"""Masked-ness score by source and derivative zero count ratio for images."""

import numpy as np
import nibabel as nb

INPUT = "/path/to/image.nii.gz"

# =============================================================================

def maskedness_score(img):
    """Compute a number that relates to likelihood of an image being masked.

    The natural range of mask score is between 0 and 1. Closer to 1 indicates
    the image might have been masked. More masked voxels (aka larger maskes)
    leads to scores closer to 1.
    """

    # Compute spatial gradient
    gra = np.asarray(np.gradient(img))
    gra = np.sum(np.abs(gra), axis=0)
    # Mask score is a ratio between an image and its derivative's zero voxels
    count1 = np.sum(img == 0)
    count2 = np.sum(gra == 0)
    score = count2 / count1
    return score


# Load data
nii = nb.load(INPUT)
img = nii.get_fdata()
dims = img.shape

# Compute mask score for the input image
score1 = maskedness_score(img)

# Simulate 'not defaced data' by shuffling voxels
img = img.flatten()
np.random.shuffle(img)
img = img.reshape(dims)
score2 = maskedness_score(img)

print("Maskedness score = {:.3f}".format(score1))
print("Maskedness score after voxel shuffling = {:.3f}".format(score2))
print("(Shuffling is done to simulate non-masked images)".format(score2))