I just want to raise this problem.
\nTo read DICOM files (*.dcm) you would need to use pydicom. However, while reading the pixel_array, for whatever reason, using it would SOMETIMES lead to an error of:
\nRuntimeError: The following handlers are available to decode the pixel data however they are missing required dependencies: GDCM (req. GDCM)\nAnd to fix this error, you would need to download/update the GDCM module via conda:
\n!conda install -c conda-forge gdcm -y\nIn other words, you would be needing internet connection for your kernel. Can there be any loophole for this? Any fix?
\nI thought of saving my model, convert all test DICOM into png, loading them all to an offline kernel and submitting the result. However, this will not fix the problem for the private data that is still in DICOM format.
\nI hope this problem gains attention.
\nEDIT:
\nI decided to use this as reference and implemented vtk instead of pydicom.
also discussed in the currently running Pulmonary Fibrosis competition. You may find helpful comments there.
\nFor now, I am just trapping the error and using a default all-zeros image. Hoping there are not too many of these in the private test dataset.
", "rawMarkdown": "also discussed in the currently running Pulmonary Fibrosis competition. You may find helpful comments there.\n\nFor now, I am just trapping the error and using a default all-zeros image. Hoping there are not too many of these in the private test dataset.", "votes": 1, "replies": [ { "id": 1006679, "postDate": "2020-09-11T13:33:46.907Z", "content": "Thanks! I saw this thread that seems to have some fix. Will try it now and see.
", "rawMarkdown": "Thanks! I saw [this thread](https://www.kaggle.com/c/osic-pulmonary-fibrosis-progression/discussion/166957) that seems to have some fix. Will try it now and see." } ] }, { "id": 1006634, "postDate": "2020-09-11T12:39:10.003Z", "content": "I just want to raise this problem.
\nTo read DICOM files (*.dcm) you would need to use pydicom. However, while reading the pixel_array, for whatever reason, using it would SOMETIMES lead to an error of:
\nRuntimeError: The following handlers are available to decode the pixel data however they are missing required dependencies: GDCM (req. GDCM)\nAnd to fix this error, you would need to download/update the GDCM module via conda:
\n!conda install -c conda-forge gdcm -y\nIn other words, you would be needing internet connection for your kernel. Can there be any loophole for this? Any fix?
\nI thought of saving my model, convert all test DICOM into png, loading them all to an offline kernel and submitting the result. However, this will not fix the problem for the private data that is still in DICOM format.
\nI hope this problem gains attention.
\nEDIT:
\nI decided to use this as reference and implemented vtk instead of pydicom.
Thank you @seraphwedd18 ! I was encountering the same problem! One solution is to manually download the gdcm library files into your input folder and do a conda offline install each time.
", "rawMarkdown": "Thank you @seraphwedd18 ! I was encountering the same problem! One solution is to manually download the gdcm library files into your input folder and do a conda offline install each time." } ], "comments": [ { "id": 1006646, "author_name": "quadcore/Richard Epstein", "author_url": "", "post_date": "2020-09-11T13:00:33.150000", "content": "also discussed in the currently running Pulmonary Fibrosis competition. You may find helpful comments there.
\nFor now, I am just trapping the error and using a default all-zeros image. Hoping there are not too many of these in the private test dataset.
", "votes": 1, "replies": [ { "id": 1006679, "author_name": "Venturillo JE", "author_url": "", "post_date": "2020-09-11T13:33:46.907000", "content": "Thanks! I saw this thread that seems to have some fix. Will try it now and see.
", "votes": 0, "replies": [] } ] }, { "id": 1006644, "author_name": "Alaa Triki", "author_url": "", "post_date": "2020-09-11T12:58:20.237000", "content": "Thank you @seraphwedd18 ! I was encountering the same problem! One solution is to manually download the gdcm library files into your input folder and do a conda offline install each time.
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1006646": "also discussed in the currently running Pulmonary Fibrosis competition. You may find helpful comments there.\n\nFor now, I am just trapping the error and using a default all-zeros image. Hoping there are not too many of these in the private test dataset.", "1006634": "I just want to raise this problem.\n\nTo read DICOM files (*.dcm) you would need to use pydicom. However, while reading the pixel_array, for whatever reason, using it would SOMETIMES lead to an error of: \n\n```\nRuntimeError: The following handlers are available to decode the pixel data however they are missing required dependencies: GDCM (req. GDCM)\n```\n\nAnd to fix this error, you would need to download/update the GDCM module via conda:\n\n```\n!conda install -c conda-forge gdcm -y\n```\n\nIn other words, you would be needing internet connection for your kernel. Can there be any loophole for this? Any fix?\n\nI thought of saving my model, convert all test DICOM into png, loading them all to an offline kernel and submitting the result. However, this will not fix the problem for the private data that is still in DICOM format.\n\nI hope this problem gains attention.\n\n\nEDIT:\nI decided to use [this](https://www.kaggle.com/eladwar/20-seconds-or-less) as reference and implemented vtk instead of pydicom.", "1006644": "Thank you @seraphwedd18 ! I was encountering the same problem! One solution is to manually download the gdcm library files into your input folder and do a conda offline install each time." } }