scale datasets.

Contribution

It presents the first large-scale automatic calibration of sparsity-based estimators using SURE, improving model selection in brain source imaging.

Findings

SURE-based tuning outperforms cross-validation in accuracy and speed

Method achieves state-of-the-art results on realistic simulations

Applicable to large datasets like Cam-CAN

Abstract

Estimators based on non-convex sparsity-promoting penalties were shown to yield state-of-the-art solutions to the magneto-/electroencephalography (M/EEG) brain source localization problem. In this paper we tackle the model selection problem of these estimators: we propose to use a proxy of the Stein's Unbiased Risk Estimator (SURE) to automatically select their regularization parameters. The effectiveness of the method is demonstrated on realistic simulations and $30$ subjects from the Cam-CAN dataset. To our knowledge, this is the first time that sparsity promoting estimators are automatically calibrated at such a scale. Results show that the proposed SURE approach outperforms cross-validation strategies and state-of-the-art Bayesian statistics methods both computationally and statistically.