Parameters for MRI brain normalization and extraction feature

[biobob]

Dear All,

I have used pyradiomic on  3D brain MRI. I use the defaulted parameters and I start change some parameters:

 label: 1
  correctMask: True
  normalize: True
  normalizeScale: 1
  interpolator: 'sitkBSpline' # This is an enumerated value, here None is not allowed
  resampledPixelSpacing: # This disables resampling, as it is interpreted as None, to enable it, specify spacing in x, y, z as [x, y , z]
  weightingNorm: # If no value

So I have this questions:

1) What is the important parameters we need to use for correct extraction of the data from brain MRI?

2) How can measere the quality of the extraction obtained?

3) Only the column start from original_ word are the feature?

Thanks in advance for any help

M

[Joost van Griethuysen]

Ad 1) and 2) I'm afraid I don't have a definitive answer for you here. My only advise is to check the examples, and to see what's used in literature. As to measuring the quality of the extraction, that really depends on what you want to measure.

Basically, all settings are 'correct' in the way that they describe you texture in a certain way. Still, some settings will lead to more robust values and some settings may produce values that will yield higher accuracy models.

As to measuring robustness, an ICC is usually used. For predictive power, you need to develop and validate your model.

ad 3) Not exactly. The output of PyRadiomics consists of all input columns + diagnostic features (starting with `diagnostic_`) + feature values (named as `<filter>_<class>_<name>`). `original_` prefix only indicates features extracted from non-filtered images.

Regards,

Joost

Op dinsdag 8 oktober 2019 12:59:25 UTC+2 schreef biobob: