Greetings from the Center for Augmented Intelligence in Imaging (CAII) of the Department of Radiology, Mayo Clinic, Florida
\nWe welcome you to participate in the “Stroke Thromboembolism Registery of Imaging and Pathology (STRIP) AI Challenge”!
\nEach year in the United States, over 700,000 individuals experience an ischemic stroke caused by a blood clot blocking an artery to the brain. A second stroke (23% of total events are recurrent) worsens the chances of the patient’s survival. However, subsequent strokes may be mitigated if the source of the thrombotic material can be determined at the time of the first event, thereby directing preventive management of future clot formation and its embolization to brain arteries.
\nThis competition is about predicting the origins of blood clot resulting in ischemic stroke using Digital Pathology Slides (aka Whole Slide Images (WSI)) of the thrombotic material extracted mechanically during acute neurovascular procedures. Collaborative efforts across 18 institutions, led by Waleed Brinjikji, MD of the Mayo Clinic Rochester allowed the compilation of this unique dataset (https://pubmed.ncbi.nlm.nih.gov/33722963/).
\nUnlike traditional images, WSIs present many additional challenges due to their high resolution and size. In addition, based on the source MSI scanner hardware, they may be saved in different file formats (e.g., .svs, .ndpi, etc. ). Hence, in order to ease some of the potential pitfalls, we have converted and standardized all collected data into the TIFF format.
\nYou could potentially help many patients by developing and sharing your classification of blood clot origins. We look forward to interacting with you towards this goal.
\nWe hope you will enjoy participating in this competition!
\nCAII Team, Mayo Clinic Florida
\nHosts:
\nMayo Clinic
\nBarbaros Selnur Erdal (@Barbaros), Associate Professor of Radiology, Department of Radiology, Mayo Clinic Florida
\nMohammed Sobhi Jabal (@Sobhi Jabal), Research Fellow, Department of Radiology, Mayo Clinic Rochester
\n*Special thanks to Dr. Waleed Brinjikji, Dr. David F. Kallmes, Dr. Ramanathan Kadirvel, and Dr. Richard D. White.
Thank you to our collaborators at Kaggle
\nAshley (@Ashley Chow)
\nRyan (@Ryan Holbrook)
Greetings from the Center for Augmented Intelligence in Imaging (CAII) of the Department of Radiology, Mayo Clinic, Florida
\nWe welcome you to participate in the “Stroke Thromboembolism Registery of Imaging and Pathology (STRIP) AI Challenge”!
\nEach year in the United States, over 700,000 individuals experience an ischemic stroke caused by a blood clot blocking an artery to the brain. A second stroke (23% of total events are recurrent) worsens the chances of the patient’s survival. However, subsequent strokes may be mitigated if the source of the thrombotic material can be determined at the time of the first event, thereby directing preventive management of future clot formation and its embolization to brain arteries.
\nThis competition is about predicting the origins of blood clot resulting in ischemic stroke using Digital Pathology Slides (aka Whole Slide Images (WSI)) of the thrombotic material extracted mechanically during acute neurovascular procedures. Collaborative efforts across 18 institutions, led by Waleed Brinjikji, MD of the Mayo Clinic Rochester allowed the compilation of this unique dataset (https://pubmed.ncbi.nlm.nih.gov/33722963/).
\nUnlike traditional images, WSIs present many additional challenges due to their high resolution and size. In addition, based on the source MSI scanner hardware, they may be saved in different file formats (e.g., .svs, .ndpi, etc. ). Hence, in order to ease some of the potential pitfalls, we have converted and standardized all collected data into the TIFF format.
\nYou could potentially help many patients by developing and sharing your classification of blood clot origins. We look forward to interacting with you towards this goal.
\nWe hope you will enjoy participating in this competition!
\nCAII Team, Mayo Clinic Florida
\nHosts:
\nMayo Clinic
\nBarbaros Selnur Erdal (@Barbaros), Associate Professor of Radiology, Department of Radiology, Mayo Clinic Florida
\nMohammed Sobhi Jabal (@Sobhi Jabal), Research Fellow, Department of Radiology, Mayo Clinic Rochester
\n*Special thanks to Dr. Waleed Brinjikji, Dr. David F. Kallmes, Dr. Ramanathan Kadirvel, and Dr. Richard D. White.
