{ "id": 375520, "title": "Yet another ADMANI model: BRAIxProtoPNet++ ", "url": "/competitions/rsna-breast-cancer-detection/discussion/375520", "author_name": "hengck23", "post_date": "2023-01-02T05:02:25.830000", "votes": 12, "comment_count": 4, "views": 0, "content": "

paper

\n

[1] Knowledge Distillation to Ensemble Global and Interpretable Prototype-based Mammogram Classification Models
\nhttps://arxiv.org/abs/2209.12420

\n

[2] This Looks Like That: Deep Learning for Interpretable Image Recognition
\nhttps://arxiv.org/pdf/1806.10574.pdf

\n
\n

kaggle notebook

\n

doing final checking …. live on 03-Jan-2023 (meanwhile you can read the paper first)
\n(https://www.kaggle.com/code/hengck23/yet-another-admani-model-braixprotopnet)

\n

\"https://i.ibb.co/SxzBc4P/8i27of.gif\"

\n
\n

Basically an advance prototype net (or i i think is a modern KNN classifier):

\n

\"Selection-446\"
\n\"Selection-444\"
\n\"Selection-443\"

", "messages": [ { "id": 2082906, "postDate": "2023-01-02T05:02:25.830Z", "content": "

paper

\n

[1] Knowledge Distillation to Ensemble Global and Interpretable Prototype-based Mammogram Classification Models
\nhttps://arxiv.org/abs/2209.12420

\n

[2] This Looks Like That: Deep Learning for Interpretable Image Recognition
\nhttps://arxiv.org/pdf/1806.10574.pdf

\n
\n

kaggle notebook

\n

doing final checking …. live on 03-Jan-2023 (meanwhile you can read the paper first)
\n(https://www.kaggle.com/code/hengck23/yet-another-admani-model-braixprotopnet)

\n

\"https://i.ibb.co/SxzBc4P/8i27of.gif\"

\n
\n

Basically an advance prototype net (or i i think is a modern KNN classifier):

\n

\"Selection-446\"
\n\"Selection-444\"
\n\"Selection-443\"

", "rawMarkdown": "**paper**\n\n[1] Knowledge Distillation to Ensemble Global and Interpretable Prototype-based Mammogram Classification Models\nhttps://arxiv.org/abs/2209.12420\n\n[2] This Looks Like That: Deep Learning for Interpretable Image Recognition\nhttps://arxiv.org/pdf/1806.10574.pdf\n\n---\n\n**kaggle notebook**\n\ndoing final checking .... live on 03-Jan-2023 (meanwhile you can read the paper first)\n(https://www.kaggle.com/code/hengck23/yet-another-admani-model-braixprotopnet)\n\n![https://i.ibb.co/SxzBc4P/8i27of.gif](https://i.ibb.co/SxzBc4P/8i27of.gif)\n\n\n\n---\n\n\nBasically an advance prototype net (or i i think is a modern KNN classifier):\n\n\"Selection-446\"\n\"Selection-444\"\n\"Selection-443\"\n\n\n\n\n\n", "votes": 12 }, { "id": 2084033, "postDate": "2023-01-03T04:45:49.863Z", "content": "

more results from paper
\n[1] AI integration improves breast cancer screening in a real-world, retrospective cohort study
\nhttps://www.medrxiv.org/content/medrxiv/early/2022/11/28/2022.11.23.22282646.full.pdf

\n

AI ensemble model consisting of:

\n\n
\n

\"dataset comprising 808,318 episodes, 577,576 clients and 3,404,326 images in the period 2013 to

\n
    \n
  1. \"
    2. \n
\n
\n

\"We evaluated our AI reader on our testing set and on external datasets previously unseen by our models.\"
\nThe are having AUC of >=92. Can we expect this quality of solution in this kaggle competition?

", "rawMarkdown": "more results from paper\n[1] AI integration improves breast cancer screening in a real-world, retrospective cohort study\nhttps://www.medrxiv.org/content/medrxiv/early/2022/11/28/2022.11.23.22282646.full.pdf\n\nAI ensemble model consisting of:\n- eight individual CNN models, \n- four off-the-shelf CNN models \n- four model architectures developed specifically for this problem\n - BRAIxMVCCL, uses two image-views of a single breast as inputs\n - BRAIxProtoP-Net++ framework\n\n---\n\n\"dataset comprising 808,318 episodes, 577,576 clients and 3,404,326 images in the period 2013 to\n2019. \"\n\n---\n\n\"We evaluated our AI reader on our testing set and on external datasets previously unseen by our models.\"\nThe are having AUC of >=92. Can we expect this quality of solution in this kaggle competition?\n", "votes": 1 }, { "id": 2084024, "postDate": "2023-01-03T04:27:17.813Z", "content": "

there seems to be smarter implementation using trasnformer:

\n

[1] ProtoPFormer: Concentrating on Prototypical Parts in Vision Transformers for Interpretable Image Recognition
\nhttps://paperswithcode.com/paper/protopformer-concentrating-on-prototypical

