Objective: One of the principal drawbacks of magnetoencephalography (MEG) is its high sensitivity to metallic artifacts, which come from implanted intracranial electrodes and dental ferromagnetic prosthesis and produce a high distortion that masks cerebral activity. The aim of this study was to develop an automatic algorithm based on blind source separation (BSS) techniques to remove metallic artifacts from MEG signals.
Approach: Three methods were evaluated: AMUSE, a second-order technique; and INFOMAX and FastICA, both based on high-order statistics. Simulated signals consisting of real artifact-free data mixed with real metallic artifacts were generated to objectively evaluate the effectiveness of BSS and the subsequent interference reduction. A completely automatic detection of metallic-related components was proposed, exploiting the known characteristics of the metallic interference: regularity and low frequency content.
Main results: The automatic procedure was applied to the simulated datasets and the three methods exhibited different performances. Results indicated that AMUSE preserved and consequently recovered more brain activity than INFOMAX and FastICA. Normalized mean squared error for AMUSE decomposition remained below 2%, allowing an effective removal of artifactual components.
Significance: To date, the performance of automatic artifact reduction has not been evaluated in MEG recordings. The proposed methodology is based on an automatic algorithm that provides an effective interference removal. This approach can be applied to any MEG dataset affected by metallic artifacts as a processing step, allowing further analysis of unusable or poor quality data.