Welcome to the Prostate cANcer graDe Assessment (PANDA) challenge! Prostate cancer is a leading cause of cancer death. Each year, pathologists assess millions of prostate tissue samples to diagnose cancer and assess its severity. This is not only labor-intensive but it is also a very difficult task for humans. An incorrect diagnosis may lead to suboptimal treatment, which causes unnecessary side effects or shortened life expectancy of the patient. In this competition, we need your help to improve prostate cancer diagnostics to the benefit of millions of men worldwide.
\n\nTo get you up to speed with the dataset of the competition, we have made a getting started notebook. All images in the dataset are whole-slide images (WSI) and the notebook contains some starting points on using these types of images.
\n\nDuring the competition, we will be following the forums and are available to answer any questions you might have. Feel free to post questions regarding the data or the (medical) background, we are here to help!
\n\nThis competition is also part of a scientific conference and we will write a paper on the results, see the page on the challenge workshop for more information.
\n\nWe would like to extend our gratitude to the pathologists who graded the test sets used in the competition: Lars Egevad, Brett Delahunt, Hemamali Samaratunga, Toyonori Tsuzuki, Christina Hulsbergen-van de Kaa, Robert Vink, and Hester van Boven.
\n\nWe are looking forward to seeing your methods for solving this important problem! Good luck with the challenge!
\n\nOn behalf of the whole organizing team,
\n\nPeter, Kimmo, Martin, Geert & Wouter
", "messages": [ { "id": 815581, "postDate": "2020-04-21T16:47:32.810Z", "content": "Welcome to the Prostate cANcer graDe Assessment (PANDA) challenge! Prostate cancer is a leading cause of cancer death. Each year, pathologists assess millions of prostate tissue samples to diagnose cancer and assess its severity. This is not only labor-intensive but it is also a very difficult task for humans. An incorrect diagnosis may lead to suboptimal treatment, which causes unnecessary side effects or shortened life expectancy of the patient. In this competition, we need your help to improve prostate cancer diagnostics to the benefit of millions of men worldwide.
\n\nTo get you up to speed with the dataset of the competition, we have made a getting started notebook. All images in the dataset are whole-slide images (WSI) and the notebook contains some starting points on using these types of images.
\n\nDuring the competition, we will be following the forums and are available to answer any questions you might have. Feel free to post questions regarding the data or the (medical) background, we are here to help!
\n\nThis competition is also part of a scientific conference and we will write a paper on the results, see the page on the challenge workshop for more information.
\n\nWe would like to extend our gratitude to the pathologists who graded the test sets used in the competition: Lars Egevad, Brett Delahunt, Hemamali Samaratunga, Toyonori Tsuzuki, Christina Hulsbergen-van de Kaa, Robert Vink, and Hester van Boven.
\n\nWe are looking forward to seeing your methods for solving this important problem! Good luck with the challenge!
\n\nOn behalf of the whole organizing team,
\n\nPeter, Kimmo, Martin, Geert & Wouter
", "rawMarkdown": "Welcome to the **Prostate cANcer graDe Assessment (PANDA)** challenge! Prostate cancer is a leading cause of cancer death. Each year, pathologists assess millions of prostate tissue samples to diagnose cancer and assess its severity. This is not only labor-intensive but it is also a very difficult task for humans. An incorrect diagnosis may lead to suboptimal treatment, which causes unnecessary side effects or shortened life expectancy of the patient. In this competition, we need your help to improve prostate cancer diagnostics to the benefit of millions of men worldwide.\n\nTo get you up to speed with the dataset of the competition, we have made a [getting started notebook](https://www.kaggle.com/wouterbulten/getting-started-with-the-panda-dataset). All images in the dataset are whole-slide images (WSI) and the notebook contains some starting points on using these types of images.\n\nDuring the competition, we will be following the forums and are available to answer any questions you might have. Feel free to post questions regarding the data or the (medical) background, we are here to help!\n\nThis competition is also part of a scientific conference and we will write a paper on the results, see the page on the [challenge workshop](https://www.kaggle.com/c/prostate-cancer-grade-assessment/overview/miccai-2020) for more information.\n\nWe would like to extend our gratitude to the pathologists who graded the test sets used in the competition: Lars Egevad, Brett Delahunt, Hemamali Samaratunga, Toyonori Tsuzuki, Christina Hulsbergen-van de Kaa, Robert Vink, and Hester van Boven.\n\nWe are looking forward to seeing your methods for solving this important problem! Good luck with the challenge!\n\nOn behalf of the whole organizing team,\n\n[Peter](https://www.kaggle.com/petstr), [Kimmo](https://www.kaggle.com/kimmokartasalo), [Martin](https://www.kaggle.com/martineklund), [Geert](https://www.kaggle.com/geertlitjens) & [Wouter](https://www.kaggle.com/wouterbulten)\n", "votes": 34 }, { "id": 847824, "postDate": "2020-05-14T16:11:10.700Z", "content": "Hi. \nI need to ask something. As in Masks of Karolinska, We have only 0, 1 and 2. Where 2 stands for cancerous part. But it doesn't provide information about Gleason score. Then how ISUP labels are assigned for Karolinska.
