Zero F-stat for gamma variate fitting

Hi, I am currently using 3dNLfim to apply gamma variate model to fit our tracer kinetics data and extract useful parameters such as time-to-peak, onset time, etc.

During this process, I've encountered an issue where all the f-test values were coming out as zeroes even though the resulting fit was very decent.

I've attached my code and snapshot of the fit vs. the signal. I am looking forward to hear back from you soon.

#!/bin/tcsh -xef

# Where is data located? It is FITC-Dextran data that was used to compute the arrival time of the tracer in miniscope paper.
# First: convert each image to text using FIJI. Maybe write a script to do this either in Python or any language.
# 600x600 ROI was used before converting to text images
# Get number of CPUs: afni_system_check.py -check_all | grep CPU

set in_path = "/media/vu/Elements/Vu/FITC/M5/Cropped_FITC_25x25"
set out_path = "/media/vu/Elements/Vu/FITC/M5/AFNI_25x25"
set prefix = "FITC"
# in ms
set tr = 50

# go and delete everything
mkdir "$out_path"
cd "$out_path"
rm ${prefix}*

cd "$in_path"

set num_img = `ls -1 | wc -l`

# make brik
to3d -prefix $prefix -orient LAI -time:tz ${num_img} 1 ${tr} ZERO -datum float *.txt
3drefit -xdel 0.005 -ydel 0.005 -zdel 0.005 -keepcen "$prefix+orig"

# move new briks to output directory
mv ${prefix}+orig* "$out_path"

# change directory
cd "$out_path"

# compute average of the first 300 time points
3dTstat -mean -prefix "${prefix}_avg_baseline" "${prefix}+orig.BRIK'[0..300]'"

# get the max and min offset
set min_offset = `3dBrickStat -min FITC_avg_baseline+orig.BRIK`
set max_offset = `3dBrickStat -max FITC_avg_baseline+orig.BRIK`

# compute constraints
set tol = `python ../calc.py "0.5 *$min_offset"`
echo "++ Tolerance: $tol"

set new_min_offset = `python ../calc.py "$min_offset - $tol"`
set max_offset = `python ../calc.py "$min_offset"`

echo "++ Offset minimum: ${new_min_offset}"
echo "++ Offset maximum: ${max_offset}"

# t0, k, r, b
3dNLfim -input "${prefix}+orig.HEAD"	\
	-inTR					\
	-signal GammaVar 			\
	-noise Constant	 		\
	-nconstr 0 $new_min_offset $max_offset \
	-nabs					\
	-sconstr 0 15 30			\
	-sconstr 1 240 250 			\
	-sconstr 2 4.5 5.1			\
	-sconstr 3 0.4 0.5			\
	-nbest 20				\
	-nrand 19999				\
	-BOTH					\
	-sfit	"${prefix}_signal_fit" 	\
	-snfit	"${prefix}_sig_and_noise_fit"	\
	-bucket 0 "${prefix}_abucket"		\
	-voxel_count				\
	-fdisp	1				\
	-jobs	16				

fit

Oh, I had not seen this post, just the email. Hopefully the F-stats are reasonable now.

This image shows somewhat of a wsinc5/Fourier type of falloff. But can you explain the 2 different baselines?

Thanks,

  • rick