Hi everyone, to help you get started I simply converted .dcm files to .jpg/.png. As these datasets are quite small comparing the actual dataset, you can easily use them to fast up your training.
256x256: RSNA-CSFD: 256x256 Jpg Dataset (~18GB).512x512: RSNA-CSFD: 512x512 Jpg Dataset (~55GB)Coming soon
", "messages": [ { "id": "1875766", "postDate": "07/29/2022 09:33:43", "content": "Hi everyone, to help you get started I simply converted .dcm files to .jpg/.png. As these datasets are quite small comparing the actual dataset, you can easily use them to fast up your training.
256x256: RSNA-CSFD: 256x256 Jpg Dataset (~18GB).512x512: RSNA-CSFD: 512x512 Jpg Dataset (~55GB)Coming soon
", "rawMarkdown": "Hi everyone, to help you get started I simply converted `.dcm` files to `.jpg`/`.png`. As these datasets are quite small comparing the actual dataset, you can easily use them to fast up your training.\n## Jpg\n* `256x256`: [RSNA-CSFD: 256x256 Jpg Dataset](https://www.kaggle.com/datasets/awsaf49/rsna-csfd-256x256-jpg-dataset) (~18GB).\n* `512x512`: [RSNA-CSFD: 512x512 Jpg Dataset](https://www.kaggle.com/datasets/awsaf49/rsna-csfd-512x512-jpg-Dataset) (~55GB)\n## Png\nComing soon", "votes": null }, { "id": "1875844", "postDate": "07/29/2022 10:45:03", "content": "Thank for sharing!!!!
", "rawMarkdown": "Thank for sharing!!!!", "votes": null }, { "id": "1875931", "postDate": "07/29/2022 12:51:40", "content": "Thanks, this will help me!
", "rawMarkdown": "Thanks, this will help me!", "votes": null }, { "id": "1875983", "postDate": "07/29/2022 13:40:56", "content": "Thanks for sharing ..
", "rawMarkdown": "Thanks for sharing ..", "votes": null }, { "id": "1876108", "postDate": "07/29/2022 15:25:23", "content": "Thanks for sharing, it will be quite useful!
\nDid you use interpolation for resizing from 512 x 512 to 256 x 256?
I also made a kernel for PNG conversion:
\n\n", "rawMarkdown": "I also made a kernel for PNG conversion:\n>https://www.kaggle.com/code/jirkaborovec/spine-fracture-load-convert-dicom-to-png", "votes": null }, { "id": "1877235", "postDate": "07/30/2022 12:47:14", "content": "https://www.kaggle.com/code/jirkaborovec/spine-fracture-load-convert-dicom-to-png
\n
@awsaf49 given your experience from the competitions, do you think converting images from 16-bit Dicom to JPEG or PNG with an 8-bit depth will distort the image's quality and characteristics?
", "rawMarkdown": "awsaf49 given your experience from the competitions, do you think converting images from 16-bit Dicom to JPEG or PNG with an 8-bit depth will distort the image's quality and characteristics?", "votes": null }, { "id": "1880097", "postDate": "08/01/2022 13:15:33", "content": "I used inter_area* interpolation.
", "rawMarkdown": "I used *inter_area** interpolation.", "votes": null }, { "id": "1881091", "postDate": "08/02/2022 08:45:51", "content": "I also created a dataset from the kernel output with some denoising, so check also:
\n\n", "rawMarkdown": "I also created a dataset from the kernel output with some denoising, so check also:\n\n>https://www.kaggle.com/datasets/jirkaborovec/cervical-spine-fracture-detection-equalized-png", "votes": null }, { "id": "1956074", "postDate": "09/26/2022 09:10:33", "content": "https://www.kaggle.com/datasets/jirkaborovec/cervical-spine-fracture-detection-equalized-png
\n
May I ask how did you overcome disk space issue? I tried to create my own dataset but ran into full disk issue, even I save images into /tmp path
", "rawMarkdown": "May I ask how did you overcome disk space issue? I tried to create my own dataset but ran into full disk issue, even I save images into /tmp path", "votes": null }, { "id": "1964131", "postDate": "09/30/2022 15:33:53", "content": "Since jpg is a lossy storage format, it seems possible that small/hairline fractures will be removed from the image. Especially since these small features will show up in the high-frequency spectrum of the images DCT, which could be removed by the jpg conversion. I'd be interested to see the results of training with raw and jpg formats
", "rawMarkdown": "Since jpg is a lossy storage format, it seems possible that small/hairline fractures will be removed from the image. Especially since these small features will show up in the high-frequency spectrum of the images DCT, which could be removed by the jpg conversion. I'd be interested to see the results of training with raw and jpg formats", "votes": null }, { "id": "1988772", "postDate": "10/15/2022 14:33:52", "content": "Anyone compares training CV loss between the original dataset and this Jpg dataset?
", "rawMarkdown": "Anyone compares training CV loss between the original dataset and this Jpg dataset?", "votes": null }, { "id": "3229704", "postDate": "06/22/2025 01:47:43", "content": "info_popup.9.png
", "rawMarkdown": "[](url)info_popup.9.png", "votes": null } ], "comments": [ { "id": 1875844, "author_name": "imvision12", "author_url": "", "post_date": "07/29/2022 10:45:03", "content": "Thank for sharing!!!!
", "votes": null, "replies": [] }, { "id": 1875931, "author_name": "andrewteplov", "author_url": "", "post_date": "07/29/2022 12:51:40", "content": "Thanks, this will help me!
", "votes": null, "replies": [] }, { "id": 1875983, "author_name": "reachkishore", "author_url": "", "post_date": "07/29/2022 13:40:56", "content": "Thanks for sharing ..
", "votes": null, "replies": [] }, { "id": 1876108, "author_name": "augustosarquisserpa", "author_url": "", "post_date": "07/29/2022 15:25:23", "content": "Thanks for sharing, it will be quite useful!
\nDid you use interpolation for resizing from 512 x 512 to 256 x 256?
I used inter_area* interpolation.
", "votes": null, "replies": [] } ] }, { "id": 1877090, "author_name": "jirkaborovec", "author_url": "", "post_date": "07/30/2022 10:42:25", "content": "I also made a kernel for PNG conversion:
\n\n", "votes": null, "replies": [ { "id": 1881091, "author_name": "jirkaborovec", "author_url": "", "post_date": "08/02/2022 08:45:51", "content": "https://www.kaggle.com/code/jirkaborovec/spine-fracture-load-convert-dicom-to-png
\n
I also created a dataset from the kernel output with some denoising, so check also:
\n\n", "votes": null, "replies": [] } ] }, { "id": 1877235, "author_name": "mohammaddehghan", "author_url": "", "post_date": "07/30/2022 12:47:14", "content": "https://www.kaggle.com/datasets/jirkaborovec/cervical-spine-fracture-detection-equalized-png
\n
@awsaf49 given your experience from the competitions, do you think converting images from 16-bit Dicom to JPEG or PNG with an 8-bit depth will distort the image's quality and characteristics?
", "votes": null, "replies": [] }, { "id": 1956074, "author_name": "ptran1203", "author_url": "", "post_date": "09/26/2022 09:10:33", "content": "May I ask how did you overcome disk space issue? I tried to create my own dataset but ran into full disk issue, even I save images into /tmp path
", "votes": null, "replies": [] }, { "id": 1964131, "author_name": "coleharlow", "author_url": "", "post_date": "09/30/2022 15:33:53", "content": "Since jpg is a lossy storage format, it seems possible that small/hairline fractures will be removed from the image. Especially since these small features will show up in the high-frequency spectrum of the images DCT, which could be removed by the jpg conversion. I'd be interested to see the results of training with raw and jpg formats
", "votes": null, "replies": [] }, { "id": 1988772, "author_name": "pierretisseur", "author_url": "", "post_date": "10/15/2022 14:33:52", "content": "Anyone compares training CV loss between the original dataset and this Jpg dataset?
", "votes": null, "replies": [] }, { "id": 3229704, "author_name": "tamamlifesupport", "author_url": "", "post_date": "06/22/2025 01:47:43", "content": "info_popup.9.png
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1875766": "Hi everyone, to help you get started I simply converted `.dcm` files to `.jpg`/`.png`. As these datasets are quite small comparing the actual dataset, you can easily use them to fast up your training.\n## Jpg\n* `256x256`: [RSNA-CSFD: 256x256 Jpg Dataset](https://www.kaggle.com/datasets/awsaf49/rsna-csfd-256x256-jpg-dataset) (~18GB).\n* `512x512`: [RSNA-CSFD: 512x512 Jpg Dataset](https://www.kaggle.com/datasets/awsaf49/rsna-csfd-512x512-jpg-Dataset) (~55GB)\n## Png\nComing soon", "1875844": "Thank for sharing!!!!", "1875931": "Thanks, this will help me!", "1875983": "Thanks for sharing ..", "1876108": "Thanks for sharing, it will be quite useful! \nDid you use interpolation for resizing from 512 x 512 to 256 x 256?", "1877090": "I also made a kernel for PNG conversion:\n>https://www.kaggle.com/code/jirkaborovec/spine-fracture-load-convert-dicom-to-png", "1877235": "awsaf49 given your experience from the competitions, do you think converting images from 16-bit Dicom to JPEG or PNG with an 8-bit depth will distort the image's quality and characteristics?", "1880097": "I used *inter_area** interpolation.", "1881091": "I also created a dataset from the kernel output with some denoising, so check also:\n\n>https://www.kaggle.com/datasets/jirkaborovec/cervical-spine-fracture-detection-equalized-png", "1956074": "May I ask how did you overcome disk space issue? I tried to create my own dataset but ran into full disk issue, even I save images into /tmp path", "1964131": "Since jpg is a lossy storage format, it seems possible that small/hairline fractures will be removed from the image. Especially since these small features will show up in the high-frequency spectrum of the images DCT, which could be removed by the jpg conversion. I'd be interested to see the results of training with raw and jpg formats", "1988772": "Anyone compares training CV loss between the original dataset and this Jpg dataset?", "3229704": "[](url)info_popup.9.png" }, "source": "meta" }