{ "id": 377208, "title": "DicomSDL & VOILUTFunction Sigmoid", "url": "/competitions/rsna-breast-cancer-detection/discussion/377208", "author_name": "Bob de Graaf", "post_date": "2023-01-10T10:42:16.830000", "votes": 14, "comment_count": 5, "views": 0, "content": "

Hi there!

\n

I might be wrong, but I think that because at least 22% of this competition's datasets use Sigmoid as its VOILUTFunction, it's important to use some kind of windowing operation that Pydicom does.
\nBut, since DicomSDL is a lot faster, I've dug into the source code of Pydicom and tried to recreate the windowing operation so I could keep using DicomSDL.

\n

I've posted some results here.

\n

I think that if you use DicomSDL, you should check it out!
\nAnd if I'm wrong, please let me know! :)

", "messages": [ { "id": 2093814, "postDate": "2023-01-10T10:42:16.830Z", "content": "

Hi there!

\n

I might be wrong, but I think that because at least 22% of this competition's datasets use Sigmoid as its VOILUTFunction, it's important to use some kind of windowing operation that Pydicom does.
\nBut, since DicomSDL is a lot faster, I've dug into the source code of Pydicom and tried to recreate the windowing operation so I could keep using DicomSDL.

\n

I've posted some results here.

\n

I think that if you use DicomSDL, you should check it out!
\nAnd if I'm wrong, please let me know! :)

", "rawMarkdown": "Hi there!\n\nI might be wrong, but I think that because at least 22% of this competition's datasets use Sigmoid as its VOILUTFunction, it's important to use some kind of windowing operation that Pydicom does.\nBut, since DicomSDL is a lot faster, I've dug into the source code of Pydicom and tried to recreate the windowing operation so I could keep using DicomSDL.\n\nI've posted some [results here](https://www.kaggle.com/code/bobdegraaf/dicomsdl-voi-lut).\n\nI think that if you use DicomSDL, you should check it out!\nAnd if I'm wrong, please let me know! :)", "votes": 13 }, { "id": 2106706, "postDate": "2023-01-19T10:11:25.167Z", "content": "

earlier it was mentioned that no voi_lut information for training data exists. I was assuming it meant 'slope is 1 and intercept is 0' and we are good to go? Now this sigmoid?
\n[https://www.kaggle.com/competitions/rsna-breast-cancer-detection/discussion/375419#2082409]
\nhow should we connect this and that?

", "rawMarkdown": "earlier it was mentioned that no voi_lut information for training data exists. I was assuming it meant 'slope is 1 and intercept is 0' and we are good to go? Now this sigmoid? \n[https://www.kaggle.com/competitions/rsna-breast-cancer-detection/discussion/375419#2082409]\nhow should we connect this and that? \n", "votes": 1, "replies": [ { "id": 2106841, "postDate": "2023-01-19T11:32:22.570Z", "content": "

Yeah, I've seen that discussion as well. I think they refer to the fact that there's no valid VOILUTSequence in all of the images. But once I saw the details of the windowing operation including the VOILUTFunction, mentioning the Sigmoid, I then checked the VOILUTFunctions of all images and saw that 22% had the SIGMOID. Then I checked if the images looked different, and as you can see in my notebook, they do.
\nI'm not an expert on VOILUT and I had little time, so all I did was check if I could figure out if I could create a similar windowing operation for the sigmoid images, so everyone could keep using DicomSDL. I hope a smarter person can explain more about VOILUT and what's happening here :)

", "rawMarkdown": "Yeah, I've seen that discussion as well. I think they refer to the fact that there's no valid VOILUTSequence in all of the images. But once I saw the details of the windowing operation including the VOILUTFunction, mentioning the Sigmoid, I then checked the VOILUTFunctions of all images and saw that 22% had the SIGMOID. Then I checked if the images looked different, and as you can see in my notebook, they do. \nI'm not an expert on VOILUT and I had little time, so all I did was check if I could figure out if I could create a similar windowing operation for the sigmoid images, so everyone could keep using DicomSDL. I hope a smarter person can explain more about VOILUT and what's happening here :) ", "votes": 3, "replies": [ { "id": 2106870, "postDate": "2023-01-19T11:59:41.137Z", "content": "

Makes sense thanks

", "rawMarkdown": "Makes sense thanks " } ] } ] }, { "id": 2093926, "postDate": "2023-01-10T13:00:22.697Z", "content": "

Works like a charm. 👍

", "rawMarkdown": "Works like a charm. 👍", "votes": 1 }, { "id": 2098960, "postDate": "2023-01-14T01:29:34.327Z", "content": "

Thanks for mentioning

", "rawMarkdown": "Thanks for mentioning", "votes": 1 } ], "comments": [ { "id": 2106706, "author_name": "GUNER", "author_url": "", "post_date": "2023-01-19T10:11:25.167000", "content": "

earlier it was mentioned that no voi_lut information for training data exists. I was assuming it meant 'slope is 1 and intercept is 0' and we are good to go? Now this sigmoid?
\n[https://www.kaggle.com/competitions/rsna-breast-cancer-detection/discussion/375419#2082409]
\nhow should we connect this and that?

", "votes": 1, "replies": [ { "id": 2106841, "author_name": "Bob de Graaf", "author_url": "", "post_date": "2023-01-19T11:32:22.570000", "content": "

Yeah, I've seen that discussion as well. I think they refer to the fact that there's no valid VOILUTSequence in all of the images. But once I saw the details of the windowing operation including the VOILUTFunction, mentioning the Sigmoid, I then checked the VOILUTFunctions of all images and saw that 22% had the SIGMOID. Then I checked if the images looked different, and as you can see in my notebook, they do.
\nI'm not an expert on VOILUT and I had little time, so all I did was check if I could figure out if I could create a similar windowing operation for the sigmoid images, so everyone could keep using DicomSDL. I hope a smarter person can explain more about VOILUT and what's happening here :)

", "votes": 3, "replies": [ { "id": 2106870, "author_name": "GUNER", "author_url": "", "post_date": "2023-01-19T11:59:41.137000", "content": "

Makes sense thanks

", "votes": 0, "replies": [] } ] } ] }, { "id": 2093926, "author_name": "Remek Kinas", "author_url": "", "post_date": "2023-01-10T13:00:22.697000", "content": "

Works like a charm. 👍

", "votes": 1, "replies": [] }, { "id": 2098960, "author_name": "Nhan Vi", "author_url": "", "post_date": "2023-01-14T01:29:34.327000", "content": "

Thanks for mentioning

", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "2093814": "Hi there!\n\nI might be wrong, but I think that because at least 22% of this competition's datasets use Sigmoid as its VOILUTFunction, it's important to use some kind of windowing operation that Pydicom does.\nBut, since DicomSDL is a lot faster, I've dug into the source code of Pydicom and tried to recreate the windowing operation so I could keep using DicomSDL.\n\nI've posted some [results here](https://www.kaggle.com/code/bobdegraaf/dicomsdl-voi-lut).\n\nI think that if you use DicomSDL, you should check it out!\nAnd if I'm wrong, please let me know! :)", "2106706": "earlier it was mentioned that no voi_lut information for training data exists. I was assuming it meant 'slope is 1 and intercept is 0' and we are good to go? Now this sigmoid? \n[https://www.kaggle.com/competitions/rsna-breast-cancer-detection/discussion/375419#2082409]\nhow should we connect this and that? \n", "2093926": "Works like a charm. 👍", "2098960": "Thanks for mentioning" } }