Electromagnetic Source Imaging via Bayesian Modeling With Smoothness in Spatial and Temporal Domains.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Reconstructing cortical activation from electroencephalography (EEG) and magnetoencephalography (MEG) is challenging due to the ill-posed inverse problem.
• Existing methods struggle with accurate spatial and temporal resolution.

Purpose of the Study:

• To introduce a novel algorithm, source imaging with smoothness in spatial and temporal domains (SI-SST), for improved E/MEG source reconstruction.
• To address the limitations of current inverse problem solutions in neuroimaging.

Main Methods:

• SI-SST decomposes current sources using spatial smoothing, sparseness encoding, and temporal basis functions (TBFs).
• It incorporates temporal domain smoothness via an autoregressive model.
• A novel fixed-point update rule is derived for overlapped cortical clusters, with alternative updates for variables and hyperparameters using variational inference.

Main Results:

• SI-SST demonstrated superior reconstruction performance in both spatial extents and temporal profiles on simulated and experimental datasets.
• The algorithm outperformed established benchmark methods.

Conclusions:

• SI-SST offers a significant advancement in accurately reconstructing cortical activation from E/MEG data.
• The proposed method provides enhanced spatial and temporal accuracy, addressing key challenges in neuroimaging source localization.