FLOBS HRF reconstruction and latency estimation — feedback on method?

Hi all,

I’m working with flexible (FLOBS) HRF modeling and . I wanted to get community feedback on how we estimated latency using the FLOBS .

FLOBS gives three PEs per voxel (PE1, PE2, PE3), which are the weights on the three optimal basis functions. These don’t correspond to amplitude/timing/dispersion like in the Gamma + derivative approach, so PE2/PE1 is not a meaningful latency measure.

Instead, we reconstructed the full HRF curve at each voxel within the ROI using:

HRF(t) = PE1 × basis1(t) + PE2 × basis2(t) + PE3 × basis3(t)

where basis1, basis2, basis3 are the 559-timepoint FLOBS waveforms spanning 0-28 seconds. We did this at every voxel within the ROI, then averaged the reconstructed curves across voxels to get one curve per subject. We then found the time-to-peak of each subject’s curve, and ran a one-sample t-test against 5.5 seconds (canonical peak) across all subjects.

Questions:

  1. Is comparing the FLOBS time-to-peak against 5.5 seconds (canonical) a valid approach for testing latency shifts, or is there a more appropriate reference or statistical method?

  2. Any other suggestions on how to properly extract and compare latency between a fixed-shape model (Gamma) and a flexible model (FLOBS)?

  3. If any comment about the approach used for constructing the HRF curve that be highly appreciated as well.

Thank you and any comments or direction would be great!