Reconstruction of Low-Quality Channel Data in Magnetoencephalography using Surface Reconstruction and Interpolation Methods

Abstract

Purpose: Magnetoencephalography (MEG) is a brain imaging method with high temporal and acceptable spatial resolution, achieved by recording neural magnetic fields. The data quality of this imaging method is compromised due to reasons such as the failure of one or more sensors. This study aims to explore the efficiency of the various data reconstruction techniques in MEG for recovering poor-quality or missing channel data. Materials and Methods: We compared three surface reconstruction methods (Mean, Median, and Trimmed mean), two partial differential equations (modified Poisson and Diffusion equation), and a Finite Element-based interpolation method using data from 11 young adults (aged 30±12). Using varying levels of simulated data loss (2%, 5%, 11%, and 16%), we assessed each method in terms of time taken for reconstruction, R-squared, root mean squared error (RMSE), and signal-to-noise ratio (SNR) compared to a reference signal. Statistical tests (P-value < 0.05) were used to analyze the relationships between the mentioned evaluation criteria. Generalized Linear Models revealed that surface reconstruction methods and finite-element interpolation outperformed partial differential equations. Results: The Trimmed mean method achieved the highest R-squared (0.882 ± 0.0610) and lowest RMSE (0.0155 ± 0.00904) with a reconstruction time of 9.5154 microseconds for a 500-millisecond epoch of MEG channel data. Conclusion: The surface reconstruction methods can recover the noisy or lost signal in MEG with a suitable error and required time. These findings support the use of robust statistical strategies for improving MEG signal quality, especially in high-density sensor arrays.