", "rawMarkdown": "Hi. \nI need to ask something. As in Masks of Karolinska, We have only 0, 1 and 2. Where 2 stands for cancerous part. But it doesn't provide information about Gleason score. Then how ISUP labels are assigned for Karolinska.", "votes": 3, "replies": [ { "id": 847861, "postDate": "2020-05-14T16:26:11.127Z", "content": "Salman it would be nice if you make a discussion topic about your doubt. So that all the competitors can read and share info. Even if the hosts answer here, it would be better if everyone in this competition read it.
", "rawMarkdown": "Salman it would be nice if you make a discussion topic about your doubt. So that all the competitors can read and share info. Even if the hosts answer here, it would be better if everyone in this competition read it." }, { "id": 847869, "postDate": "2020-05-14T16:31:47.453Z", "content": "I've created a topic about this but still no answers.... :(
", "rawMarkdown": "I've created a topic about this but still no answers.... :(", "votes": 1 } ] }, { "id": 820825, "postDate": "2020-04-25T18:29:16.827Z", "content": "Hey @wouterbulten and organizing team. First of all thank you for the dataset and the competition.
\n\nI would like to clarify gleason_score labeling approach.
In Additional Resource page there is a comment:\n\nThe minority pattern must account for at least 5% of the total area to be included\n\nIn Getting started with the PANDA dataset notebook there is a note:
\nNote that, eventhough a biopsy contains cancer, not all epithelial tissue has to be cancerous. Biopsies can contain a mix of cancerous and healthy tissue.\n
But in train['gleason_score'].value_counts() there is no single WSI were non zero Gleason pattern coexists with 0/negative Gleasson pattern: \n\n3+3 2666\n0+0 1925\n3+4 1342\n4+3 1243\n4+4 1126\nnegative 967\n4+5 849\n5+4 248\n5+5 127\n3+5 80\n5+3 43\n
Based on the statement The minority pattern must account for at least 5% of the total area to be included we can conclude that 3+3 gleason_score means there is less than 5% of WSI area with marker different than 3 or there is no such area at all. Can you please explain how should we read the note in notebooks ?
Good question. To understand this, you have to know a bit how the pathologist assigns the label to a slide. The grading system can be quite complex. Shortly described: when a pathologist looks at a biopsy, he/she determines whether some of the tissue is cancerous and, if so, what pattern(s) is/are present. For example, it could be the case that of the epithelial tissue (the glands in the prostate), 20% is healthy, and 80% is cancerous. The 80% should then be classified in the different growth patterns. Some examples:
\n\nNote that in all these cases, there can still be healthy tissue present. The ISUP score 0 in our dataset is only given to biopsies without cancerous tissue.
\n\nThe comment in the notebook is there to make you aware that even though the biopsy-level label is cancerous, not all tissue has to be cancerous. It could even be the case that only a few glands of the biopsy are cancerous. You can take a look at some of the label masks and see that regions of the biopsy will be labelled as benign.
\n\nI hope that answers your question!
