Enhanced Neural Decoding With Optically Pumped Magnetometer MEG Using Multivariate Pattern Analysis

Multivariate pattern analysis (MVPA) provides a sensitive means to decode distributed neural activity from magnetoencephalography (MEG) by jointly exploiting temporal and spatial information. Optically pumped magnetometer (OPM)-based MEG enables close-to-scalp measurements and is therefore expected to capture neuromagnetic fields with higher spatial resolution compared to conventional systems based on superconducting quantum interference devices (SQUIDs). In this study, we conducted a within-subject comparison of OPM- and SQUID-MEG using time-resolved MVPA to decode neural responses to visual objects presented as images and corresponding words, recorded from the same participants under an identical experimental paradigm and matched preprocessing. To isolate the contribution of spatial sampling, decoding performance was evaluated under controlled sensor counts and spatial-frequency content, quantified using a spherical harmonic expansion of the sensor topographies. A complementary full-array SQUID analysis was also included as a reference. The results indicated that OPM-MEG achieved higher decoding accuracy than SQUID-MEG in the sensor-matched comparison, with the advantage being particularly evident for word decoding, where OPM also exceeded the full-array SQUID across a broad range of sensor counts. Further analysis of spatial-frequency content revealed that OPM-MEG benefited from the inclusion of higher-order spatial components. These findings demonstrate that OPM-MEG enhances the recoverability of fine-grained neural representations for multivariate decoding, with implications for cognitive neuroscience research and neural engineering applications such as brain-computer interfaces.