I think segmentation would be quite useful for this competition. In previous RSNA competitions, the organisers had provided ground truth for segmentation. As they did not provide any for this competition, unsupervised segmentation could be an idea.
\nI find this repository quite helpful. Specially SSL_ALPNet implementation, as it is used for abdominal organ segmentation for CT and MRI.
\nI think it would be helpful for others also. If you have any other ideas for segmentation, please pitch in.
", "messages": [ { "id": 2855990, "postDate": "2024-06-05T05:30:07.687Z", "content": "I think segmentation would be quite useful for this competition. In previous RSNA competitions, the organisers had provided ground truth for segmentation. As they did not provide any for this competition, unsupervised segmentation could be an idea.
\nI find this repository quite helpful. Specially SSL_ALPNet implementation, as it is used for abdominal organ segmentation for CT and MRI.
\nI think it would be helpful for others also. If you have any other ideas for segmentation, please pitch in.
", "rawMarkdown": "I think segmentation would be quite useful for this competition. In previous RSNA competitions, the organisers had provided ground truth for segmentation. As they did not provide any for this competition, unsupervised segmentation could be an idea.\n\nI find [this repository](https://github.com/Mirsadeghi/Awesome-Unsupervised-Segmentation?tab=readme-ov-file) quite helpful. Specially [SSL_ALPNet implementation](https://github.com/cheng-01037/Self-supervised-Fewshot-Medical-Image-Segmentation), as it is used for abdominal organ segmentation for CT and MRI.\n\nI think it would be helpful for others also. If you have any other ideas for segmentation, please pitch in. ", "votes": 8 }, { "id": 2911813, "postDate": "2024-07-08T14:54:08.390Z", "content": "The label coordinate file is kind of like the ground truth for segmentation -- it's not exactly segmentation, but you can predict the important location for model, or draw a fix sized bounding box around it. Either way would be helpful and more reliable than unsupervised training.
\nAlso, there are supervised segmentation datasets available in other competitions. I remember it's in the discussion, it called SPIDER.
", "rawMarkdown": "The label coordinate file is kind of like the ground truth for segmentation -- it's not exactly segmentation, but you can predict the important location for model, or draw a fix sized bounding box around it. Either way would be helpful and more reliable than unsupervised training. \n\nAlso, there are supervised segmentation datasets available in other competitions. I remember it's in the discussion, it called SPIDER.", "votes": 4, "replies": [ { "id": 2913312, "postDate": "2024-07-09T12:01:48.373Z", "content": "Thank you for your suggestion
", "rawMarkdown": "Thank you for your suggestion" }, { "id": 2914841, "postDate": "2024-07-10T07:38:41.920Z", "content": "I have been working on Segmentation with SPIDER and Zenodo dataset. Here is the link of my notebook (https://www.kaggle.com/code/tabassumnova/segmentation-unet-inference-zenodo-spider)
", "rawMarkdown": "I have been working on Segmentation with SPIDER and Zenodo dataset. Here is the link of my notebook ([https://www.kaggle.com/code/tabassumnova/segmentation-unet-inference-zenodo-spider](url))", "votes": 1 } ] }, { "id": 2911518, "postDate": "2024-07-08T11:07:57.873Z", "content": "@tabassumnova Have you tried to implement this? What were the results? Your response will be highly valuable.
", "rawMarkdown": "@tabassumnova Have you tried to implement this? What were the results? Your response will be highly valuable.\n", "replies": [ { "id": 2913311, "postDate": "2024-07-09T12:01:06.150Z", "content": "No, I have not tried it yet
", "rawMarkdown": "No, I have not tried it yet" } ] }, { "id": 2858885, "postDate": "2024-06-06T17:41:07.080Z", "content": "Problem is we don't have ground truth segmentation labels.
", "rawMarkdown": "Problem is we don't have ground truth segmentation labels.", "replies": [ { "id": 2859689, "postDate": "2024-06-07T07:21:34.240Z", "content": "According to their paper, the network can perform self supervised segmentation which does not need ground truth
", "rawMarkdown": "According to their paper, the network can perform self supervised segmentation which does not need ground truth", "votes": 3 } ] }, { "id": 2877830, "postDate": "2024-06-18T15:28:01.827Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2865664, "postDate": "2024-06-10T22:28:42.247Z", "rawMarkdown": "", "isDeleted": true, "replies": [ { "id": 2867188, "postDate": "2024-06-11T17:20:30.853Z", "content": "I want to perform segmentation because if we have segments of the spine then we can look for a specific pathology/condition in that specific part. We don’t have to search in the whole image.