", "rawMarkdown": "Good question. To understand this, you have to know a bit how the pathologist assigns the label to a slide. The grading system can be quite complex. Shortly described: when a pathologist looks at a biopsy, he/she determines whether some of the tissue is cancerous and, if so, what pattern(s) is/are present. For example, it could be the case that of the epithelial tissue (the glands in the prostate), 20% is healthy, and 80% is cancerous. The 80% should then be classified in the different growth patterns. Some examples:\n\n- All of the cancerous tissue can be Gleason 3, so the Gleason score would then be 3+3, and the ISUP grade would be 1.\n- 60% of the cancerous tissue could be Gleason 3, and 40% could be 4. This would get a Gleason score of 3+4 and ISUP score 2. \n- In some cases, there are three patterns present. For example, 60% Gleason 3, 30% Gleason 4, and 10% Gleason 5. In those edge cases, the grading system states that the higher Gleason pattern should be used. So, in this case: 3+5 (even though %G5 < %G4) with an ISUP score of 4. This is important because such a minor aggressive component can have a worse prognosis.\n\nNote that in all these cases, there can still be healthy tissue present. The ISUP score 0 in our dataset is only given to biopsies without cancerous tissue.\n\nThe comment in the notebook is there to make you aware that even though the biopsy-level label is cancerous, not all tissue has to be cancerous. It could even be the case that only a few glands of the biopsy are cancerous. You can take a look at some of the label masks and see that regions of the biopsy will be labelled as benign.\n\nI hope that answers your question!", "votes": 6 }, { "id": 821779, "postDate": "2020-04-26T12:17:41.200Z", "content": "Yes, thank you. Perfect explanation.
", "rawMarkdown": "Yes, thank you. Perfect explanation. " }, { "id": 830669, "postDate": "2020-05-02T19:11:47.997Z", "content": "Hello, what is negative means in the gleason_score? Can we treat them as 0's?
", "rawMarkdown": "Hello, what is negative means in the gleason_score? Can we treat them as 0's?" }, { "id": 830808, "postDate": "2020-05-02T23:29:39.877Z", "content": "I think so. Usually, negative means benign.
", "rawMarkdown": "I think so. Usually, negative means benign. " } ] }, { "id": 945044, "postDate": "2020-07-25T14:32:28.953Z", "content": "great work
", "rawMarkdown": "great work", "votes": 1 }, { "id": 924940, "postDate": "2020-07-11T18:06:52.723Z", "content": "In Additional Resources the following is mentioned:
\n\n\n\n\nThe minority pattern must account for at least 5% of the total area to be included, e.g. if pattern 3 is less than 5%, the Gleason score 4 + 3 will instead be 4 + 4. Further, the highest grade should always be part of the score. For example, a biopsy that contains 60% Gleason 4, 37% Gleason 3 and 3% Gleason 5 should get a score of 4 + 5 = 9
\n
So from that, am I supposed to presume the following:
\n\nThe majority pattern has to be included in the grading even if it's < 5%.
The Gleason scores 4+3 and 3+4 have different ISUP grades because of the difference in position of the majority pattern.
If there is only one pattern present in the slide, the ISUP grade has to be calculated by doubling its Gleason score.
Could we use model trained from any public notebook of the competition for example there is one guy who did public his notebook having score of 85.\nWe use of following notebook to create one model .\nhttps://www.kaggle.com/micheomaano/tpu-training-tensorflow-iafoos-method-42x256x256x3/\nusing data from\nhttps://www.kaggle.com/micheomaano/tf-record-256-256-48
\n\nand then ensemble with our models.
", "rawMarkdown": "Could we use model trained from any public notebook of the competition for example there is one guy who did public his notebook having score of 85.\nWe use of following notebook to create one model .\nhttps://www.kaggle.com/micheomaano/tpu-training-tensorflow-iafoos-method-42x256x256x3/\nusing data from\nhttps://www.kaggle.com/micheomaano/tf-record-256-256-48\n\nand then ensemble with our models.", "votes": 1 }, { "id": 825279, "postDate": "2020-04-28T22:36:19.500Z", "content": "Hello,@wouterbulten and organizing team. \n In my country,according to the guidline of prostate cancer, 98% of the cancerous tissue could be Gleason 3, and 2% could be 4, this would get a Gleason score of 3+4 and ISUP score 2.\nIs it same to this competition?
", "rawMarkdown": "Hello,@wouterbulten and organizing team. \n In my country,according to the guidline of prostate cancer, 98% of the cancerous tissue could be Gleason 3, and 2% could be 4, this would get a Gleason score of 3+4 and ISUP score 2.\nIs it same to this competition?", "votes": 2, "replies": [ { "id": 825878, "postDate": "2020-04-29T09:51:27.987Z", "content": "For the test set the pathologists followed the ISUP 2014 guidelines to grade the biopsies.
", "rawMarkdown": "For the test set the pathologists followed the ISUP 2014 guidelines to grade the biopsies.", "votes": 2 }, { "id": 825966, "postDate": "2020-04-29T11:10:17.970Z", "content": "Thank you! I understand.
", "rawMarkdown": "Thank you! I understand." } ] }, { "id": 816037, "postDate": "2020-04-22T04:15:03.603Z", "content": "Thanks for the getting starter notebook.
