Nishemo

Nishemo is a utility for generating model hemodynamic filters based on the Gauss, Poisson, gamma or identity functions.

Invocation

java  Nishemo 

Nishemo window

General Fields

• Output file
  Name of the ascii file that will contain the filter values.
  Filter notes: This program will always produce a zero centered odd length filter, so that it can be easily used by the nisfilter program, which uses the Numerical Recipes in C conv() function. The duration that is entered (see below) is always measured from the starting point t=0. The user does not need to deal with the filter values for timepoints where t < 0. These are simply automatically generated so as to make a temporally centered filter. The filter text file has 2 colunms. First column is the timepoint (in seconds) when the sample was taken. Second column is that value of the filter at that timepoint.
• Filter duration
  Temporal duration of the filter that will be constructed, in seconds (not TR units).
  Note: the number of elements in the filter for timepoints > 0 will be (filter duration)/(sampling interval).
• Sampling interval
  The temporal duration of each element in the filter. In discrete signal processing terms the sampling interval is equal to 1/fs, where fs is the sampling frequency. Typically, one would enter the TR for fMRI time series data in this field.
  

Hemodynamic Functions

• gauss
  Select Gaussian model function. The equation for the Gaussian model is:
  s(t) = [1/(sqrt(2*pi*sigma²))] * exp(- (t-mean)²/sigma²)
  Where t is the timepoint, sigma is the width (variance) of the Gaussian distribution, and mean is the mean (ie center) of the Gaussian distribution. exp is the exponential function (e^x).
  The model parameters to be entered for the Gaussian function in the Filter Parameters section of the GUI are:
  • mean: The mean of the Gaussian distribution. This corresponds to the lag of the hemodynamic filter.
  • FWHM: The Full Width Half Max of the Gaussian distribution. This controls the dispersion of the hemodynamic filter. FWHM is a function of the sigma variable in the equation:
    sigma = fwhm / (2.0 * sqrt(-2.0 * log(0.5))) log is natural log.
  
  
• poisson
  Select Poisson model function. The equation for the Poisson model is:
  s(t) = lambda^t * [exp(-lambda) / t!]
  Where t is the timepoint, exp is the exponential function (e^x), ^ is the exponentiation operator (x^y means x raised to the y power), and ! is the factorial function.
  The model parameter to be entered for the Poisson function in the Filter Parameters section of the GUI is:
  • lambda: lambda is a single parameter controlling both the lag and dispersion of the Poisson function. Lag is equal to dispersion because in a Poisson distribution mean is equal to variance. Note that because of the factorial function, which is only defined for integer values, t must be an integer value. This means that the Sampling Interval entered in the first section of the GUI must be an integer if you select the Poisson function. Scans with a non-integral TR cannot use the Poisson model.
  
  
• gamma
  Select gamma model function. The equation for the gamma model is:
  s(t) = [ theta2 * [(theta2*t)^(theta1 - 1)] * exp(- theta2*t)] / gamma(theta1)
  Where t is the timepoint, ^ is the exponentiation operator, exp is the exponential function (e^x), and gamma is the mathematical gamma function.
  The model parameters to be entered for the gamma function in the Filter Parameters section of the GUI are:
  • theta1 and theta2. Theta1 and Theta2 do not uniquely determine lag and dispersion. ie, there are different combinations of theta1 and theta2 that may yield essentially the same lag and dispersion values. Very roughly, you can think of theta1 as a shape parameter and theta2 as a scale parameter. Or, lag is determined by theta1/theta2 and dispersion is determined by theta1/theta2²
  
  
• identity
  Select identity model function. The equation for the identity filter is:
  s(t) = 0 when t does not equal shift
  s(t) = scale when t=shift
  
  The model parameters to be entered for the identity function in the Filter Parameters section of the GUI are:
  • shift: how many units (units are in sampling interval units) to shift the signal
  • scale: signal is scaled by this factor.
  Notes: The effect of convolving a signal with this filter is to shift and scale the given signal:
  s(t) = s(t-shift)*scale
  The identity filter is primarily used for debugging purposes, but, it could be used to rescale a dataset.
  

Options

• Use output run window
  Check this box if you want the output run window displayed. This will show the command line call made to the nishemo C program, and any error diagnostics from that program.
• View results
  Check this box to have a plot of the selected filter displayed. Note that for continuous functions (Gauss, gamma) the filter plot (green color) will be overlaid on a densely sampled version (red color) of the filter, to make the discrete sampling more clear. The green dots are your actual filter values. The green lines just connect the green dots for visual effect.
  If you use the spreadsheet button on the graph window, you will get a number of columns in the spreadsheet, all with filter values. This is because a number of copies of the filter were overlaid in the graph to get the densely sampled red background, and, to get the dots on top of the curves. The spreadsheet shows everything that went into the graph. The last column should always be your actual filter values.
  
• Overwrite output file
  Check this box if you would like to automatically overwrite the filter output file with the current model selection. This is useful if you are experimenting with many model parameters and only want to save your most recent choice.

References

Boynton, G.M., Engel, S.A., Glover, G.H., Heeger, D.J., "Linear Systems Analysis of Functional Magnetic Resonance Imaging in Human V1.", The Journal of Neuroscience, July 1, 1996, 16(13):4207-4221.
Lange, N., Zeger,S., "Non-linear Fourier Time Series Analysis for Human Brain Mapping by Functional Magnetic Resonance Imaging.", Journal of the Royal Statistical Society, Applied Statistics, 1997, 46 No. 1 pp 1-29.
Press, W.,H., et al, Numerical Recipes in C, Cambridge University Press, 1988.
Rajapakse, J.C., Kruggel, F., Maisog, J.M., Cramon, D., "Modeling Hemodynamic Response for Analysis of Functional MRI Time-Series.", Human Brain Mapping, 6:283-300(1998).

• Menubar
• Toolbar
• Output window