Thank you to our collaborators at Kaggle
\nAshley (@Ashley Chow)
\nRyan (@Ryan Holbrook)
Thanks for participating!
", "rawMarkdown": "Thanks for participating!", "votes": 5 }, { "id": 1990878, "postDate": "2022-10-16T19:15:34.973Z", "content": "I also tried this.
", "rawMarkdown": "I also tried this. \n", "votes": 1 }, { "id": 1848793, "postDate": "2022-07-08T22:36:36.177Z", "content": "Hi @barbaroserdal -
\nThis may be a dumb question, but can doctors do this task?
\nIn other words, are there physiological indications of stroke etiology/origin?
Are there papers that discuss this that you can link to?
", "rawMarkdown": "Hi @barbaroserdal - \n\nThis may be a dumb question, but can doctors do this task? \nIn other words, are there physiological indications of stroke etiology/origin?\n\nAre there papers that discuss this that you can link to?", "votes": 3, "replies": [ { "id": 1851901, "postDate": "2022-07-11T16:08:53.407Z", "content": "Hi Darien,
\nYes, it would be challenging for a pathologist. But algorithmic learning is possible.
", "rawMarkdown": "Hi Darien,\n\nYes, it would be challenging for a pathologist. But algorithmic learning is possible.", "votes": 4 } ] }, { "id": 1931550, "postDate": "2022-09-08T18:56:52.387Z", "content": "Just had done another medical competition. So really glad to be able to work on another one :)
", "rawMarkdown": "Just had done another medical competition. So really glad to be able to work on another one :)", "votes": 1 }, { "id": 1859152, "postDate": "2022-07-17T12:36:24.400Z", "content": "Hi I am a doctor and I want to join some group here where we can learn to build the algorithm. I can be of any medical related help if required. I will be happy if any group of expert data scientists can chip me in. Many thanks
", "rawMarkdown": "Hi I am a doctor and I want to join some group here where we can learn to build the algorithm. I can be of any medical related help if required. I will be happy if any group of expert data scientists can chip me in. Many thanks ", "votes": 1, "replies": [ { "id": 1865116, "postDate": "2022-07-21T14:35:02.817Z", "content": "Hi Ayesha, if you are still looking for a group then I would be happy to form a team. Please PM me if you are interested.
", "rawMarkdown": "Hi Ayesha, if you are still looking for a group then I would be happy to form a team. Please PM me if you are interested." } ] }, { "id": 1848246, "postDate": "2022-07-08T13:22:20.683Z", "content": "Thanks for setting up this competition! I have a few queries on the background information on the dataset e.g. IRB status. Is there a contact email address where I can send these queries? @barbaroserdal maybe? (I'm unable to message you here on Kaggle unfortunately)
", "rawMarkdown": "Thanks for setting up this competition! I have a few queries on the background information on the dataset e.g. IRB status. Is there a contact email address where I can send these queries? @barbaroserdal maybe? (I'm unable to message you here on Kaggle unfortunately)", "votes": 1, "replies": [ { "id": 1851898, "postDate": "2022-07-11T16:03:18.623Z", "content": "Hi Anton,
\nThe data was collected with IRBs in place (across multiple institutions). Following the completion of the competition, all metadata will be publicly disseminated (it was collected as part of an NIH Study). We are in the data dissemination stage.
\nThanks,
\nBarbaros
Hi, I am new to Kaggle. I have access to a GPU. Is it allowed in the rules to download the data and train the model in my computer (to see how the model performs) and then upload the notebook to do the same thing within Kaggle? This is because GPU resurces in Kaggle are limited.
", "rawMarkdown": "Hi, I am new to Kaggle. I have access to a GPU. Is it allowed in the rules to download the data and train the model in my computer (to see how the model performs) and then upload the notebook to do the same thing within Kaggle? This is because GPU resurces in Kaggle are limited.", "votes": 1, "replies": [ { "id": 1848270, "postDate": "2022-07-08T13:37:55.157Z", "content": "Hi @swallowroot ,
\nYou are allowed to download the data and train on your own computer. Note however that this is a code competition, which means your submission will be a notebook that produces predictions on a hidden test set. (Your submission isn't just a file of predictions, in other words.)