", "rawMarkdown": "Thanks for the getting starter notebook." }, { "id": 834285, "postDate": "2020-05-05T13:06:18.150Z", "content": "will test images also have mask info ?
", "rawMarkdown": "will test images also have mask info ?" }, { "id": 815754, "postDate": "2020-04-21T20:09:42.863Z", "content": "Thanks the team for the starter and the Dataset.
", "rawMarkdown": "Thanks the team for the starter and the Dataset." }, { "id": 829819, "postDate": "2020-05-02T05:33:53.350Z", "rawMarkdown": "", "votes": -1, "isDeleted": true }, { "id": 816051, "postDate": "2020-04-22T04:33:04.493Z", "content": "Thanks for the notebook
", "rawMarkdown": "Thanks for the notebook", "votes": 1 }, { "id": 836863, "postDate": "2020-05-07T10:01:45.623Z", "content": "Thanks for the notebook.
", "rawMarkdown": "Thanks for the notebook." } ], "comments": [ { "id": 847824, "author_name": "Salman", "author_url": "", "post_date": "2020-05-14T16:11:10.700000", "content": "Hi. \nI need to ask something. As in Masks of Karolinska, We have only 0, 1 and 2. Where 2 stands for cancerous part. But it doesn't provide information about Gleason score. Then how ISUP labels are assigned for Karolinska.
", "votes": 3, "replies": [ { "id": 847861, "author_name": "Marília Prata", "author_url": "", "post_date": "2020-05-14T16:26:11.127000", "content": "Salman it would be nice if you make a discussion topic about your doubt. So that all the competitors can read and share info. Even if the hosts answer here, it would be better if everyone in this competition read it.
", "votes": 0, "replies": [] }, { "id": 847869, "author_name": "Salman", "author_url": "", "post_date": "2020-05-14T16:31:47.453000", "content": "I've created a topic about this but still no answers.... :(
", "votes": 1, "replies": [] } ] }, { "id": 820825, "author_name": "SM", "author_url": "", "post_date": "2020-04-25T18:29:16.827000", "content": "Hey @wouterbulten and organizing team. First of all thank you for the dataset and the competition.
\n\nI would like to clarify gleason_score labeling approach.
In Additional Resource page there is a comment:\n\nThe minority pattern must account for at least 5% of the total area to be included\n\nIn Getting started with the PANDA dataset notebook there is a note:
\nNote that, eventhough a biopsy contains cancer, not all epithelial tissue has to be cancerous. Biopsies can contain a mix of cancerous and healthy tissue.\n
But in train['gleason_score'].value_counts() there is no single WSI were non zero Gleason pattern coexists with 0/negative Gleasson pattern: \n\n3+3 2666\n0+0 1925\n3+4 1342\n4+3 1243\n4+4 1126\nnegative 967\n4+5 849\n5+4 248\n5+5 127\n3+5 80\n5+3 43\n
Based on the statement The minority pattern must account for at least 5% of the total area to be included we can conclude that 3+3 gleason_score means there is less than 5% of WSI area with marker different than 3 or there is no such area at all. Can you please explain how should we read the note in notebooks ?
Good question. To understand this, you have to know a bit how the pathologist assigns the label to a slide. The grading system can be quite complex. Shortly described: when a pathologist looks at a biopsy, he/she determines whether some of the tissue is cancerous and, if so, what pattern(s) is/are present. For example, it could be the case that of the epithelial tissue (the glands in the prostate), 20% is healthy, and 80% is cancerous. The 80% should then be classified in the different growth patterns. Some examples:
\n\nNote that in all these cases, there can still be healthy tissue present. The ISUP score 0 in our dataset is only given to biopsies without cancerous tissue.
\n\nThe comment in the notebook is there to make you aware that even though the biopsy-level label is cancerous, not all tissue has to be cancerous. It could even be the case that only a few glands of the biopsy are cancerous. You can take a look at some of the label masks and see that regions of the biopsy will be labelled as benign.
\n\nI hope that answers your question!
", "votes": 6, "replies": [] }, { "id": 821779, "author_name": "SM", "author_url": "", "post_date": "2020-04-26T12:17:41.200000", "content": "Yes, thank you. Perfect explanation.
", "votes": 0, "replies": [] }, { "id": 830669, "author_name": "Sedat", "author_url": "", "post_date": "2020-05-02T19:11:47.997000", "content": "Hello, what is negative means in the gleason_score? Can we treat them as 0's?
