I am making a separate topic on this for visibility.
\nThere has already been discussing regarding PerFrameFunctionalGroupsSequence, which the test set has been updated to include. This is for extracting ImagePositionPatient information.
Multiframe DICOMs also contain SharedFunctionalGroupsSequence which contains ImageOrientationPatient and PixelSpacing information. However, this is not present in the test set based on my submissions. All multiframe DICOMs in the training set contain this attribute. It would be helpful if this attribute was also available in the test set.
I am making a separate topic on this for visibility.
\nThere has already been discussing regarding PerFrameFunctionalGroupsSequence, which the test set has been updated to include. This is for extracting ImagePositionPatient information.
Multiframe DICOMs also contain SharedFunctionalGroupsSequence which contains ImageOrientationPatient and PixelSpacing information. However, this is not present in the test set based on my submissions. All multiframe DICOMs in the training set contain this attribute. It would be helpful if this attribute was also available in the test set.
The metadata for multiframe files might still be in there but then organized in the same way for singleframe files, meaning it might just be under their 'normal' names. We reasoned that they took all the metadata out of each file and then only put back the allowed tags without keeping the original structure of dicom metadata. We dont know for sure, but would seem logical to us. Hope this helps!
", "rawMarkdown": "The metadata for multiframe files might still be in there but then organized in the same way for singleframe files, meaning it might just be under their 'normal' names. We reasoned that they took all the metadata out of each file and then only put back the allowed tags without keeping the original structure of dicom metadata. We dont know for sure, but would seem logical to us. Hope this helps!", "votes": 1 }, { "id": 3296096, "postDate": "2025-09-30T06:59:41.180Z", "content": "To be honest, I've already trained an orientation classifier.
\nSeemed easier to me than looking for potentially missing dicom fields :)
If you are willing to share, that would be extremely helpful!
", "rawMarkdown": "If you are willing to share, that would be extremely helpful!" } ] }, { "id": 3296796, "postDate": "2025-10-01T16:16:44.463Z", "content": "Unfortunately its too late for any additional changes. I will make a separate post about this.
", "rawMarkdown": "Unfortunately its too late for any additional changes. I will make a separate post about this." } ], "comments": [ { "id": 3297154, "author_name": "Willem Dieleman", "author_url": "", "post_date": "2025-10-02T14:03:50.090000", "content": "The metadata for multiframe files might still be in there but then organized in the same way for singleframe files, meaning it might just be under their 'normal' names. We reasoned that they took all the metadata out of each file and then only put back the allowed tags without keeping the original structure of dicom metadata. We dont know for sure, but would seem logical to us. Hope this helps!
", "votes": 1, "replies": [] }, { "id": 3296096, "author_name": "Theo Viel", "author_url": "", "post_date": "2025-09-30T06:59:41.180000", "content": "To be honest, I've already trained an orientation classifier.
\nSeemed easier to me than looking for potentially missing dicom fields :)
If you are willing to share, that would be extremely helpful!
", "votes": 0, "replies": [] } ] }, { "id": 3296796, "author_name": "Evan Calabrese", "author_url": "", "post_date": "2025-10-01T16:16:44.463000", "content": "Unfortunately its too late for any additional changes. I will make a separate post about this.
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "3295924": "I am making a separate topic on this for visibility.\n\nThere has already been discussing regarding `PerFrameFunctionalGroupsSequence`, which the test set has been updated to include. This is for extracting `ImagePositionPatient` information. \n\nMultiframe DICOMs also contain `SharedFunctionalGroupsSequence` which contains `ImageOrientationPatient` and `PixelSpacing` information. However, this is not present in the test set based on my submissions. All multiframe DICOMs in the training set contain this attribute. It would be helpful if this attribute was also available in the test set.\n\n", "3297154": "The metadata for multiframe files might still be in there but then organized in the same way for singleframe files, meaning it might just be under their 'normal' names. We reasoned that they took all the metadata out of each file and then only put back the allowed tags without keeping the original structure of dicom metadata. We dont know for sure, but would seem logical to us. Hope this helps!", "3296096": "To be honest, I've already trained an orientation classifier.\nSeemed easier to me than looking for potentially missing dicom fields :) ", "3296796": "Unfortunately its too late for any additional changes. I will make a separate post about this." } }