\nA common strategy is to upload a trained model as a dataset and attach that dataset to your submission notebook so you don't have to retrain every time you submit. See the pairs of Training / Inference notebooks in the Code tab from this or previous competitions for examples.
\nHope this helps!
", "rawMarkdown": "Hi @swallowroot ,\n\nYou are allowed to download the data and train on your own computer. Note however that this is a code competition, which means your submission will be a notebook that produces predictions on a hidden test set. (Your submission isn't just a file of predictions, in other words.)\n\nA common strategy is to upload a trained model as a dataset and attach that dataset to your submission notebook so you don't have to retrain every time you submit. See the pairs of Training / Inference notebooks in the Code tab from this or previous competitions for examples.\n\nHope this helps!", "votes": 7 }, { "id": 1848301, "postDate": "2022-07-08T14:00:35.373Z", "content": "Thanks @ryanholbrook that's helpful!
", "rawMarkdown": "Thanks @ryanholbrook that's helpful!" } ] }, { "id": 1889065, "postDate": "2022-08-08T00:50:46.737Z", "content": "Frankly, a poor effort at the \"converted and standardized\" all data into tiff format. It is unmanageable data at the size being 300GB+. I and im sure many other google colab users cannot even download it. You should've offered a subset of some lower res slides - contrary to popular beflied deep learning does not always require a lot of data to get decent results, especially with pretrained models. Poorly managed competition.
", "rawMarkdown": "Frankly, a poor effort at the \"converted and standardized\" all data into tiff format. It is unmanageable data at the size being 300GB+. I and im sure many other google colab users cannot even download it. You should've offered a subset of some lower res slides - contrary to popular beflied deep learning does not always require a lot of data to get decent results, especially with pretrained models. Poorly managed competition." }, { "id": 1868700, "postDate": "2022-07-24T06:45:47.433Z", "content": "Hello @barbaroserdal,
\nThank you for organizing this competition, I am new to Kaggle and already learning so much! I have a question on the data set and how these tissue slides where stained originally. I have read that using Hematoxylin and eosin (H&E) for staining is the most common way to stain tissues. Were these tissues treated with H & E for staining? across different centers?
Thank you much!
", "rawMarkdown": "Hello @barbaroserdal,\nThank you for organizing this competition, I am new to Kaggle and already learning so much! I have a question on the data set and how these tissue slides where stained originally. I have read that using Hematoxylin and eosin (H&E) for staining is the most common way to stain tissues. Were these tissues treated with H & E for staining? across different centers?\n\nThank you much!", "votes": 2, "replies": [ { "id": 1933234, "postDate": "2022-09-10T11:34:57.473Z", "content": "Any answer on this? It is sort of important.
", "rawMarkdown": "Any answer on this? It is sort of important." }, { "id": 1950738, "postDate": "2022-09-22T14:06:27.187Z", "content": "This does not seem H&E. If you look at the paper mentioned in the first post, it's MSB.
", "rawMarkdown": "This does not seem H&E. If you look at the paper mentioned in the first post, it's MSB.", "votes": 1 }, { "id": 1950773, "postDate": "2022-09-22T14:24:36.280Z", "content": "@vdellamea thank you, I somehow missed that link
", "rawMarkdown": "@vdellamea thank you, I somehow missed that link" } ] }, { "id": 1861343, "postDate": "2022-07-19T01:27:09.190Z", "content": "Hey thanks for putting up this competition.
\nI have a question, regarding the image generation in its different centers.
\nAs per I understood, the images were generated in eleven different centers (with different file formats, different equipments, models, vendors, parameters, and so on.. ).
\nIt would be very important to understand a bit of how the images were generated, and to figure out how they differ from one center to another. This has a huge impact on the model structure and strategy. For instance, centers 11, 4 and 7 account for more than 60% of the images generated, thus, this could create a bias on the model, if such particularities are not tackled properly.
I have posted this question in the forum, too, please take a look.
\n https://www.kaggle.com/competitions/mayo-clinic-strip-ai/discussion/337811
Hope to hear from you.
