1) block study design

• AB...A design (start and and with the same condition to minilmalize effects of monotonic drift)
• optimal task block length - 15-20 sec (assuming 2-3 sec of possible delay, may be less for sensory motor studies), this result is independent from the length of imaging series.

2) choice of preprocessing steps

• spatial smoothing init_gaus_filt(1) the default value of 1 applies the gaussian filter of half widthh at half maximum equal to pixel  size
• high-pass filtering (a.k.a Chris filter) at 0.35 of the stimulus frequency (Chris default value is 0.5, so you have to use it with increased number of OnOff Blocks ChrisFilt(OnOffNUmber*1.4)

other available options:
 
• temporal normalization - does NOT help
• polynominal (linear, quadratic ...) trend_subtr drift removal help (but less than high-pass filter)

3) Statistics (applied to a SINGLE run)

• Cross-correlation, t_statistic, and Mann-Whitney are best
• signal change (%), Fourier and split statistics are BAD
• skew correction is helpful only for unbalanced AB..AB design

4) Spatial smoothing

• Initial Gaussian smoothing is BETTER that final cluster filtering
• multifiltering (adding maps with and without smoothing is BEST (but complicated)

5) Between identical run consolidation

• Mean of individal t-maps is better than taking lowest or highest value (it has an additional advantage of being insensitive to changes in number of runs).
• Running statistics without regard to breaks in imaging runs is BAD
• averaging runs works BEST (assuming that runs are IDENTICAL)Ý this can be applied by applying theÝ postprocessing function ColapseSeries
• this function reduces data set, to images collected in the first data series, those images are replaced by averages of images collected at the same time in all imaging series