![\"result_table\"](\"https://f.easyuploader.app/20210821051434_56624470.jpg\")
Found this interesting and insightful paper on the advantages/disadvantages for different augmentation techniques on MRI data for brain-tumor segmentation.
\nI want to put it here in case others may find it useful, and thank you to the authors of the paper for their work.
\n", "messages": [ { "id": "1483726", "postDate": "08/20/2021 20:04:30", "content": "Found this interesting and insightful paper on the advantages/disadvantages for different augmentation techniques on MRI data for brain-tumor segmentation.
\nI want to put it here in case others may find it useful, and thank you to the authors of the paper for their work.
\n", "rawMarkdown": "Found this interesting and insightful paper on the advantages/disadvantages for different augmentation techniques on MRI data for brain-tumor segmentation.\n\nI want to put it here in case others may find it useful, and thank you to the authors of the paper for their work.\n\n[Click here for the paper](https://www.frontiersin.org/articles/10.3389/fncom.2019.00083/full)", "votes": null }, { "id": "1483734", "postDate": "08/20/2021 20:16:27", "content": "Thank @dinodeep for sharing this useful paper.
\nAs all participants probably feel, this competition has very little training, public, and private data, and a large shake is expected to occur. Therefore, I think there is no doubt that data augmentation is one of the most important keys to build a robust model. The model can be roughly divided into 2D or 3D approaches. I would also like to share the following paper that examines what is effective augmentation in 3D CNNs using the BraTS 2020 dataset.
\n\n\nWhat is the best data augmentation for 3D brain tumor segmentation?
\n
\n 26 Oct 2020
\n Marco Domenico Cirillo, David Abramian, Anders Eklund
\n https://arxiv.org/pdf/2010.13372v2.pdf
This paper is about segmentation for brain tumors, but 3D CNN representations such as used in segmentation should also be effective for MGMT classification in this competition, so I think it is worth trying the augmentation that is effective for 3D segmentation in our classification task.
\nThe standard 3D U-Net was the winner at BraTS 2020, and this paper uses it to examine several augmentations and concludes that brightness, elastic deformation, scaling, flipping, and 90-deg rotation are effective in this order.
\nOne point to note is that this paper has very little preprocessing. I think it is quite possible that careful preprocessing (e.g. intensity normalization, axis alignment, …) could change the order of magnitude of the effects of these augmentations.
GitHub 👇
\nhttps://github.com/mdciri/3D-augmentation-techniques
Thank you for sharing this as well. Upvoting. ⬆️
", "rawMarkdown": "Thank you for sharing this as well. Upvoting. ⬆️", "votes": null }, { "id": "1483844", "postDate": "08/20/2021 23:19:19", "content": "Thank you @maxwell110 for sharing this! Your informative response gives me a better idea at how to post about my findings in a discussion post. Thanks again!
", "rawMarkdown": "Thank you @maxwell110 for sharing this! Your informative response gives me a better idea at how to post about my findings in a discussion post. Thanks again!", "votes": null } ], "comments": [ { "id": 1483734, "author_name": "maxwell110", "author_url": "", "post_date": "08/20/2021 20:16:27", "content": "Thank @dinodeep for sharing this useful paper.
\nAs all participants probably feel, this competition has very little training, public, and private data, and a large shake is expected to occur. Therefore, I think there is no doubt that data augmentation is one of the most important keys to build a robust model. The model can be roughly divided into 2D or 3D approaches. I would also like to share the following paper that examines what is effective augmentation in 3D CNNs using the BraTS 2020 dataset.
\n\n\nWhat is the best data augmentation for 3D brain tumor segmentation?
\n
\n 26 Oct 2020
\n Marco Domenico Cirillo, David Abramian, Anders Eklund
\n https://arxiv.org/pdf/2010.13372v2.pdf
This paper is about segmentation for brain tumors, but 3D CNN representations such as used in segmentation should also be effective for MGMT classification in this competition, so I think it is worth trying the augmentation that is effective for 3D segmentation in our classification task.
\nThe standard 3D U-Net was the winner at BraTS 2020, and this paper uses it to examine several augmentations and concludes that brightness, elastic deformation, scaling, flipping, and 90-deg rotation are effective in this order.
\nOne point to note is that this paper has very little preprocessing. I think it is quite possible that careful preprocessing (e.g. intensity normalization, axis alignment, …) could change the order of magnitude of the effects of these augmentations.
GitHub 👇
\nhttps://github.com/mdciri/3D-augmentation-techniques
Thank you for sharing this as well. Upvoting. ⬆️
", "votes": null, "replies": [] }, { "id": 1483844, "author_name": "dinodeep", "author_url": "", "post_date": "08/20/2021 23:19:19", "content": "Thank you @maxwell110 for sharing this! Your informative response gives me a better idea at how to post about my findings in a discussion post. Thanks again!
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1483726": "Found this interesting and insightful paper on the advantages/disadvantages for different augmentation techniques on MRI data for brain-tumor segmentation.\n\nI want to put it here in case others may find it useful, and thank you to the authors of the paper for their work.\n\n[Click here for the paper](https://www.frontiersin.org/articles/10.3389/fncom.2019.00083/full)", "1483734": "Thank @dinodeep for sharing this useful paper.\n\nAs all participants probably feel, this competition has very little training, public, and private data, and a large shake is expected to occur. Therefore, I think there is no doubt that data augmentation is one of the most important keys to build a robust model. The model can be roughly divided into 2D or 3D approaches. I would also like to share the following paper that examines what is effective augmentation in 3D CNNs using the BraTS 2020 dataset.\n\n\n> What is the best data augmentation for 3D brain tumor segmentation?\n26 Oct 2020\nMarco Domenico Cirillo, David Abramian, Anders Eklund\nhttps://arxiv.org/pdf/2010.13372v2.pdf\n\n\nThis paper is about segmentation for brain tumors, but 3D CNN representations such as used in segmentation should also be effective for MGMT classification in this competition, so I think it is worth trying the augmentation that is effective for 3D segmentation in our classification task.\n\nThe standard 3D U-Net was the winner at BraTS 2020, and this paper uses it to examine several augmentations and concludes that `brightness, elastic deformation, scaling, flipping, and 90-deg rotation are effective in this order`.\nOne point to note is that this paper has very little preprocessing. I think it is quite possible that careful preprocessing (e.g. intensity normalization, axis alignment, ...) could change the order of magnitude of the effects of these augmentations.\n\nGitHub 👇\nhttps://github.com/mdciri/3D-augmentation-techniques\n\n", "1483767": "Thank you for sharing this as well. Upvoting. ⬆️", "1483844": "Thank you @maxwell110 for sharing this! Your informative response gives me a better idea at how to post about my findings in a discussion post. Thanks again!" }, "source": "meta" }