To investigate the variability of HRF across layers, we employed a simple but powerful blind deconvolution technique to recover the latent neural signal at each vertex. Our results showed that all three parameters of region-specific laminar HRF (response height, time-to-peak, and FWHM) varied in reference to cortical depths, and were significantly greater in superficial layers than deeper layers. This finding matches findings from previous HRF studies in animals. Tian et al. found both the onset of the BOLD response and the initial dip rely on cortical depth, and the fastest response was in deeper layers within the rat primary somatosensory cortex [49]. In addition, Yu et al. showed the onsets at different layers coincided with the neural inputs with line-scanning fMRI both in rat somatosensory cortex and motor cortex [47]. We have demonstrated that this is a general fact for almost all cortical regions. The comparison of functional connectivity before and after deconvolution showed the importance and necessity of recovering latent neural signals before any resting state functional connectivity analysis is performed at the laminar level. The functional connectivity post-deconvolution in the latent neural space aligned more closely with the underlying anatomical connections compared to FC obtained on BOLD data.