\nBest regards,
\nFelipe
Hi @barbaroserdal can people working at Mayo take part?
", "rawMarkdown": "Hi @barbaroserdal can people working at Mayo take part?", "votes": 2, "replies": [ { "id": 1857166, "postDate": "2022-07-15T23:11:17.300Z", "content": "Hi @habsid, thanks for your question! Just chiming in here - as per the Competition Rules under the \"2. ELIGIBILITY\":
\n\"B. Unless otherwise stated in the Specific Competition Rules above or prohibited by internal policies of the Competition Entities, employees, interns, contractors, officers and directors of Competition Entities may enter and participate in the Competition, but are not eligible to win any Prizes.\"Competition Entities\" means the Competition Sponsor, Kaggle Inc., and their respective parent companies, subsidiaries and affiliates. If you are such a participant from a Competition Entity, you are subject to all applicable internal policies of your employer with respect to your participation.\"
Hope it helps answer. Thanks so much!
", "rawMarkdown": "Hi @habsid, thanks for your question! Just chiming in here - as per the Competition Rules under the \"2. ELIGIBILITY\":\n\"B. Unless otherwise stated in the Specific Competition Rules above or prohibited by internal policies of the Competition Entities, **employees, interns, contractors, officers and directors of Competition Entities may enter and participate in the Competition, but are not eligible to win any Prizes.**\"Competition Entities\" means the Competition Sponsor, Kaggle Inc., and their respective parent companies, subsidiaries and affiliates. If you are such a participant from a Competition Entity, you are subject to all applicable internal policies of your employer with respect to your participation.\"\n\nHope it helps answer. Thanks so much!", "votes": 1 } ] }, { "id": 1847108, "postDate": "2022-07-07T16:32:55.487Z", "content": "Thanks for organising this competition. Hope to learn a lot!
", "rawMarkdown": "Thanks for organising this competition. Hope to learn a lot!", "votes": 2 }, { "id": 1865564, "postDate": "2022-07-22T00:51:19.090Z", "content": "Mr. Barbaros, of the 700,000 individuals that experience ischemic Stroke, how many of them are females and Males? From my observations more attention is giving to males than females when it comes to stroke awareness. The actually cause of a stoke cannot be determined. This is what is known as a \"cryptogenic\" stroke or a stroke of unknown cause. It's estimated that 25-30% of ischemic strokes are cryptogenic or unexpected. (Medtronic)
\nSince strokes are unexpected lets folks pray to God for good health, eat healthily, and exercise once or twice a week. My advice is; for males one pushup a day, and eat one banana or a grip fruit, for females 15 munities walk in the park or 15 jumping jack will do.
The best solution to preventing Stroke caused by a Blood clot is to consume about three Grapefruit a day, Florida is a Tropical State a Grapefruit plantation should be planted to address this Stroke issue.
", "rawMarkdown": "The best solution to preventing Stroke caused by a Blood clot is to consume about three Grapefruit a day, Florida is a Tropical State a Grapefruit plantation should be planted to address this Stroke issue. ", "votes": -3 }, { "id": 2001058, "postDate": "2022-10-23T18:32:30.793Z", "content": "Still fighting ………..!
", "rawMarkdown": "Still fighting ...........!" }, { "id": 1979967, "postDate": "2022-10-09T21:31:17.570Z", "content": "Hi Barbaros, if my late submission score which could have been a second best score on the leadeboard. Will it contribute towards the goal of this competition in any way?
", "rawMarkdown": "Hi Barbaros, if my late submission score which could have been a second best score on the leadeboard. Will it contribute towards the goal of this competition in any way?" }, { "id": 1959330, "postDate": "2022-09-28T05:04:56.667Z", "content": "Hi everyone.. I was just going through the data and wanted to know, how everyone processed and created model for this large amount of data just in case you don't have a personal GPU with you.
", "rawMarkdown": "Hi everyone.. I was just going through the data and wanted to know, how everyone processed and created model for this large amount of data just in case you don't have a personal GPU with you.", "replies": [ { "id": 1960614, "postDate": "2022-09-28T16:52:25.557Z", "content": "As far as I have understood there are two ways for reducing computational complexity caused by dataset size:
\n1) Downscale whole slide images - which will likely result in loss of training signal but you can build and test end to end models faster.