", "votes": 0, "replies": [] }, { "id": 830808, "author_name": "Chage", "author_url": "", "post_date": "2020-05-02T23:29:39.877000", "content": "I think so. Usually, negative means benign.
", "votes": 0, "replies": [] } ] }, { "id": 945044, "author_name": "Meesala sai dhanush", "author_url": "", "post_date": "2020-07-25T14:32:28.953000", "content": "great work
", "votes": 1, "replies": [] }, { "id": 924940, "author_name": "Kishore Badyakar", "author_url": "", "post_date": "2020-07-11T18:06:52.723000", "content": "In Additional Resources the following is mentioned:
\n\n\n\n\nThe minority pattern must account for at least 5% of the total area to be included, e.g. if pattern 3 is less than 5%, the Gleason score 4 + 3 will instead be 4 + 4. Further, the highest grade should always be part of the score. For example, a biopsy that contains 60% Gleason 4, 37% Gleason 3 and 3% Gleason 5 should get a score of 4 + 5 = 9
\n
So from that, am I supposed to presume the following:
\n\nThe majority pattern has to be included in the grading even if it's < 5%.
The Gleason scores 4+3 and 3+4 have different ISUP grades because of the difference in position of the majority pattern.
If there is only one pattern present in the slide, the ISUP grade has to be calculated by doubling its Gleason score.
Could we use model trained from any public notebook of the competition for example there is one guy who did public his notebook having score of 85.\nWe use of following notebook to create one model .\nhttps://www.kaggle.com/micheomaano/tpu-training-tensorflow-iafoos-method-42x256x256x3/\nusing data from\nhttps://www.kaggle.com/micheomaano/tf-record-256-256-48
\n\nand then ensemble with our models.
", "votes": 1, "replies": [] }, { "id": 825279, "author_name": "Chage", "author_url": "", "post_date": "2020-04-28T22:36:19.500000", "content": "Hello,@wouterbulten and organizing team. \n In my country,according to the guidline of prostate cancer, 98% of the cancerous tissue could be Gleason 3, and 2% could be 4, this would get a Gleason score of 3+4 and ISUP score 2.\nIs it same to this competition?
", "votes": 2, "replies": [ { "id": 825878, "author_name": "Wouter Bulten", "author_url": "", "post_date": "2020-04-29T09:51:27.987000", "content": "For the test set the pathologists followed the ISUP 2014 guidelines to grade the biopsies.
", "votes": 2, "replies": [] }, { "id": 825966, "author_name": "Chage", "author_url": "", "post_date": "2020-04-29T11:10:17.970000", "content": "Thank you! I understand.
", "votes": 0, "replies": [] } ] }, { "id": 816037, "author_name": "Prashant Banerjee", "author_url": "", "post_date": "2020-04-22T04:15:03.603000", "content": "Thanks for the getting starter notebook.
", "votes": 0, "replies": [] }, { "id": 834285, "author_name": "Rajnish Chauhan", "author_url": "", "post_date": "2020-05-05T13:06:18.150000", "content": "will test images also have mask info ?
", "votes": 0, "replies": [] }, { "id": 815754, "author_name": "Marília Prata", "author_url": "", "post_date": "2020-04-21T20:09:42.863000", "content": "Thanks the team for the starter and the Dataset.
", "votes": 0, "replies": [] }, { "id": 829819, "author_name": "", "author_url": "", "post_date": "2020-05-02T05:33:53.350000", "content": "", "votes": -1, "replies": [] }, { "id": 816051, "author_name": "Eswar Chand", "author_url": "", "post_date": "2020-04-22T04:33:04.493000", "content": "Thanks for the notebook
", "votes": 1, "replies": [] }, { "id": 836863, "author_name": "cscmedimg", "author_url": "", "post_date": "2020-05-07T10:01:45.623000", "content": "Thanks for the notebook.
