Changes to pixel_array do not save to new files

[cheffley6]

Describe the issue

When I attempt to modify a FileDataset's pixel_array, and then save that FileDataset to a path on my local system, and then execute a pydicom.read_file on the new .dcm file, I expect the changes to the previous FileDataset's pixel_array to be present in the FileDataset generated by executing pydicom.read_file on the new .dcm file, but the changes are not present. Instead, it appears as though a copy of the original state of the FileDataset is copied to the new .dcm file.

Here is my env_info:

module       | version
------       | -------
platform     | macOS-10.16-x86_64-i386-64bit
Python       | 3.8.13 | packaged by conda-forge | (default, Mar 25 2022, 06:06:49)  [Clang 12.0.1 ]
pydicom      | 2.3.0
gdcm         | 2.8.9
jpeg_ls      | _module not found_
numpy        | 1.22.3
PIL          | 8.4.0
pylibjpeg    | _module not found_
openjpeg     | _module not found_
libjpeg      | _module not found_

Steps to reproduce

filename = "<path to a dcm file>"
dcm = pydicom.read_file(filename)

dcm.pixel_array[0:1, 0:1] = [[5]]
dcm.save_as("temp_file.dcm")

print(dcm.pixel_array)

second_dcm = pydicom.read_file("temp_file.dcm")
print(second_dcm.pixel_array)

[mrbean-bremen]

pixel_array is not the same as PixelData - it is a NumPy array containing the pixel data in uncompressed form.
See the documentation for more information. Basically, you can write back pixel_data for uncompressed transfer syntaxes by converting them into a byte array and setting it to the PixelData element, so in the simplest case it would be:

dcm = pydicom.read_file(filename)
dcm.pixel_array[0:1, 0:1] = [[5]]
dcm.PixelData = dcm.pixel_array.tobytes()
dcm.save_as("temp_file.dcm")

For compressed transfer syntaxes, the process is more complicated, as pydicom generally does not support compressing image data (except for the RLE transfer syntax).