def kde_normalize(img, mask=None, contrast='t1', norm_value=1):
"""
use kernel density estimation to find the peak of the white
matter in the histogram of a skull-stripped image. Normalize
the WM of the non-skull-stripped image to norm_value
Args:
img (nibabel.nifti1.Nifti1Image): target MR image
mask (nibabel.nifti1.Nifti1Image): brain mask of img
contrast (str): contrast of img (T1,T2,FA,MD)
norm_value (float): value at which to place WM peak
Returns:
normalized (nibabel.nifti1.Nifti1Image): WM normalized img
"""
if mask is not None:
voi = img.get_fdata()[mask.get_fdata() == 1].flatten()
else:
voi = img.get_fdata()[img.get_fdata() > img.get_fdata().mean()].flatten()
if contrast.lower() in ['t1', 'flair', 'last']:
wm_peak = hist.get_last_mode(voi)
elif contrast.lower() in ['t2', 'largest']:
wm_peak = hist.get_largest_mode(voi)
elif contrast.lower() in ['md', 'first']:
wm_peak = hist.get_first_mode(voi)
else:
raise NormalizationError(
'Contrast {} not valid, needs to be `t1`,`t2`,`flair`,`md`,`first`,`largest`,`last`'.format(contrast))
normalized = nib.Nifti1Image((img.get_fdata() / wm_peak) * norm_value,
img.affine, img.header)
return normalized
def whitestripe(img,
contrast,
mask=None,
width=0.05,
width_l=None,
width_u=None):
"""
find the "(normal appearing) white (matter) stripe" of the input MR image
and return the indices
Args:
img (nibabel.nifti1.Nifti1Image): target MR image
contrast (str): contrast of img (e.g., T1)
mask (nibabel.nifti1.Nifti1Image): brainmask for img (None is default, for skull-stripped img)
width (float): width quantile for the "white (matter) stripe"
width_l (float): lower bound for width (default None, derives from width)
width_u (float): upper bound for width (default None, derives from width)
Returns:
ws_ind (np.ndarray): the white stripe indices (boolean mask)
"""
if width_l is None and width_u is None:
width_l = width
width_u = width
img_data = img.get_data()
if mask is not None:
mask_data = mask.get_data()
masked = img_data * mask_data
voi = img_data[mask_data == 1]
else:
masked = img_data
voi = img_data[img_data > img_data.mean()]
if contrast.lower() in ['t1', 'last']:
mode = hist.get_last_mode(voi)
elif contrast.lower() in ['t2', 'flair', 'largest']:
mode = hist.get_largest_mode(voi)
elif contrast.lower() in ['md', 'first']:
mode = hist.get_first_mode(voi)
else:
raise NormalizationError(
'Contrast {} not valid, needs to be `t1`,`t2`,`flair`,`md`,`first`,`largest`,`last`'
.format(contrast))
img_mode_q = np.mean(voi < mode)
ws = np.percentile(voi, (max(img_mode_q - width_l, 0) * 100,
min(img_mode_q + width_u, 1) * 100))
ws_ind = np.logical_and(masked > ws[0], masked < ws[1])
if len(ws_ind) == 0:
raise NormalizationError(
'WhiteStripe failed to find any valid indices!')
return ws_ind