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "815581": "Welcome to the **Prostate cANcer graDe Assessment (PANDA)** challenge! Prostate cancer is a leading cause of cancer death. Each year, pathologists assess millions of prostate tissue samples to diagnose cancer and assess its severity. This is not only labor-intensive but it is also a very difficult task for humans. An incorrect diagnosis may lead to suboptimal treatment, which causes unnecessary side effects or shortened life expectancy of the patient. In this competition, we need your help to improve prostate cancer diagnostics to the benefit of millions of men worldwide.\n\nTo get you up to speed with the dataset of the competition, we have made a [getting started notebook](https://www.kaggle.com/wouterbulten/getting-started-with-the-panda-dataset). All images in the dataset are whole-slide images (WSI) and the notebook contains some starting points on using these types of images.\n\nDuring the competition, we will be following the forums and are available to answer any questions you might have. Feel free to post questions regarding the data or the (medical) background, we are here to help!\n\nThis competition is also part of a scientific conference and we will write a paper on the results, see the page on the [challenge workshop](https://www.kaggle.com/c/prostate-cancer-grade-assessment/overview/miccai-2020) for more information.\n\nWe would like to extend our gratitude to the pathologists who graded the test sets used in the competition: Lars Egevad, Brett Delahunt, Hemamali Samaratunga, Toyonori Tsuzuki, Christina Hulsbergen-van de Kaa, Robert Vink, and Hester van Boven.\n\nWe are looking forward to seeing your methods for solving this important problem! Good luck with the challenge!\n\nOn behalf of the whole organizing team,\n\n[Peter](https://www.kaggle.com/petstr), [Kimmo](https://www.kaggle.com/kimmokartasalo), [Martin](https://www.kaggle.com/martineklund), [Geert](https://www.kaggle.com/geertlitjens) & [Wouter](https://www.kaggle.com/wouterbulten)\n", "847824": "Hi. \nI need to ask something. As in Masks of Karolinska, We have only 0, 1 and 2. Where 2 stands for cancerous part. But it doesn't provide information about Gleason score. Then how ISUP labels are assigned for Karolinska.", "820825": "Hey @wouterbulten and organizing team. First of all thank you for the dataset and the competition. \n\nI would like to clarify `gleason_score` labeling approach. \n\nIn `Additional Resource` page there is a comment:\n```\nThe minority pattern must account for at least 5% of the total area to be included\n```\nIn `Getting started with the PANDA dataset` notebook there is a note:\n\n```\nNote that, eventhough a biopsy contains cancer, not all epithelial tissue has to be cancerous. Biopsies can contain a mix of cancerous and healthy tissue.\n```\n\nBut in `train['gleason_score'].value_counts()` there is no single WSI were non zero Gleason pattern coexists with 0/negative Gleasson pattern: \n```\n3+3 2666\n0+0 1925\n3+4 1342\n4+3 1243\n4+4 1126\nnegative 967\n4+5 849\n5+4 248\n5+5 127\n3+5 80\n5+3 43\n```\n\nBased on the statement `The minority pattern must account for at least 5% of the total area to be included` we can conclude that `3+3` gleason_score means there is less than 5% of WSI area with marker different than `3` or there is no such area at all. Can you please explain how should we read the note in notebooks ? \n\n", "945044": "great work", "924940": "In [Additional Resources](https://www.kaggle.com/c/prostate-cancer-grade-assessment/overview/additional-resources) the following is mentioned:\n\n> The minority pattern must account for at least 5% of the total area to be included, e.g. if pattern 3 is less than 5%, the Gleason score 4 + 3 will instead be 4 + 4. Further, the highest grade should always be part of the score. For example, a biopsy that contains 60% Gleason 4, 37% Gleason 3 and 3% Gleason 5 should get a score of 4 + 5 = 9\n\nSo from that, am I supposed to presume the following:\n\n1. The majority pattern has to be included in the grading even if it's `< 5%`.\n\n2. The Gleason scores 4+3 and 3+4 have different ISUP grades because of the difference in position of the majority pattern.\n\n3. If there is only one pattern present in the slide, the ISUP grade has to be calculated by doubling its Gleason score.\n\n", "913157": "Could we use model trained from any public notebook of the competition for example there is one guy who did public his notebook having score of 85.\nWe use of following notebook to create one model .\nhttps://www.kaggle.com/micheomaano/tpu-training-tensorflow-iafoos-method-42x256x256x3/\nusing data from\nhttps://www.kaggle.com/micheomaano/tf-record-256-256-48\n\nand then ensemble with our models.", "825279": "Hello,@wouterbulten and organizing team. \n In my country,according to the guidline of prostate cancer, 98% of the cancerous tissue could be Gleason 3, and 2% could be 4, this would get a Gleason score of 3+4 and ISUP score 2.\nIs it same to this competition?", "816037": "Thanks for the getting starter notebook.", "834285": "will test images also have mask info ?", "815754": "Thanks the team for the starter and the Dataset.", "829819": "", "816051": "Thanks for the notebook", "836863": "Thanks for the notebook." } }