\n

Here, the cls token is replacing the global-net.
\nInteractive of global and local is in-built because of the attention mechanism in transformer

\n

\"Selection-460.png\"

", "rawMarkdown": "there seems to be smarter implementation using trasnformer:\n\n[1] ProtoPFormer: Concentrating on Prototypical Parts in Vision Transformers for Interpretable Image Recognition \nhttps://paperswithcode.com/paper/protopformer-concentrating-on-prototypical\n\nHere, the cls token is replacing the global-net.\nInteractive of global and local is in-built because of the attention mechanism in transformer\n\n[![Selection-460.png](https://i.postimg.cc/Hk0PHtnM/Selection-460.png)](https://postimg.cc/G42jxGk3)\n", "votes": 1 }, { "id": 2082951, "postDate": "2023-01-02T06:20:34.393Z", "content": "

good background reading for protoytpe net

\n

This Looks Like That: Deep Learning for Interpretable Image Recognition (AI Paper Summary)
\nhttps://www.youtube.com/watch?v=v3direJ7ZWU

\n

detail explanation
\nThis Looks Like That: Deep Learning for Interpretable Image Recognition (NeurIPS 2019)
\nhttps://www.youtube.com/watch?v=k3IQnRsl9U4

\n

application to breast cancer mammography screening
\nAlina Barnett - Interpretable Image Recognition
\n\"https://i.ibb.co/GTGW0QJ/Selection-447.png\"
\nhttps://www.youtube.com/watch?v=-IkQ5CbVTkE

\n

https://arxiv.org/pdf/2107.05605.pdf
\nInterpretable Mammographic Image Classification using Case-Based Reasoning and Deep Learning
\nhttps://github.com/alinajadebarnett/iaiabl

\n

related:
\nPrototpyical Networks for Few-shot Learning (not quite the same but very related)
\nhttps://www.kaggle.com/competitions/humpback-whale-identification/discussion/81085
\nhttps://arxiv.org/pdf/1703.05175.pdf

", "rawMarkdown": "good background reading for protoytpe net\n\nThis Looks Like That: Deep Learning for Interpretable Image Recognition (AI Paper Summary)\nhttps://www.youtube.com/watch?v=v3direJ7ZWU\n\n\ndetail explanation\nThis Looks Like That: Deep Learning for Interpretable Image Recognition (NeurIPS 2019)\nhttps://www.youtube.com/watch?v=k3IQnRsl9U4\n\napplication to breast cancer mammography screening \nAlina Barnett - Interpretable Image Recognition\n![https://i.ibb.co/GTGW0QJ/Selection-447.png](https://i.ibb.co/GTGW0QJ/Selection-447.png)\nhttps://www.youtube.com/watch?v=-IkQ5CbVTkE\n\nhttps://arxiv.org/pdf/2107.05605.pdf\nInterpretable Mammographic Image Classification using Case-Based Reasoning and Deep Learning\nhttps://github.com/alinajadebarnett/iaiabl\n\n\nrelated:\nPrototpyical Networks for Few-shot Learning (not quite the same but very related)\nhttps://www.kaggle.com/competitions/humpback-whale-identification/discussion/81085\nhttps://arxiv.org/pdf/1703.05175.pdf\n", "votes": 1 }, { "id": 2096905, "postDate": "2023-01-12T10:36:22.003Z", "content": "

Thank you for putting this all together!
\nReally cool!

\n

The Devastator.

", "rawMarkdown": "Thank you for putting this all together!\nReally cool!\n\nThe Devastator.\n" } ], "comments": [ { "id": 2084033, "author_name": "hengck23", "author_url": "", "post_date": "2023-01-03T04:45:49.863000", "content": "

more results from paper
\n[1] AI integration improves breast cancer screening in a real-world, retrospective cohort study
\nhttps://www.medrxiv.org/content/medrxiv/early/2022/11/28/2022.11.23.22282646.full.pdf

\n

AI ensemble model consisting of:

\n\n
\n

\"dataset comprising 808,318 episodes, 577,576 clients and 3,404,326 images in the period 2013 to

\n
    \n
  1. \"
    2. \n
\n
\n

\"We evaluated our AI reader on our testing set and on external datasets previously unseen by our models.\"
\nThe are having AUC of >=92. Can we expect this quality of solution in this kaggle competition?

", "votes": 1, "replies": [] }, { "id": 2084024, "author_name": "hengck23", "author_url": "", "post_date": "2023-01-03T04:27:17.813000", "content": "

there seems to be smarter implementation using trasnformer:

\n

[1] ProtoPFormer: Concentrating on Prototypical Parts in Vision Transformers for Interpretable Image Recognition
\nhttps://paperswithcode.com/paper/protopformer-concentrating-on-prototypical

\n

Here, the cls token is replacing the global-net.
\nInteractive of global and local is in-built because of the attention mechanism in transformer

\n

\"Selection-460.png\"

", "votes": 1, "replies": [] }, { "id": 2082951, "author_name": "hengck23", "author_url": "", "post_date": "2023-01-02T06:20:34.393000", "content": "

good background reading for protoytpe net

\n

This Looks Like That: Deep Learning for Interpretable Image Recognition (AI Paper Summary)
\nhttps://www.youtube.com/watch?v=v3direJ7ZWU

