For my pipeline I've been segmenting and croping ROIs from competition dicoms. The resulting dataset is ~30 GB.
\n\n", "messages": [ { "id": 3273989, "postDate": "2025-08-23T18:23:01.687Z", "content": "For my pipeline I've been segmenting and croping ROIs from competition dicoms. The resulting dataset is ~30 GB.
\n\n", "rawMarkdown": "For my pipeline I've been segmenting and croping ROIs from competition dicoms. The resulting dataset is ~30 GB.\n\n[Notebook](https://www.kaggle.com/code/sacuscreed/rsna-extracted-rois-demo?scriptVersionId=257734711)\n\n[Dataset](https://www.kaggle.com/datasets/sacuscreed/rsna-extracted-rois)", "votes": 13 }, { "id": 3274231, "postDate": "2025-08-24T11:29:48.890Z", "content": "DISCLAIMER: I've been stacking volumes by instance number. But examining the results some of them look like reversed in depth so after investigating and assuming in each series all slices are almost equall oriented (same IOP) the actual sorting should be by (IPP). So, if I'm right, to train this data one should to first check wich of them have IPP ordering reversed with respect instance number and flip his depth. I came back to the basis of my pipeline so I can't check it right now by myself.
", "rawMarkdown": "DISCLAIMER: I've been stacking volumes by instance number. But examining the results some of them look like reversed in depth so after investigating and assuming in each series all slices are almost equall oriented (same IOP) the actual sorting should be by (IPP). So, if I'm right, to train this data one should to first check wich of them have IPP ordering reversed with respect instance number and flip his depth. I came back to the basis of my pipeline so I can't check it right now by myself.", "votes": 1, "replies": [ { "id": 3274236, "postDate": "2025-08-24T11:56:35.063Z", "content": "@sacuscreed Thank you for providing this. Really nice work.
", "rawMarkdown": "@sacuscreed Thank you for providing this. Really nice work.", "votes": 1 } ] }, { "id": 3281146, "postDate": "2025-09-03T18:10:45.043Z", "content": "Thx, really helpful
", "rawMarkdown": "Thx, really helpful " }, { "id": 3274993, "postDate": "2025-08-25T18:24:37.580Z", "content": "Given the different voxel spacings and shape sizes, how did you derive the ROIs, if I may ask?
", "rawMarkdown": "Given the different voxel spacings and shape sizes, how did you derive the ROIs, if I may ask?", "replies": [ { "id": 3275007, "postDate": "2025-08-25T18:59:19.937Z", "content": "In one word, resizing. I have a 3D example on codes.
", "rawMarkdown": "In one word, resizing. I have a 3D example on codes.", "replies": [ { "id": 3275357, "postDate": "2025-08-26T11:30:02.053Z", "content": "Thanks. I did extract all metadata and my reasoning would be to calculate each dimension in mm (which is what the voxel spacing is expressed in, unless I'm mistaken?) and find a common ground in resizing, so that we can extract ROI pieces which cover the same area/volume (depending on view, 2D/3D) and are brought to the same dimensions in pixels?
\nWould you mind pointing me to your 3D example? I'd be very curious to take a look.
", "rawMarkdown": "Thanks. I did extract all metadata and my reasoning would be to calculate each dimension in mm (which is what the voxel spacing is expressed in, unless I'm mistaken?) and find a common ground in resizing, so that we can extract ROI pieces which cover the same area/volume (depending on view, 2D/3D) and are brought to the same dimensions in pixels? \n\nWould you mind pointing me to your 3D example? I'd be very curious to take a look." }, { "id": 3275386, "postDate": "2025-08-26T12:20:33.007Z", "content": "https://www.kaggle.com/code/sacuscreed/rsna-from-binary-2d-to-full-3d-segmentation
", "rawMarkdown": "https://www.kaggle.com/code/sacuscreed/rsna-from-binary-2d-to-full-3d-segmentation", "votes": 1 }, { "id": 3275464, "postDate": "2025-08-26T15:11:14.713Z", "content": "@sacuscreed this is smart, thank you for sharing!
