Issue with Multi-Frame DICOMs in Dicom Image Redactor

[erencanuysal]

Hello,

I’m encountering an issue when using the DICOM Image Redactor with multi-frame DICOM images (where the value in DICOM tag [0x0028, 0x0008] is more than one). The de-identification process works fine for DICOMs with a single frame (pixel_array.shape ≤ 3), but for multi-frame images (where pixel_array.shape > 3), the redaction seems to fail or not function as expected.

Specifically, the issue occurs in the _save_pixel_array_as_png() method of DicomImageRedactorEngine. The code attempts to flatten the pixel array for multi-frame images and save them as PNG files, leading to errors:

if is_greyscale:
    with open(f"{output_dir}/{output_file_name}.png", "wb") as png_file:
                w = png.Writer(shape[1], shape[0], greyscale=True)
                w.write(png_file, pixel_array
else:
    with open(f"{output_dir}/{output_file_name}.png", "wb") as png_file:
                w = png.Writer(shape[1], shape[0], greyscale=False)
                # Semi-flatten the pixel array to RGB representation in 2D
                pixel_array = np.reshape(pixel_array, (shape[0], shape[1] * 3))
                w.write(png_file, pixel_array)

For multi-frame images, such as with a pixel array shape of (135, 600, 800, 3), the code throws the following error:
ValueError: cannot reshape array of size 194400000 into shape (135,1800)

The root cause seems to be the redact_and_return_bbox() method not handling multi-frame DICOMs correctly, especially when working with pixel arrays that contain more than one frame.

To Reproduce

To reproduce the issue, simply use DicomImageRedactorEngine with a multi-frame DICOM image. The de-identification process fails for images with pixel arrays of shape (N, M, P, Q) where N > 1.

Expected Behavior

The DicomImageRedactorEngine should handle multi-frame DICOM images by processing each frame independently or appropriately handling the reshaping of the pixel array.

Question

Is there a solution to handle multi-frame DICOMs in the current implementation? Alternatively, I would be happy to contribute to resolving this issue.

Thank you!

[omri374]

Thanks for raising this @erencanuysal. Any contribution would be appreciated, and we'd be happy to help with reviewing and guidance.