{ "id": 337651, "title": "All you need is Multiple Instance Learning", "url": "/competitions/mayo-clinic-strip-ai/discussion/337651", "author_name": "RabotniKuma", "post_date": "2022-07-17T01:47:33.596000", "votes": 39, "comment_count": 5, "views": 0, "content": "

One of the most challenging aspects of this competition is how to exclude the background and incorporate the areas of interest into the model. This problem can be solved by extracting tissue tiles(-> My notebook).

\n

However, the next problem is how to associate multiple tiles that may not contain a signal with a single label. Multiple Instance Learning (MIL) solves this problem.

\n

\"\"

\n

I published my implementation of a simple MIL CNN model(-> My notebook )

", "messages": [ { "id": 1858508, "postDate": "2022-07-17T01:47:33.597Z", "content": "

One of the most challenging aspects of this competition is how to exclude the background and incorporate the areas of interest into the model. This problem can be solved by extracting tissue tiles(-> My notebook).

\n

However, the next problem is how to associate multiple tiles that may not contain a signal with a single label. Multiple Instance Learning (MIL) solves this problem.

\n

\"\"

\n

I published my implementation of a simple MIL CNN model(-> My notebook )

", "rawMarkdown": "One of the most challenging aspects of this competition is how to exclude the background and incorporate the areas of interest into the model. This problem can be solved by extracting tissue tiles([-> My notebook](https://www.kaggle.com/code/analokamus/a-fast-tile-generation)).\n\nHowever, the next problem is how to associate multiple tiles that may not contain a signal with a single label. Multiple Instance Learning (MIL) solves this problem.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1973217%2F1005d61e62b31ba4f5bac52a354fae79%2Fkaggle-mayo.001.png?generation=1658022399002616&alt=media)\n\nI published my implementation of a simple MIL CNN model([-> My notebook](https://www.kaggle.com/code/analokamus/a-sample-of-multi-instance-learning-model/notebook) )", "votes": 37 }, { "id": 1858914, "postDate": "2022-07-17T09:25:34.637Z", "content": "

Fantastic work, I stand to learn a lot from this! Great notebook! Thanks for sharing!

", "rawMarkdown": "Fantastic work, I stand to learn a lot from this! Great notebook! Thanks for sharing!", "votes": 1 }, { "id": 1859279, "postDate": "2022-07-17T14:05:13.450Z", "content": "

Thanks for sharing this, can vision transformers be also used for this challenge assuming the background patches are removed in a preprocessing step. Looking forward to any ideas or suggestions.

", "rawMarkdown": "Thanks for sharing this, can vision transformers be also used for this challenge assuming the background patches are removed in a preprocessing step. Looking forward to any ideas or suggestions. ", "replies": [ { "id": 1877677, "postDate": "2022-07-30T22:41:20.647Z", "rawMarkdown": "", "votes": 1, "isDeleted": true }, { "id": 1892677, "postDate": "2022-08-10T08:45:53.180Z", "rawMarkdown": "", "votes": 1, "isDeleted": true } ] }, { "id": 1858643, "postDate": "2022-07-17T05:20:32.103Z", "content": "

@analokamus thanks for sharing!!

", "rawMarkdown": "@analokamus thanks for sharing!! " } ], "comments": [ { "id": 1858914, "author_name": "Ravi Ramakrishnan", "author_url": "", "post_date": "2022-07-17T09:25:34.637000", "content": "

Fantastic work, I stand to learn a lot from this! Great notebook! Thanks for sharing!

", "votes": 1, "replies": [] }, { "id": 1859279, "author_name": "OUA", "author_url": "", "post_date": "2022-07-17T14:05:13.450000", "content": "

Thanks for sharing this, can vision transformers be also used for this challenge assuming the background patches are removed in a preprocessing step. Looking forward to any ideas or suggestions.

", "votes": 0, "replies": [ { "id": 1877677, "author_name": "", "author_url": "", "post_date": "2022-07-30T22:41:20.647000", "content": "", "votes": 1, "replies": [] }, { "id": 1892677, "author_name": "", "author_url": "", "post_date": "2022-08-10T08:45:53.180000", "content": "", "votes": 1, "replies": [] } ] }, { "id": 1858643, "author_name": "Arun Purakkatt", "author_url": "", "post_date": "2022-07-17T05:20:32.103000", "content": "

@analokamus thanks for sharing!!

", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1858508": "One of the most challenging aspects of this competition is how to exclude the background and incorporate the areas of interest into the model. This problem can be solved by extracting tissue tiles([-> My notebook](https://www.kaggle.com/code/analokamus/a-fast-tile-generation)).\n\nHowever, the next problem is how to associate multiple tiles that may not contain a signal with a single label. Multiple Instance Learning (MIL) solves this problem.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F1973217%2F1005d61e62b31ba4f5bac52a354fae79%2Fkaggle-mayo.001.png?generation=1658022399002616&alt=media)\n\nI published my implementation of a simple MIL CNN model([-> My notebook](https://www.kaggle.com/code/analokamus/a-sample-of-multi-instance-learning-model/notebook) )", "1858914": "Fantastic work, I stand to learn a lot from this! Great notebook! Thanks for sharing!", "1859279": "Thanks for sharing this, can vision transformers be also used for this challenge assuming the background patches are removed in a preprocessing step. Looking forward to any ideas or suggestions. ", "1858643": "@analokamus thanks for sharing!! " } }