{ "id": 271524, "title": "Scan 00998 a bad data?", "url": "/competitions/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/271524", "author_name": "", "post_date": "2021-09-11T02:22:04.647673400Z", "votes": 13, "comment_count": 3, "views": 0, "content": "

I visualized the scan 00998 FLAIR, and it looks something like this. This scan looks like it has way too much brightness that it looks like there is nothing useful in this scan for the model to learn. Compared to the other scans.
\nDo you think we should exclude this while training ?

", "messages": [ { "id": "1509190", "postDate": "09/11/2021 02:22:04", "content": "

I visualized the scan 00998 FLAIR, and it looks something like this. This scan looks like it has way too much brightness that it looks like there is nothing useful in this scan for the model to learn. Compared to the other scans.
\nDo you think we should exclude this while training ?

", "rawMarkdown": "I visualized the scan 00998 FLAIR, and it looks something like [this](https://drive.google.com/file/d/1eSHKpQ0WnUXXDApHm6NTKJbwre-MSrGW/view?usp=sharing). This scan looks like it has way too much brightness that it looks like there is nothing useful in this scan for the model to learn. Compared to the other [scans](https://drive.google.com/file/d/1LSGm3qokt11Uiapi3_Oxx2IN1jRJM_Vu/view?usp=sharing).\nDo you think we should exclude this while training ?", "votes": null }, { "id": "1510041", "postDate": "09/12/2021 02:49:44", "content": "

I don't have any problems exporting study 00998. The FLAIR series is somewhat low quality, but can be exported decently. How did you normalize it?

\n

Try using a LUT like in this notebook -> https://www.kaggle.com/davidbroberts/standardizing-mr-images

", "rawMarkdown": "I don't have any problems exporting study 00998. The FLAIR series is somewhat low quality, but can be exported decently. How did you normalize it? \n\nTry using a LUT like in this notebook -> https://www.kaggle.com/davidbroberts/standardizing-mr-images", "votes": null }, { "id": "1510244", "postDate": "09/12/2021 08:13:51", "content": "

Thanks, Roberts!
\nbut my doubt is that the scan looks way too bland that Is there even anything inside it for the model to learn? I can see nothing but the shape of the brain in that study.
\nI'm not normalizing anything It's just a plot of images from the png dataset. and it's the same for all 4 mri types.
\ndoes normalizing change anything? I'll try doing it the way you did it in your notebook

", "rawMarkdown": "Thanks, Roberts!\nbut my doubt is that the scan looks way too bland that Is there even anything inside it for the model to learn? I can see nothing but the shape of the brain in that study.\nI'm not normalizing anything It's just a plot of images from the png dataset. and it's the same for all 4 mri types.\ndoes normalizing change anything? I'll try doing it the way you did it in your notebook", "votes": null }, { "id": "1510504", "postDate": "09/12/2021 13:21:49", "content": "

I didn't realize you were using a PNG dataset. Try re-exporting it from DICOM. It isn't normalized properly. You're not seeing how the images actually look.

", "rawMarkdown": "I didn't realize you were using a PNG dataset. Try re-exporting it from DICOM. It isn't normalized properly. You're not seeing how the images actually look.", "votes": null } ], "comments": [ { "id": 1510041, "author_name": "davidbroberts", "author_url": "", "post_date": "09/12/2021 02:49:44", "content": "

I don't have any problems exporting study 00998. The FLAIR series is somewhat low quality, but can be exported decently. How did you normalize it?

\n

Try using a LUT like in this notebook -> https://www.kaggle.com/davidbroberts/standardizing-mr-images

", "votes": null, "replies": [ { "id": 1510244, "author_name": "theunrealsamurai", "author_url": "", "post_date": "09/12/2021 08:13:51", "content": "

Thanks, Roberts!
\nbut my doubt is that the scan looks way too bland that Is there even anything inside it for the model to learn? I can see nothing but the shape of the brain in that study.
\nI'm not normalizing anything It's just a plot of images from the png dataset. and it's the same for all 4 mri types.
\ndoes normalizing change anything? I'll try doing it the way you did it in your notebook

", "votes": null, "replies": [] }, { "id": 1510504, "author_name": "davidbroberts", "author_url": "", "post_date": "09/12/2021 13:21:49", "content": "

I didn't realize you were using a PNG dataset. Try re-exporting it from DICOM. It isn't normalized properly. You're not seeing how the images actually look.

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1509190": "I visualized the scan 00998 FLAIR, and it looks something like [this](https://drive.google.com/file/d/1eSHKpQ0WnUXXDApHm6NTKJbwre-MSrGW/view?usp=sharing). This scan looks like it has way too much brightness that it looks like there is nothing useful in this scan for the model to learn. Compared to the other [scans](https://drive.google.com/file/d/1LSGm3qokt11Uiapi3_Oxx2IN1jRJM_Vu/view?usp=sharing).\nDo you think we should exclude this while training ?", "1510041": "I don't have any problems exporting study 00998. The FLAIR series is somewhat low quality, but can be exported decently. How did you normalize it? \n\nTry using a LUT like in this notebook -> https://www.kaggle.com/davidbroberts/standardizing-mr-images", "1510244": "Thanks, Roberts!\nbut my doubt is that the scan looks way too bland that Is there even anything inside it for the model to learn? I can see nothing but the shape of the brain in that study.\nI'm not normalizing anything It's just a plot of images from the png dataset. and it's the same for all 4 mri types.\ndoes normalizing change anything? I'll try doing it the way you did it in your notebook", "1510504": "I didn't realize you were using a PNG dataset. Try re-exporting it from DICOM. It isn't normalized properly. You're not seeing how the images actually look." }, "source": "meta" }