{ "id": 145304, "title": "Best Score from Previous Competitions", "url": "/competitions/prostate-cancer-grade-assessment/discussion/145304", "author_name": "Nanashi", "post_date": "2020-04-22T16:51:27.311000", "votes": 17, "comment_count": 2, "views": 0, "content": "

AUTOMATIC PROSTATE GLEASON GRADING CHALLENGE 2019

\n\n

> This challenge is part of the MICCAI 2019 Conference

\n\n

As you can check at results the LB was:\n|Rank |UserID |Score| \n| ------------- |:-------------:| -----:|
\n|1 |YujinHu |0.845151814|
\n|2 |nitinsinghal |0.792585244|
\n|3 |ternaus| 0.789663211|
\n|4 |zhangjingmri|0.778060956|
\n|5 |sdsy888 |0.759776281|

\n\n

The metric is a combination of Cohen's kappa and the F1-scores.\nscore= Cohen's kappa + (macro-averaged F1-score + micro-averaged F1-score)/2.

\n\n

However, the tasks is a bit different:\n- Task 1: Pixel-level Gleason grade prediction\n- Task 2: Core-level Gleason score prediction

\n\n

Considering that a \"simple\" model trained on 256x256 images could score 0.55 panda-submit-test... are the top scores going to be above 0.8 here too? For kappa that's really cool :)

\n\n

EDIT:\nThe best public score now using SEResnet is 0.64: https://www.kaggle.com/rohitsingh9990/panda-resnext-inference

\n\n
\n

I guess we'll see kappa score > 0.8 at the LB

\n
", "messages": [ { "id": 816849, "postDate": "2020-04-22T16:51:27.310Z", "content": "

AUTOMATIC PROSTATE GLEASON GRADING CHALLENGE 2019

\n\n

> This challenge is part of the MICCAI 2019 Conference

\n\n

As you can check at results the LB was:\n|Rank |UserID |Score| \n| ------------- |:-------------:| -----:|
\n|1 |YujinHu |0.845151814|
\n|2 |nitinsinghal |0.792585244|
\n|3 |ternaus| 0.789663211|
\n|4 |zhangjingmri|0.778060956|
\n|5 |sdsy888 |0.759776281|

\n\n

The metric is a combination of Cohen's kappa and the F1-scores.\nscore= Cohen's kappa + (macro-averaged F1-score + micro-averaged F1-score)/2.

\n\n

However, the tasks is a bit different:\n- Task 1: Pixel-level Gleason grade prediction\n- Task 2: Core-level Gleason score prediction

\n\n

Considering that a \"simple\" model trained on 256x256 images could score 0.55 panda-submit-test... are the top scores going to be above 0.8 here too? For kappa that's really cool :)

\n\n

EDIT:\nThe best public score now using SEResnet is 0.64: https://www.kaggle.com/rohitsingh9990/panda-resnext-inference

\n\n
\n

I guess we'll see kappa score > 0.8 at the LB

\n
", "rawMarkdown": "###[AUTOMATIC PROSTATE GLEASON GRADING CHALLENGE 2019](https://gleason2019.grand-challenge.org/Home/)\n> This challenge is part of the MICCAI 2019 Conference\n\nAs you can check at **results** the LB was:\n|Rank\t|UserID\t|Score|\t\n| ------------- |:-------------:| -----:|\t\t\t\t\t\t\n|1\t|YujinHu\t|**0.845151814**|\t\t\n|2\t|nitinsinghal\t|0.792585244|\t\t\t\n|3\t|ternaus|\t0.789663211|\t\t\t\t\n|4\t|zhangjingmri|0.778060956|\t\t\t\n|5\t|sdsy888\t|0.759776281|\t\t\t\t\t\t\t\n\nThe metric is a combination of Cohen's kappa and the F1-scores.\n`score= Cohen's kappa + (macro-averaged F1-score + micro-averaged F1-score)/2.`\n\nHowever, the tasks is a bit different:\n- Task 1: Pixel-level Gleason grade prediction\n- Task 2: Core-level Gleason score prediction\n\nConsidering that a \"simple\" model trained on 256x256 images could score 0.55 [panda-submit-test](https://www.kaggle.com/yasufuminakama/panda-submit-test)... are the top scores going to be above 0.8 here too? For **kappa** that's really cool :)\n\nEDIT:\nThe best public score now using SEResnet is 0.64: https://www.kaggle.com/rohitsingh9990/panda-resnext-inference\n\n> I guess we'll see kappa score > 0.8 at the LB ", "votes": 15 }, { "id": 818810, "postDate": "2020-04-24T06:15:23.147Z", "content": "

Thanks for sharing!!

", "rawMarkdown": "Thanks for sharing!!" }, { "id": 816875, "postDate": "2020-04-22T17:14:02.370Z", "content": "

thanks for sharing :)

", "rawMarkdown": "thanks for sharing :)" } ], "comments": [ { "id": 818810, "author_name": "NaturalNeuralNetwork", "author_url": "", "post_date": "2020-04-24T06:15:23.147000", "content": "

Thanks for sharing!!

", "votes": 0, "replies": [] }, { "id": 816875, "author_name": "Alberto Maria Falletta", "author_url": "", "post_date": "2020-04-22T17:14:02.370000", "content": "

thanks for sharing :)

", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "816849": "###[AUTOMATIC PROSTATE GLEASON GRADING CHALLENGE 2019](https://gleason2019.grand-challenge.org/Home/)\n> This challenge is part of the MICCAI 2019 Conference\n\nAs you can check at **results** the LB was:\n|Rank\t|UserID\t|Score|\t\n| ------------- |:-------------:| -----:|\t\t\t\t\t\t\n|1\t|YujinHu\t|**0.845151814**|\t\t\n|2\t|nitinsinghal\t|0.792585244|\t\t\t\n|3\t|ternaus|\t0.789663211|\t\t\t\t\n|4\t|zhangjingmri|0.778060956|\t\t\t\n|5\t|sdsy888\t|0.759776281|\t\t\t\t\t\t\t\n\nThe metric is a combination of Cohen's kappa and the F1-scores.\n`score= Cohen's kappa + (macro-averaged F1-score + micro-averaged F1-score)/2.`\n\nHowever, the tasks is a bit different:\n- Task 1: Pixel-level Gleason grade prediction\n- Task 2: Core-level Gleason score prediction\n\nConsidering that a \"simple\" model trained on 256x256 images could score 0.55 [panda-submit-test](https://www.kaggle.com/yasufuminakama/panda-submit-test)... are the top scores going to be above 0.8 here too? For **kappa** that's really cool :)\n\nEDIT:\nThe best public score now using SEResnet is 0.64: https://www.kaggle.com/rohitsingh9990/panda-resnext-inference\n\n> I guess we'll see kappa score > 0.8 at the LB ", "818810": "Thanks for sharing!!", "816875": "thanks for sharing :)" } }