Best practice for ordering T1w preprocessing steps (Denoise, N4, lesion filling)

[ChristineFarrugia]

Hi everyone,

I’m planning to preprocess T1-weighted images using the following steps:

• ANTs DenoiseImage
• Truncation of outlier intensities
• ANTs N4BiasFieldCorrection
• Contrast harmonization (custom script)
• ANTs LesionFilling

Could anyone advise whether this is a reasonable order for applying these steps, or suggest any modifications?

Thanks in advance!
Christine

[cookpa]

This looks reasonable to me. If you have extreme outliers then DenoiseImage might work better if you truncate those first.

You might also try out the ANTsPyNet inpainting

https://github.com/ANTsX/ANTsPyNet/blob/master/antspynet/utilities/inpainting.py

[ChristineFarrugia]

Many thanks!

[ChristineFarrugia]

Hi,
Sorry to bother you again, but I’m wondering what the best method might be for detecting extreme outliers in a fairly large set of images. I’ve been thinking about flagging intensities outside the range of 3×[1st percentile, 99th percentile], but that’s more of a rough idea than a rigorous approach. Are there any established rules of thumb for this kind of problem?

[ntustison]

If you're looking to preprocess the images to remove outliers, as @cookpa mentioned, you might want to do an initial truncation (below for the ImageMath call). And your numbers of 0.01 and 0.99 are closer to what's actually used in practice than the defaults listed in the function call.

% ImageMath | grep TruncateImageIntensity
  TruncateImageIntensity: 
      Usage        : TruncateImageIntensity InputImage.ext {lowerQuantile=0.05} {upperQuantile=0.95} {numberOfBins=65} {binary-maskImage}

[ChristineFarrugia]

I was initially thinking of running a test to check whether any truncation is required, but from you've said I guess I can just truncate above 0.99 and below 0.01 and keep things simple. Thanks again!