\n2) Tile creation and selection - in this method you create tiles from images and have criteria for selecting tiles with maximum signal, and then train an MIL based method which is more complicated but may do better (or not, but there's a better chance). Tile creation and selection is key, this will reduce your training time based on pre-selected best guess part of the data.
I have only tried method two and struggled throughout - so that you know how much salt to take these statements with :)
\nAll the best.
", "rawMarkdown": "As far as I have understood there are two ways for reducing computational complexity caused by dataset size:\n\n1) Downscale whole slide images - which will likely result in loss of training signal but you can build and test end to end models faster.\n2) Tile creation and selection - in this method you create tiles from images and have criteria for selecting tiles with maximum signal, and then train an MIL based method which is more complicated but may do better (or not, but there's a better chance). Tile creation and selection is key, this will reduce your training time based on pre-selected best guess part of the data. \n\nI have only tried method two and struggled throughout - so that you know how much salt to take these statements with :)\n\nAll the best.", "votes": 2 }, { "id": 1961210, "postDate": "2022-09-29T04:40:59.490Z", "content": "Thanks for the detailed explanation.. I will surely experiment with tile creation and selection.
", "rawMarkdown": "Thanks for the detailed explanation.. I will surely experiment with tile creation and selection.", "votes": 1 } ] }, { "id": 1951816, "postDate": "2022-09-23T09:04:21.520Z", "content": "From the paper, it seems all slides were acquired with the same scanner, however the magnification is not specified: was it 20x or 40x? The TIFF files have an improper DPI.
", "rawMarkdown": "From the paper, it seems all slides were acquired with the same scanner, however the magnification is not specified: was it 20x or 40x? The TIFF files have an improper DPI. " }, { "id": 1936821, "postDate": "2022-09-13T05:06:44.240Z", "content": "Hello @barbaroserdal ,
\nLet me ask a question about private test data.
\nIs there only \"CE\" and \"LAA\" labels for private test data? In other words, would there be no labels like \"other\" or \"unknown\"?
", "rawMarkdown": "Hello @barbaroserdal ,\n\nLet me ask a question about private test data.\n\nIs there only \"CE\" and \"LAA\" labels for private test data? In other words, would there be no labels like \"other\" or \"unknown\"?" }, { "id": 1928042, "postDate": "2022-09-06T07:41:43.267Z", "content": "Hi guys
\nI released a dataset that contains resized images from this competition(5GB) . Check it out link
Thank you for your organising. This competition is so interesting and I enjoy it !
", "rawMarkdown": "Thank you for your organising. This competition is so interesting and I enjoy it ! " }, { "id": 1872069, "postDate": "2022-07-26T16:45:00.097Z", "content": "I had some difficulties making submissions,
\ncan someone please check and lemme know?
\nhttps://www.kaggle.com/competitions/mayo-clinic-strip-ai/discussion/339590
Thank you
", "rawMarkdown": "I had some difficulties making **submissions**, \ncan someone please check and lemme know?\nhttps://www.kaggle.com/competitions/mayo-clinic-strip-ai/discussion/339590\n\nThank you" }, { "id": 1860512, "postDate": "2022-07-18T11:17:08.713Z", "content": "I am really going to enjoy this competition. I hope to learn a lot and make a good solution. Thank you for organising!
", "rawMarkdown": "I am really going to enjoy this competition. I hope to learn a lot and make a good solution. Thank you for organising!" }, { "id": 1869158, "postDate": "2022-07-24T14:36:10.173Z", "rawMarkdown": "", "isDeleted": true } ], "comments": [ { "id": 1847228, "author_name": "Barbaros", "author_url": "", "post_date": "2022-07-07T18:16:41.167000", "content": "Thanks for participating!
", "votes": 5, "replies": [] }, { "id": 1990878, "author_name": "Dr. Asif Khan", "author_url": "", "post_date": "2022-10-16T19:15:34.973000", "content": "I also tried this.
", "votes": 1, "replies": [] }, { "id": 1848793, "author_name": "Darien Schettler", "author_url": "", "post_date": "2022-07-08T22:36:36.177000", "content": "Hi @barbaroserdal -
\nThis may be a dumb question, but can doctors do this task?
