Brain Imaging & Analysis Center

usage: -fithrf flag nparams latency pctsigmin

FITHRF fits a biphasic hemodynamic response function to each voxel in a time series data set. This command assumes that the input data set contains a single response time course, typically generated by averaging multiple individual responses (see the MOVIE command).

• FLAG - Add values to combine options (octal values in parentheses) Default= 013
  • 1 - Calculate HRF fitting parameters (8 params - see below) and return those as output
    • 8 params: delay,rise,peak,width,fall,undershoot,recovery,correl
    • 10 params: delay,rise,peak,width,fall,undershoot,recovery,correl,dip,depth
  • 2 - Loop time course; treat 1st point as though it follows last point in time
  • 4 - Calculate cumulative HRF time parameters; changes meaning of HRF fitting parameter output to:
    • 8 params: delay,startofpeak,width,endofpeak,undershoot,backtobaseline,correl
  • 8 (010) - Smooth input time course before fitting
  • 16 (020) - Smooth slopes
  • 32 (040) - Calculate mean HRF
  • 64 (0100) - Calculate basic time course statistics (9 params - see below) and return those as output
  • 128 (0200) - Correlate
  • 512 (01000) - Output is input replaced by idealized time course based on fitted HRF parameters
  • 1024 (02000) - Two crossings
  • 2048 (04000)- Substitute slopes

• NPARAMS - Number of fitting parameters, Minimum=4 (default= 7,18; if FLAG&4, NPARAMS=9; if FLAG&32, NPARAMS=8)
• LATENCY - (default= 0)
• PCTSIGMIN - (default= 80%)

The hemodynamic response is estimated by fitting 7 time course parameters (see below):

• delay,rise,peak,width,fall,undershoot,recovery

Of these, all are time values except PEAK and UNDERSHOOT, which are amplitude parameters.

Each input time course is fitted independently by estimating the 7 parameters using the following steps:

1. identify maximum, minimum, and mean of time course
2. identify time points where the curve crosses the mean value
3. starting at the peak, find preceding mean-crossing
4. linearly extrapolate slope of curve at rising mean-crossing to find time at base of rise and at peak of rise
5. define delay as time at base of rise and rise as time at onset of peak
6. repeat for the falling phase (find the falling mean-crossing and extrapolate its slope to find the end of the peak and the end of the fall)
7. calculate the time width of the peak (end of peak minus onset of peak)
8. calculate the amplitude of the peak
9. calculate the amplitude of falling undershoot (pre-rise level - post-fall level)
10. calculate recovery time from undershoot minimum to pre-rise baseline amplitude

Once the fitting parameters are have been estimated the fit is checked by using those parameters to generate an idealized hemodynamic response curve and then calculating the correlation between the original time course and the idealized fit. The calculated correlation is reported as goodness-of-fit on a 100 point scale (i.e., the correlation coefficient multiplied by 100).

The first step in the fitting process is to generate a standardized set of statistical parameters for each time course. If you select the FLAG 64 option, those statistical parameters will be returned as output instead of the standard HRF fitting parameters.

The time course statistics returned with that option are:

• vmean - mean valuee
• pctsignal - (vmax-vmin)/vmean X 100
• vmin - minimum value
• vmax - maximum value
• imin - time index of minimum value
• imax - time index of maximum value
• crossings - number of times the time course crosses the mean value
• v0min - minimum non-zero value
• v0max - maximum non-zero value

Primary fitting parameters:

• delay - time from beginning of time course to start of response
• rise - time from beginning of rise to reach peak
• peak - amplitude of peak (peak-baseline)
• width - time from onset of peak to end of peak
• fall - time from end of peak to minimum
• undershoot - amplitude of undershoot (baseline-minimum)
• recovery - time from undershoot minimum to return to baseline

Additional HRF parameters not usually fitted automatically:

• dip - time from delay to pre-rise initial dip minimum
• depth - initial dip amplitude (baseline-minimum)

Goodness of fit parameter:

• correl - correlation coefficient X 100, comparing initial time course to idealized fit

See Also:
fScan Home, fScan Manual, CmdLineOptions, Manual Help