Representation similarity analysis based on spatial projection: Decoding of semantic congruity with OPM-MEG and EEG.

Semantic processing is one of the core functions of human higher-level cognition. The neural mechanism of the decoding process is attached to great significance for understanding language, cognition, and clinical applications. The key issue in decoding semantic representation is how to enhance the sensitivity of task-related neural signals to effectively distinguish the spatial patterns under different stimuli. However, current studies are limited by the sensitivity to task-irrelevant noise, the trade-offs in spatial and temporal resolution inherent in neuroimaging modalities. Therefore, this study proposed a novel representation similarity analysis method based on spatial projection. By integrating optically pumped magnetometer magnetoencephalography (OPM-MEG) with multi-channel electroencephalography (EEG), we conducted a semantic congruity decoding analysis. The results showed that the optimized RSA framework significantly delineated the deflection of neural representations to different stimuli. Notably, semantic processing exhibited clear frequency-band specificity, where low-frequency bands (particularly δ and θ) accounted for a substantially larger proportion of semantic representation variance compared to higher frequencies. Further analysis revealed that semantic processing involved the refined spatiotemporal evolution of neural patterns, and semantic incongruity significantly enhanced the similarity of patterns. In addition, the source localization results not only verified the classical language network, but also found the involvement of the limbic system (such as the parahippocampal gyrus and insula), suggesting that semantic processing involved a wider range of memory and regulatory networks. This study not only expands the understanding to semantic neural representations from the multi-dimensional aspects, but also highlights the potential of spatial projection optimization combined with multivariate analysis for high-precision neural decoding.