Error in seq.default(logpmin, logpmax, length.out = Nh) : 'to' must be a finite number

[matthigger]

I'm struggling with what looks like a numerical issue. Running this stat file and mask seem to get me in trouble:

library(oro.nifti) 
library(pTFCE) 
Z=readNIfTI('stat.nii.gz') 
MASK=readNIfTI('mask.nii.gz') 
pTFCE=ptfce(Z, MASK) 

Estimating smoothness based on the data...
|=============================================================| 100%
Error in seq.default(logpmin, logpmax, length.out = Nh) :
'to' must be a finite number

here are the files I'm working with. I checked to ensure all values are above board, the statistic is positive across the whole mask.

mask.nii.gz
stat.nii.gz

(Also, thanks for the great software! the download / install / test example all went beautifully)

[spisakt]

Hi @matthigger,
Are you sure your input is a valid z-score image?
It does not look like any real data application I have seen before.
You have some unrealistically high Z-scores, that's why pTFCE crashes with overflow.

Depending on hardware, Z-score greater than about 35 will cause this error.
If you are sure that you want to enhance this map, consider truncating Z values at a reasonable level.
It shouldn't affect the results of the enhancement very much.

Cheers,
Tamas

[matthigger]

Hi @spisakt,

You're right, they aren't Z scores, I'm simulating effects to measure sensitivity. I suspect that these super high scores (T^2 stats) occur only in the obvious effects which are easily recovered, so I bet I'll only need to threshold for those straightforward examples which aren't so interesting anyways.

Thanks again for the speedy response. Cheers,

-Matt

[spisakt]

Hi Matt,
Please note that pTFCE is only valid with Z-scores as input!
Otherwise you are right, practically no enhancement is needed for very high peaks, thus truncating peaks to e.g. 30 will still do the job and enhnace low-intensity border regions equally well.

Cheers,
Tamas

[matthigger]

Hi Tamas,

Sorry, I misunderstood that bit. I'll be sure to represent it as a Z score in the analysis.

Best,

-Matt