\nIn other words, are there physiological indications of stroke etiology/origin?
Are there papers that discuss this that you can link to?
", "votes": 3, "replies": [ { "id": 1851901, "author_name": "Barbaros", "author_url": "", "post_date": "2022-07-11T16:08:53.407000", "content": "Hi Darien,
\nYes, it would be challenging for a pathologist. But algorithmic learning is possible.
", "votes": 4, "replies": [] } ] }, { "id": 1931550, "author_name": "Bilal Aamer", "author_url": "", "post_date": "2022-09-08T18:56:52.387000", "content": "Just had done another medical competition. So really glad to be able to work on another one :)
", "votes": 1, "replies": [] }, { "id": 1859152, "author_name": "ayesha abbasi", "author_url": "", "post_date": "2022-07-17T12:36:24.400000", "content": "Hi I am a doctor and I want to join some group here where we can learn to build the algorithm. I can be of any medical related help if required. I will be happy if any group of expert data scientists can chip me in. Many thanks
", "votes": 1, "replies": [ { "id": 1865116, "author_name": "prasad", "author_url": "", "post_date": "2022-07-21T14:35:02.817000", "content": "Hi Ayesha, if you are still looking for a group then I would be happy to form a team. Please PM me if you are interested.
", "votes": 0, "replies": [] } ] }, { "id": 1848246, "author_name": "Anton Schwaighofer", "author_url": "", "post_date": "2022-07-08T13:22:20.683000", "content": "Thanks for setting up this competition! I have a few queries on the background information on the dataset e.g. IRB status. Is there a contact email address where I can send these queries? @barbaroserdal maybe? (I'm unable to message you here on Kaggle unfortunately)
", "votes": 1, "replies": [ { "id": 1851898, "author_name": "Barbaros", "author_url": "", "post_date": "2022-07-11T16:03:18.623000", "content": "Hi Anton,
\nThe data was collected with IRBs in place (across multiple institutions). Following the completion of the competition, all metadata will be publicly disseminated (it was collected as part of an NIH Study). We are in the data dissemination stage.
\nThanks,
\nBarbaros
Hi, I am new to Kaggle. I have access to a GPU. Is it allowed in the rules to download the data and train the model in my computer (to see how the model performs) and then upload the notebook to do the same thing within Kaggle? This is because GPU resurces in Kaggle are limited.
", "votes": 1, "replies": [ { "id": 1848270, "author_name": "Ryan Holbrook", "author_url": "", "post_date": "2022-07-08T13:37:55.157000", "content": "Hi @swallowroot ,
\nYou are allowed to download the data and train on your own computer. Note however that this is a code competition, which means your submission will be a notebook that produces predictions on a hidden test set. (Your submission isn't just a file of predictions, in other words.)
\nA common strategy is to upload a trained model as a dataset and attach that dataset to your submission notebook so you don't have to retrain every time you submit. See the pairs of Training / Inference notebooks in the Code tab from this or previous competitions for examples.
\nHope this helps!
", "votes": 7, "replies": [] }, { "id": 1848301, "author_name": "swallowroot", "author_url": "", "post_date": "2022-07-08T14:00:35.373000", "content": "Thanks @ryanholbrook that's helpful!
", "votes": 0, "replies": [] } ] }, { "id": 1889065, "author_name": "ankit", "author_url": "", "post_date": "2022-08-08T00:50:46.737000", "content": "Frankly, a poor effort at the \"converted and standardized\" all data into tiff format. It is unmanageable data at the size being 300GB+. I and im sure many other google colab users cannot even download it. You should've offered a subset of some lower res slides - contrary to popular beflied deep learning does not always require a lot of data to get decent results, especially with pretrained models. Poorly managed competition.
", "votes": 0, "replies": [] }, { "id": 1868700, "author_name": "aldop_7895t", "author_url": "", "post_date": "2022-07-24T06:45:47.433000", "content": "Hello @barbaroserdal,
\nThank you for organizing this competition, I am new to Kaggle and already learning so much! I have a question on the data set and how these tissue slides where stained originally. I have read that using Hematoxylin and eosin (H&E) for staining is the most common way to stain tissues. Were these tissues treated with H & E for staining? across different centers?