", "rawMarkdown": "@sacuscreed this is smart, thank you for sharing!" } ] } ] }, { "id": 3274900, "postDate": "2025-08-25T15:48:01.063Z", "content": "@sacuscreed can training with reduced data with my models and has results approach the original data , thank you in advance
", "rawMarkdown": "@sacuscreed can training with reduced data with my models and has results approach the original data , thank you in advance" } ], "comments": [ { "id": 3274231, "author_name": "Ángel Jacinto Sánchez Ruiz", "author_url": "", "post_date": "2025-08-24T11:29:48.890000", "content": "DISCLAIMER: I've been stacking volumes by instance number. But examining the results some of them look like reversed in depth so after investigating and assuming in each series all slices are almost equall oriented (same IOP) the actual sorting should be by (IPP). So, if I'm right, to train this data one should to first check wich of them have IPP ordering reversed with respect instance number and flip his depth. I came back to the basis of my pipeline so I can't check it right now by myself.
", "votes": 1, "replies": [ { "id": 3274236, "author_name": "Tom", "author_url": "", "post_date": "2025-08-24T11:56:35.063000", "content": "@sacuscreed Thank you for providing this. Really nice work.
", "votes": 1, "replies": [] } ] }, { "id": 3281146, "author_name": "Taha_Alshatiri", "author_url": "", "post_date": "2025-09-03T18:10:45.043000", "content": "Thx, really helpful
", "votes": 0, "replies": [] }, { "id": 3274993, "author_name": "Andrei Zamfir", "author_url": "", "post_date": "2025-08-25T18:24:37.580000", "content": "Given the different voxel spacings and shape sizes, how did you derive the ROIs, if I may ask?
", "votes": 0, "replies": [ { "id": 3275007, "author_name": "Ángel Jacinto Sánchez Ruiz", "author_url": "", "post_date": "2025-08-25T18:59:19.937000", "content": "In one word, resizing. I have a 3D example on codes.
", "votes": 0, "replies": [ { "id": 3275357, "author_name": "Andrei Zamfir", "author_url": "", "post_date": "2025-08-26T11:30:02.053000", "content": "Thanks. I did extract all metadata and my reasoning would be to calculate each dimension in mm (which is what the voxel spacing is expressed in, unless I'm mistaken?) and find a common ground in resizing, so that we can extract ROI pieces which cover the same area/volume (depending on view, 2D/3D) and are brought to the same dimensions in pixels?
\nWould you mind pointing me to your 3D example? I'd be very curious to take a look.
", "votes": 0, "replies": [] }, { "id": 3275386, "author_name": "Ángel Jacinto Sánchez Ruiz", "author_url": "", "post_date": "2025-08-26T12:20:33.007000", "content": "https://www.kaggle.com/code/sacuscreed/rsna-from-binary-2d-to-full-3d-segmentation
", "votes": 1, "replies": [] }, { "id": 3275464, "author_name": "Tom", "author_url": "", "post_date": "2025-08-26T15:11:14.713000", "content": "@sacuscreed this is smart, thank you for sharing!
", "votes": 0, "replies": [] } ] } ] }, { "id": 3274900, "author_name": "work work", "author_url": "", "post_date": "2025-08-25T15:48:01.063000", "content": "@sacuscreed can training with reduced data with my models and has results approach the original data , thank you in advance
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "3273989": "For my pipeline I've been segmenting and croping ROIs from competition dicoms. The resulting dataset is ~30 GB.\n\n[Notebook](https://www.kaggle.com/code/sacuscreed/rsna-extracted-rois-demo?scriptVersionId=257734711)\n\n[Dataset](https://www.kaggle.com/datasets/sacuscreed/rsna-extracted-rois)", "3274231": "DISCLAIMER: I've been stacking volumes by instance number. But examining the results some of them look like reversed in depth so after investigating and assuming in each series all slices are almost equall oriented (same IOP) the actual sorting should be by (IPP). So, if I'm right, to train this data one should to first check wich of them have IPP ordering reversed with respect instance number and flip his depth. I came back to the basis of my pipeline so I can't check it right now by myself.", "3281146": "Thx, really helpful ", "3274993": "Given the different voxel spacings and shape sizes, how did you derive the ROIs, if I may ask?", "3274900": "@sacuscreed can training with reduced data with my models and has results approach the original data , thank you in advance" } }