Behavior N4Biasfieldcorrection

I have a behavior which is strange using N4Biasfieldcorrection on a brain with skull. As mentionned in another post, i check range of image intensities and rescale to the range [10, 100] then unscaled…

1. First case: default parameters and brain with skull. Results show hole in brain. (attached result), range seems to affect result
2. Same brain but skull-stripped. Results seems OK (attached result)
3. First case but using ANTs in command line linux. Result is OK

Use of last version of ANTspy from source.

import numpy as np
import ants
from typing import Tuple

def _min_max_scaling(image: np.ndarray, scaling_range: Tuple[int, int] = (10, 100)) -> Tuple[np.ndarray, float, float]:
    image_scaled = np.copy(image).astype(np.float32)
    min_ = (scaling_range[0] - np.min(image)).astype(np.float32)
    scale_ = ((scaling_range[1] - scaling_range[0]) / (np.max(image) - np.min(image))).astype(np.float32)
    image_scaled *= scale_
    image_scaled += min_
    print(np.min(image_scaled))
    return image_scaled, min_, scale_

def _invert_min_max_scaling(image_scaled: np.ndarray, scale_: float, min_: float) -> np.ndarray:
    image_scaled -= min_
    image_scaled /= scale_
    return image_scaled

image = ants.image_read("sample_t1.nii.gz", reorient=True)
image_scaled, min_, scale_ = _min_max_scaling(image.numpy())
image = image.new_image_like(image_scaled)
img_n4 = ants.n4_bias_field_correction(image, verbose=True)

image_unscaled = _invert_min_max_scaling(img_n4.numpy(), scale_, min_)
img_n4 = image.new_image_like(image_unscaled)
ants.image_write(img_n4, "sample_t1_n4.nii.gz")

[ntustison]

There is no 'right" scaling range just as there is no absolute intensity range in MRI. Just make sure the range is greater than at least 1 as the intensities are log transformed.

[Alxaline]

Thank you @ntustison