It seems that some of the SOPInstanceUIDs in train_localizers.csv are multi-frame DICOMs (i.e., pixel_array attribute is a 3D array), however only the (x, y) coordinates are provided so we cannot determine the slice location of the aneurysm center.
\nI have attached a file with these instances.
", "messages": [ { "id": 3255566, "postDate": "2025-07-29T01:03:13.747Z", "content": "It seems that some of the SOPInstanceUIDs in train_localizers.csv are multi-frame DICOMs (i.e., pixel_array attribute is a 3D array), however only the (x, y) coordinates are provided so we cannot determine the slice location of the aneurysm center.
\nI have attached a file with these instances.
", "rawMarkdown": "It seems that some of the SOPInstanceUIDs in train_localizers.csv are multi-frame DICOMs (i.e., pixel_array attribute is a 3D array), however only the (x, y) coordinates are provided so we cannot determine the slice location of the aneurysm center.\n\nI have attached a file with these instances.", "votes": 40 }, { "id": 3255836, "postDate": "2025-07-29T14:09:24.670Z", "content": "Thanks Ian for pointing this out. Would your preference be to provide the frame number for these cases or to try to convert them to single slice DICOMs? Not sure which will be more feasible but would be great to have your perspective.
", "rawMarkdown": "Thanks Ian for pointing this out. Would your preference be to provide the frame number for these cases or to try to convert them to single slice DICOMs? Not sure which will be more feasible but would be great to have your perspective.", "votes": 2, "replies": [ { "id": 3255852, "postDate": "2025-07-29T14:42:51.567Z", "content": "I think providing the frame number is fine and more applicable to real-world clinical imaging where some DICOMs may be multi-frame. It's pretty straightforward to check and account for multi-frame DICOMs.
", "rawMarkdown": "I think providing the frame number is fine and more applicable to real-world clinical imaging where some DICOMs may be multi-frame. It's pretty straightforward to check and account for multi-frame DICOMs. ", "votes": 4 } ] }, { "id": 3255806, "postDate": "2025-07-29T12:57:19.190Z", "content": "Thanks @vaillant. We are looking into it and will update the dataset as necessary.
", "rawMarkdown": "Thanks @vaillant. We are looking into it and will update the dataset as necessary.", "votes": 2, "replies": [ { "id": 3258467, "postDate": "2025-07-30T18:01:36.073Z", "content": "Thanks @ryanholbrook and @evancalabrese for your comments.
\nWould you be able to post once the dataset update is complete? Cheers 😀
", "rawMarkdown": "Thanks @ryanholbrook and @evancalabrese for your comments.\n\nWould you be able to post once the dataset update is complete? Cheers 😀", "votes": 1, "replies": [ { "id": 3258517, "postDate": "2025-07-30T19:40:27.133Z", "content": "Yes, we will make a post about any updates once they are available.
", "rawMarkdown": "Yes, we will make a post about any updates once they are available.", "votes": 1 } ] } ] }, { "id": 3258045, "postDate": "2025-07-30T01:31:24.857Z", "content": "You can use instance uid to get the specific dicom file, then you are able to find out the image position.
", "rawMarkdown": "You can use instance uid to get the specific dicom file, then you are able to find out the image position.", "replies": [ { "id": 3262544, "postDate": "2025-08-03T20:25:12.060Z", "content": "Hello!
\nThere are many slices there. You cant choose any point Z in 3d array.
https://www.kaggle.com/code/antonsibilev/heck-localizer-targets-v1
\nHere is explanation about.
\nTrying to find way to get Z. But all my ways so tricky but still trying.
Thanks Ian for pointing this out. Would your preference be to provide the frame number for these cases or to try to convert them to single slice DICOMs? Not sure which will be more feasible but would be great to have your perspective.
", "votes": 2, "replies": [ { "id": 3255852, "author_name": "Ian Pan", "author_url": "", "post_date": "2025-07-29T14:42:51.567000", "content": "I think providing the frame number is fine and more applicable to real-world clinical imaging where some DICOMs may be multi-frame. It's pretty straightforward to check and account for multi-frame DICOMs.
", "votes": 4, "replies": [] } ] }, { "id": 3255806, "author_name": "Ryan Holbrook", "author_url": "", "post_date": "2025-07-29T12:57:19.190000", "content": "Thanks @vaillant. We are looking into it and will update the dataset as necessary.
", "votes": 2, "replies": [ { "id": 3258467, "author_name": "Bartley", "author_url": "", "post_date": "2025-07-30T18:01:36.073000", "content": "Thanks @ryanholbrook and @evancalabrese for your comments.
\nWould you be able to post once the dataset update is complete? Cheers 😀
", "votes": 1, "replies": [ { "id": 3258517, "author_name": "Ryan Holbrook", "author_url": "", "post_date": "2025-07-30T19:40:27.133000", "content": "Yes, we will make a post about any updates once they are available.
", "votes": 1, "replies": [] } ] } ] }, { "id": 3258045, "author_name": "ZhongYi", "author_url": "", "post_date": "2025-07-30T01:31:24.857000", "content": "You can use instance uid to get the specific dicom file, then you are able to find out the image position.
", "votes": 0, "replies": [ { "id": 3262544, "author_name": "Anton Sibilev", "author_url": "", "post_date": "2025-08-03T20:25:12.060000", "content": "Hello!
\nThere are many slices there. You cant choose any point Z in 3d array.
https://www.kaggle.com/code/antonsibilev/heck-localizer-targets-v1
\nHere is explanation about.
\nTrying to find way to get Z. But all my ways so tricky but still trying.