", "rawMarkdown": "I want to perform segmentation because if we have segments of the spine then we can look for a specific pathology/condition in that specific part. We don’t have to search in the whole image. ", "votes": 1 } ] } ], "comments": [ { "id": 2911813, "author_name": "LLLEEEOOOH", "author_url": "", "post_date": "2024-07-08T14:54:08.390000", "content": "The label coordinate file is kind of like the ground truth for segmentation -- it's not exactly segmentation, but you can predict the important location for model, or draw a fix sized bounding box around it. Either way would be helpful and more reliable than unsupervised training.
\nAlso, there are supervised segmentation datasets available in other competitions. I remember it's in the discussion, it called SPIDER.
", "votes": 4, "replies": [ { "id": 2913312, "author_name": "Tabassum_Nova", "author_url": "", "post_date": "2024-07-09T12:01:48.373000", "content": "Thank you for your suggestion
", "votes": 0, "replies": [] }, { "id": 2914841, "author_name": "Tabassum_Nova", "author_url": "", "post_date": "2024-07-10T07:38:41.920000", "content": "I have been working on Segmentation with SPIDER and Zenodo dataset. Here is the link of my notebook (https://www.kaggle.com/code/tabassumnova/segmentation-unet-inference-zenodo-spider)
", "votes": 1, "replies": [] } ] }, { "id": 2911518, "author_name": "Devsya ", "author_url": "", "post_date": "2024-07-08T11:07:57.873000", "content": "@tabassumnova Have you tried to implement this? What were the results? Your response will be highly valuable.
", "votes": 0, "replies": [ { "id": 2913311, "author_name": "Tabassum_Nova", "author_url": "", "post_date": "2024-07-09T12:01:06.150000", "content": "No, I have not tried it yet
", "votes": 0, "replies": [] } ] }, { "id": 2858885, "author_name": "Masavarapu Appala Naidu", "author_url": "", "post_date": "2024-06-06T17:41:07.080000", "content": "Problem is we don't have ground truth segmentation labels.
", "votes": 0, "replies": [ { "id": 2859689, "author_name": "Tabassum_Nova", "author_url": "", "post_date": "2024-06-07T07:21:34.240000", "content": "According to their paper, the network can perform self supervised segmentation which does not need ground truth
", "votes": 3, "replies": [] } ] }, { "id": 2877830, "author_name": "", "author_url": "", "post_date": "2024-06-18T15:28:01.827000", "content": "", "votes": 0, "replies": [] }, { "id": 2865664, "author_name": "", "author_url": "", "post_date": "2024-06-10T22:28:42.247000", "content": "", "votes": 0, "replies": [ { "id": 2867188, "author_name": "Tabassum_Nova", "author_url": "", "post_date": "2024-06-11T17:20:30.853000", "content": "I want to perform segmentation because if we have segments of the spine then we can look for a specific pathology/condition in that specific part. We don’t have to search in the whole image.
", "votes": 1, "replies": [] } ] } ], "raw_markdown_by_id": { "2855990": "I think segmentation would be quite useful for this competition. In previous RSNA competitions, the organisers had provided ground truth for segmentation. As they did not provide any for this competition, unsupervised segmentation could be an idea.\n\nI find [this repository](https://github.com/Mirsadeghi/Awesome-Unsupervised-Segmentation?tab=readme-ov-file) quite helpful. Specially [SSL_ALPNet implementation](https://github.com/cheng-01037/Self-supervised-Fewshot-Medical-Image-Segmentation), as it is used for abdominal organ segmentation for CT and MRI.\n\nI think it would be helpful for others also. If you have any other ideas for segmentation, please pitch in. ", "2911813": "The label coordinate file is kind of like the ground truth for segmentation -- it's not exactly segmentation, but you can predict the important location for model, or draw a fix sized bounding box around it. Either way would be helpful and more reliable than unsupervised training. \n\nAlso, there are supervised segmentation datasets available in other competitions. I remember it's in the discussion, it called SPIDER.", "2911518": "@tabassumnova Have you tried to implement this? What were the results? Your response will be highly valuable.\n", "2858885": "Problem is we don't have ground truth segmentation labels.", "2877830": "", "2865664": "" } }