Scaling intensities of images to a common standard scale. The intensity standardization method by Nyul et al. should theoretically work with all types of images (not only medical images). This implementation was built for bone scintigraphy scans, which are basically grayscale images.
Import nyul.py:
import nyulLoad your data and store it in a list:
data = [img_1, img_2, ..., img_N]where data is a list of your images as numpy.ndarrays with shape: (height, width).
Compute the standard scale for a given intensity-landmark configuration:
standard_scale, percs = learn_standard_scale(data, i_min=1, i_max=99, i_s_min=1, i_s_max=100, l_percentile=10, u_percentile=90, step=10)Apply the standard scale to a target image, i.e. normalize the image intensities of the target image to the standard scale:
img_norm = apply_standard_scale(input_image, standard_scale, percs, interp_type='linear')where input_image is the image to be normalized as (numpy.ndarray) with shape: (height, width).
Refer to the jupyter notebook for a detailed explanation of the fundamental problem and the usage:
demo.ipynbRefer to the docstrings for a detailed explanation of the functions and its parameters.
Implementation based on:
- https://github.com/sergivalverde/MRI_intensity_normalization
- https://github.com/jcreinhold/intensity-normalization
- https://gitlab.com/eferrante/nyul/-/tree/c01ad9afc89e9de18cd5ff911877bcbf49777476
It is suggested to use one of those packages for (brain) MRI scans, since they support NIfTI file format.
[1] Nyúl, László G., and Jayaram K. Udupa. "On standardizing the MR image intensity scale" Magn Reson Med. (1999) Dec;42(6):1072-81.
[2] Shah, Mohak, et al. "Evaluating intensity normalization on MRIs of human brain with multiple sclerosis" Medical image analysis 15.2 (2011): 267-282.