Okay, just a quick update on this issue. It appears that there's something about running KK through the ANTsPy or ANTsR interface which causes a "failure", i.e. instances where the thickness value(s) blow up beyond what is expected. Although KK is deterministic, the problem occurs completely at random, occurs with a single thread or multiple threads, and the deviation from the correct or expected thickness image occurs in the first couple of optimization iterations. Also, as a check, it does not occur when running it via the command line.

Below are minimal examples comparing the three scripts running KK on the same data in a loop of 10 iterations: shell, python, and R. On my machine, the shell script produces 10 images which are exactly the same and have the expected values. The R and Python scripts produce the expected image anywhere from 50% to 80% of the time (very loose estimate).

I'll be diving deeper and seeing which exact internal image variable deviates but just wanted to provide an update. Also, any thoughts are welcome on why running KK through R and Python causes this.

Shell

#! /usr/bin/sh

export ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS=8

inDirectory=./

seg=${inDirectory}/seg.nii.gz
gm=${inDirectory}/gm.nii.gz
wm=${inDirectory}/wm.nii.gz

iter=3

for i in {0..9};
  do 
    echo "KKsh $i"
    kk=${inDirectory}/sh_kk_${i}.nii.gz
    KellyKapowski -d 3 \
                  -s $seg \
                  -g $gm \
                  -w $wm \
                  -v 1 \
                  -m 1.5 \
                  -r 0.0025 \
                  -t 10 \
                  -c [${iter}] \
                  -x 0 \
                  -o $kk 
  done

Python

import ants
import os

os.environ['ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS'] = '8'

in_directory = "./"

seg = ants.image_read(in_directory + "seg.nii.gz")
gm = ants.image_read(in_directory + "gm.nii.gz")
wm = ants.image_read(in_directory + "wm.nii.gz")

iter=3

for i in range(10):
    print("KKpy " + str(i))
    kk = ants.kelly_kapowski(s=seg, g=gm, w=wm,
                            its=iter, r=0.0025, m=1.5, x=0, t=10,
                            verbose=1)
    ants.image_write(kk, in_directory + "py_kk_" + str(i) + ".nii.gz")

R

library( ANTsR )

Sys.setenv(ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS = "8")

inDirectory <- "./"

seg <- antsImageRead( paste0( inDirectory, "seg.nii.gz" ) )
gm <- antsImageRead( paste0( inDirectory, "gm.nii.gz" ) )
wm <- antsImageRead( paste0( inDirectory, "wm.nii.gz" ) )

iter <- 3

for( i in seq.int(9) )
  {
  kk <- kellyKapowski( s = seg, g = gm, w = wm, its = iter, 
                       r = 0.0025, m = 1.5, x = FALSE, t = 10, 
                       verbose = TRUE )
  antsImageWrite( kk, paste0( inDirectory, "r_kk_", i, ".nii.gz") )   
  }

gm.nii.gz
seg.nii.gz
wm.nii.gz