In that paper, they use a method similar to the one used in our 3ddelay program, that computes the maximal correlation of shifted time series. You may want to take a look at that program. Also consider the approach used here by Bellgowan, Saad and Bandettini:
[
www.pnas.org]
Also we have recently published a paper using 3dNLfim, our non-linear fitting program, to compute latencies among HRF curves using a different model that worked well with our data. Many of the details are in the supplementary materials.
[
dx.doi.org]