\n

detail explanation
\nThis Looks Like That: Deep Learning for Interpretable Image Recognition (NeurIPS 2019)
\nhttps://www.youtube.com/watch?v=k3IQnRsl9U4

\n

application to breast cancer mammography screening
\nAlina Barnett - Interpretable Image Recognition
\n\"https://i.ibb.co/GTGW0QJ/Selection-447.png\"
\nhttps://www.youtube.com/watch?v=-IkQ5CbVTkE

\n

https://arxiv.org/pdf/2107.05605.pdf
\nInterpretable Mammographic Image Classification using Case-Based Reasoning and Deep Learning
\nhttps://github.com/alinajadebarnett/iaiabl

\n

related:
\nPrototpyical Networks for Few-shot Learning (not quite the same but very related)
\nhttps://www.kaggle.com/competitions/humpback-whale-identification/discussion/81085
\nhttps://arxiv.org/pdf/1703.05175.pdf

", "votes": 1, "replies": [] }, { "id": 2096905, "author_name": "The Devastator", "author_url": "", "post_date": "2023-01-12T10:36:22.003000", "content": "

Thank you for putting this all together!
\nReally cool!

\n

The Devastator.

", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2082906": "**paper**\n\n[1] Knowledge Distillation to Ensemble Global and Interpretable Prototype-based Mammogram Classification Models\nhttps://arxiv.org/abs/2209.12420\n\n[2] This Looks Like That: Deep Learning for Interpretable Image Recognition\nhttps://arxiv.org/pdf/1806.10574.pdf\n\n---\n\n**kaggle notebook**\n\ndoing final checking .... live on 03-Jan-2023 (meanwhile you can read the paper first)\n(https://www.kaggle.com/code/hengck23/yet-another-admani-model-braixprotopnet)\n\n![https://i.ibb.co/SxzBc4P/8i27of.gif](https://i.ibb.co/SxzBc4P/8i27of.gif)\n\n\n\n---\n\n\nBasically an advance prototype net (or i i think is a modern KNN classifier):\n\n\"Selection-446\"\n\"Selection-444\"\n\"Selection-443\"\n\n\n\n\n\n", "2084033": "more results from paper\n[1] AI integration improves breast cancer screening in a real-world, retrospective cohort study\nhttps://www.medrxiv.org/content/medrxiv/early/2022/11/28/2022.11.23.22282646.full.pdf\n\nAI ensemble model consisting of:\n- eight individual CNN models, \n- four off-the-shelf CNN models \n- four model architectures developed specifically for this problem\n - BRAIxMVCCL, uses two image-views of a single breast as inputs\n - BRAIxProtoP-Net++ framework\n\n---\n\n\"dataset comprising 808,318 episodes, 577,576 clients and 3,404,326 images in the period 2013 to\n2019. \"\n\n---\n\n\"We evaluated our AI reader on our testing set and on external datasets previously unseen by our models.\"\nThe are having AUC of >=92. Can we expect this quality of solution in this kaggle competition?\n", "2084024": "there seems to be smarter implementation using trasnformer:\n\n[1] ProtoPFormer: Concentrating on Prototypical Parts in Vision Transformers for Interpretable Image Recognition \nhttps://paperswithcode.com/paper/protopformer-concentrating-on-prototypical\n\nHere, the cls token is replacing the global-net.\nInteractive of global and local is in-built because of the attention mechanism in transformer\n\n[![Selection-460.png](https://i.postimg.cc/Hk0PHtnM/Selection-460.png)](https://postimg.cc/G42jxGk3)\n", "2082951": "good background reading for protoytpe net\n\nThis Looks Like That: Deep Learning for Interpretable Image Recognition (AI Paper Summary)\nhttps://www.youtube.com/watch?v=v3direJ7ZWU\n\n\ndetail explanation\nThis Looks Like That: Deep Learning for Interpretable Image Recognition (NeurIPS 2019)\nhttps://www.youtube.com/watch?v=k3IQnRsl9U4\n\napplication to breast cancer mammography screening \nAlina Barnett - Interpretable Image Recognition\n![https://i.ibb.co/GTGW0QJ/Selection-447.png](https://i.ibb.co/GTGW0QJ/Selection-447.png)\nhttps://www.youtube.com/watch?v=-IkQ5CbVTkE\n\nhttps://arxiv.org/pdf/2107.05605.pdf\nInterpretable Mammographic Image Classification using Case-Based Reasoning and Deep Learning\nhttps://github.com/alinajadebarnett/iaiabl\n\n\nrelated:\nPrototpyical Networks for Few-shot Learning (not quite the same but very related)\nhttps://www.kaggle.com/competitions/humpback-whale-identification/discussion/81085\nhttps://arxiv.org/pdf/1703.05175.pdf\n", "2096905": "Thank you for putting this all together!\nReally cool!\n\nThe Devastator.\n" } }