How to optimize the pixel threshold during dicom to png conversion?

[rasunag27]

Dear all,

I am trying to convert Dicom files to png format. For this purpose, I extracted the code from this issue issue. I changed the saving using cv2.imwrite. Here is my code:

image_path = '/home/Head_images'
for n, image in enumerate(Head_images):
    ds = dicom.read_file(path + '/' + image, force = True)
    shape = ds.pixel_array.shape
    image_2d = ds.pixel_array.astype(float)
    image_2d_scaled = (np.maximum(image_2d, 0) / image_2d.max()) * 255.0
    image_2d_scaled = np.uint8(image_2d_scaled)
    
    cv2.imwrite(os.path.join(image_path, image)+'.png', image_2d_scaled)

From the code, I am able to convert the dicom to png format, but the color-code are in a reverse way. I have attached the original which is required and the one which I have obtained from the code run. What needs to be changed in order to make the code as in original image?

Original_image_link: Original_image
Image_from_code: Image_obtained

Any leads will be appreciated.

Regards,
Sunag R A.

[scaramallion]

You have a couple of issues:

• You're not taking into account that the sample values (SV) for CT may be signed
• You're downsampling based on the maximum SV in each image, rather than the maximum possible range. If one image has an SV range 0 to 10 and another has a range 0 to 2048 then you're not going to get a consistent looking volume.

if "PixelData" not in ds:
    raise AttributeError("This script only works with Pixel Data")

arr = ds.pixel_array
# Apply rescale and windowing prior to processing

arr = arr.astype("float32")
if ds.PixelRepresentation == 1:
    # Sample values are signed, range -2 ** (BitsStored - 1) to 2 ** (BitsStored - 1) - 1
    # Convert to unsigned in range 0 to 2 ** BitsStored - 1
    arr += 2 ** (ds.BitsStored - 1)

# Sample values are in range 0 to 2 ** BitsStored - 1
# Downsample to range 0 to 255
arr = arr / (2 ** ds.BitsStored - 1) * 255
arr = np.clip(arr, 0, 255, dtype='u1', casting='unsafe')

You may also want to perform rescale and windowing operations prior to processing.