Dicom to PNG Conversion - output image is completely black

user_11016875 0
Viewed 1528

While conversion of dicom to png using Python, without getting any error. Output file is complete Black, although image array have variable values.

Code:

import pydicom
import cv2
import os

dicom = pydicom.dcmread(inputdir + f)
# print("dicom", dicom)
img = dicom.pixel_array
print("img", img)
cv2.imwrite(outdir + f.replace('.dcm','.png'),img)

Image array (for reference):

[[585 585 570 ... 570 572 570]
[585 585 585 ... 572 575 572]
[585 585 585 ... 553 568 575]
...
[854 854 854 ... 778 783 787]
[854 854 856 ... 783 785 787]
[854 856 856 ... 785 790 759]]
pydicom opencv python
2 Answers

According to docs.opencv.org, cv2.imwrite generally "prefers" the image data to be in 8-bit representation (i.e. value ranges from 0 to 255 only).

In general, only 8-bit single-channel or 3-channel (with 'BGR' channel order) images can be saved using this function ...

I noticed that your image data is more than 8-bit, hence you will need to either (a) scale it then cast it to np.uint8, or (b) to reduce the bit representation to 8 bit.

Example of (a) scaling:

import numpy as np  # assuming that you have numpy installed
img = np.array(img, dtype = float) 
img = (img - img.min()) / (img.max() - img.min()) * 255.0  
img = img.astype(np.uint8)
cv2.imwrite(outdir + f.replace('.dcm','.png'),img)

or, Example of (b) bitshifting to 8-bit:

bit_depth = 10  #  assuming you know the bit representation
img = np.array(img, dtype = np.uint16)
img = img >> (bit_depth - 8)  
img = img.astype(np.uint8)
cv2.imwrite(outdir + f.replace('.dcm','.png'),img)

But since you are saving your image in PNG format, here is shorter way...

, with these exceptions:

  • 16-bit unsigned (CV_16U) images can be saved in the case of PNG, JPEG 2000, and TIFF formats
  • ...

you can just cast your image to np.uint16 :)

img = img.astype(np.uint16)
cv2.imwrite(outdir + f.replace('.dcm','.png'),img)
user_15509960

Adding the next line to code after "dicom.pixel_array" will do the trick.

## Rescaling grey scale between 0-255
scaled_img = (np.maximum(img,0) / img.max()) * 255.0
user_110168750
Related