My RGB ultrasound image has strange colours

[Dharanikanna]

I'm trying to read medical images from the Dicom file using the python language and pydicom library. I can able to read the file but the image has some noise. If I read the same file using weasis(Dicom reader) it looks fine.

import pydicom as dicom
import matplotlib.pyplot as plt
import cv2 as cv

data=dicom.dcmread('image-00001.dcm')
image=(data.pixel_array[0])
plt.imshow(image,cmap=plt.cm.bone)
plt.show()

Please suggest an idea to resolve this problem.

Thanks in advance!

[mrbean-bremen]

Can you show how the real image looks (in Weasis), or post an anonymized version of the image? I'm not sure what you mean by "noise".
It could be a problem with the wrong color space, or just a missing LUT, but without more information it is difficult to say.

[mrbean-bremen]

I actually don't see anything here that looks wrong. PhotometricInterpretation is RGB, but the colorspace seems to be something else. As PhotometricInterpretation and colorspace don't seem to match, I don't know how to convert the colorspace to get the correct result.

I you use pillow, you could also try to install gdcm (pip install python-gdcm) to check if the output is different if gdcm does the decoding.

You seem to have an anonymized version of the DICOM file - maybe you can attach the file here so we can have a look.

[scaramallion]

You probably need to apply a YCbCr to RGB colour space conversion using convert_color_space().

from pydicom import dcmread
from pydicom.pixel_data_handlers import convert_color_space

ds = dcmread(...)
rgb = convert_color_space(ds.pixel_array, "YBR_FULL", "RGB", per_frame=True)

[Dharanikanna]

Thank you @scaramallion , Your suggestion resolved the issue. @mrbean-bremen Thank you for your support and suggestions.

If possible please explain why this issue happens while reading the image from the Dicom file?

[scaramallion]

The JPEG standard (ISO/IEC 10918-1) is agnostic as to the colour space that the encoded image data should be in. In other words, it can be monochrome, RGB, CMYK, YCCC, YCbCr or any other colour space. However, RGB data is usually more efficiently compressed if it's first converted to the YCbCr colour space. The problem is that the JPEG standard also doesn't have any mandatory way of tracking what exactly that colour space was at the time of encoding, although there are a couple of optional ways to do so.

The DICOM standard has a mandatory element that's intended to be used for this: (0028,0004) Photometric Interpretation. If this value is set correctly (and it sometimes isn't), then it's supposed to be the colour space of the JPEG encoded data (i.e. after any colour space transforms have occurred) as currently present in the dataset.

When you use Dataset.pixel_array, Photometric Interpretation will (hopefully) be the colour space of the returned pixel array. There are a few reasons we do this:

• This is actually the raw data present in the dataset and we try to avoid implicitly modifying the dataset information
• Colour space conversions are lossy (for JPEG 10918), and the user should decide if the data loss is acceptable
• The user may want the pixel data in the encoded colour space (this is useful when transcoding, for example)

So if Photometric Interpretation is "YBR_FULL_422", the returned ndarray will be in the YCbCr colour space, and to get to RGB you have to perform your own colour space conversion using convert_color_space() or similar third-party function.

[Dharanikanna]

Thank you @scaramallion this was really very helpful information.

[hubutui]

Hi, guys. I was going to report a similar issue when I got here.
Here is an anonymized sample file. Note the DICOM modality is ES (Endoscopy), but the camera is not inside the body yet, so there is no patient information in the image.

code to read and save jpg image:

from pydicom import dcmread
from PIL import Image

filename = "output.dcm"
ds = dcmread(filename)
# ds.PhotometricInterpretation = "YUV"
Image.fromarray(ds.pixel_array).save(f"output.pydicom.{pydicom.__version__}.jpg")

1. pydicom == 2.1.2 works as expected.
2. pydicom > 2.1.2, I got an image with wrong color space.
3. pydicom > 2.1.2, with ds.PhotometricInterpretation = "YUV", it works as expected.

@Dharanikanna Could you also try pydicom 2.1.2?
@scaramallion @mrbean-bremen Something changes for pydicom > 2.1.2 might also lead to this issue?

[scaramallion]

Your source image data was converted to YCbCr colorspace prior to JPEG compression so you need to convert the output back to RGB. The value of the Photometric Interpretation is incorrect in the dataset, it should be YBR_FULL:

from pydicom import dcmread
from pydicom.pixel_data_handlers import convert_color_space

ds = dcmread("output.dcm")
ybr = ds.pixel_array
rgb = convert_color_space(ybr, "YBR_FULL", "RGB")

[hubutui]

Yes, but I just don't know why pydicom 2.1.2 works. I used to think it's a new bug introduced after 2.1.2.