The initial decrease in 7T-BOLD signals detected by hyperalignment contains information to decode facial expressions

Abstract

The initial decrease in the blood oxygenation level-dependent (BOLD) signal reflects primary neuronal activity more than the later hemodynamic positive peak responses. Moreover, ultra-high field BOLD has high sensitivity for the initial de-oxygenation signal. However, it is not fully understood how much information about task events and cognitive processes the initial decrease in the BOLD signal contains. Multivoxel pattern analysis (MVPA) of the BOLD signal has enabled the quantification of information contained in the activity patterns, but it has mainly relied on the positive peak responses. Here, we applied a signal-based functional inter-individual alignment algorithm (i.e., hyper-alignment) to a 7T-BOLD timeseries scanned while participants conducted a facial expression discrimination task. We found that the MVPA decoding accuracy in the bilateral amygdala 2 s after the face onset was significantly beyond chance. Furthermore, we confirmed that the voxels contributing to the decoding accuracy at 2 s displayed a decreasing hemodynamics response. These results demonstrated that the initial decrease in 7T-BOLD signals contains finer information about task events and cognitive processes than thought previously.