Thank you much!
", "votes": 2, "replies": [ { "id": 1933234, "author_name": "cosmosaa", "author_url": "", "post_date": "2022-09-10T11:34:57.473000", "content": "Any answer on this? It is sort of important.
", "votes": 0, "replies": [] }, { "id": 1950738, "author_name": "MITEL-UNIUD", "author_url": "", "post_date": "2022-09-22T14:06:27.187000", "content": "This does not seem H&E. If you look at the paper mentioned in the first post, it's MSB.
", "votes": 1, "replies": [] }, { "id": 1950773, "author_name": "cosmosaa", "author_url": "", "post_date": "2022-09-22T14:24:36.280000", "content": "@vdellamea thank you, I somehow missed that link
", "votes": 0, "replies": [] } ] }, { "id": 1861343, "author_name": "Felipe Dutra", "author_url": "", "post_date": "2022-07-19T01:27:09.190000", "content": "Hey thanks for putting up this competition.
\nI have a question, regarding the image generation in its different centers.
\nAs per I understood, the images were generated in eleven different centers (with different file formats, different equipments, models, vendors, parameters, and so on.. ).
\nIt would be very important to understand a bit of how the images were generated, and to figure out how they differ from one center to another. This has a huge impact on the model structure and strategy. For instance, centers 11, 4 and 7 account for more than 60% of the images generated, thus, this could create a bias on the model, if such particularities are not tackled properly.
I have posted this question in the forum, too, please take a look.
\n https://www.kaggle.com/competitions/mayo-clinic-strip-ai/discussion/337811
Hope to hear from you.
\nBest regards,
\nFelipe
Hi @barbaroserdal can people working at Mayo take part?
", "votes": 2, "replies": [ { "id": 1857166, "author_name": "Ashley Chow", "author_url": "", "post_date": "2022-07-15T23:11:17.300000", "content": "Hi @habsid, thanks for your question! Just chiming in here - as per the Competition Rules under the \"2. ELIGIBILITY\":
\n\"B. Unless otherwise stated in the Specific Competition Rules above or prohibited by internal policies of the Competition Entities, employees, interns, contractors, officers and directors of Competition Entities may enter and participate in the Competition, but are not eligible to win any Prizes.\"Competition Entities\" means the Competition Sponsor, Kaggle Inc., and their respective parent companies, subsidiaries and affiliates. If you are such a participant from a Competition Entity, you are subject to all applicable internal policies of your employer with respect to your participation.\"
Hope it helps answer. Thanks so much!
", "votes": 1, "replies": [] } ] }, { "id": 1847108, "author_name": "KH_ARJUN", "author_url": "", "post_date": "2022-07-07T16:32:55.487000", "content": "Thanks for organising this competition. Hope to learn a lot!
", "votes": 2, "replies": [] }, { "id": 1865564, "author_name": "Jonathan Savice", "author_url": "", "post_date": "2022-07-22T00:51:19.090000", "content": "Mr. Barbaros, of the 700,000 individuals that experience ischemic Stroke, how many of them are females and Males? From my observations more attention is giving to males than females when it comes to stroke awareness. The actually cause of a stoke cannot be determined. This is what is known as a \"cryptogenic\" stroke or a stroke of unknown cause. It's estimated that 25-30% of ischemic strokes are cryptogenic or unexpected. (Medtronic)
\nSince strokes are unexpected lets folks pray to God for good health, eat healthily, and exercise once or twice a week. My advice is; for males one pushup a day, and eat one banana or a grip fruit, for females 15 munities walk in the park or 15 jumping jack will do.
The best solution to preventing Stroke caused by a Blood clot is to consume about three Grapefruit a day, Florida is a Tropical State a Grapefruit plantation should be planted to address this Stroke issue.
", "votes": -3, "replies": [] }, { "id": 2001058, "author_name": "Dr. Asif Khan", "author_url": "", "post_date": "2022-10-23T18:32:30.793000", "content": "Still fighting ………..!
", "votes": 0, "replies": [] }, { "id": 1979967, "author_name": "Vijay Jadhav", "author_url": "", "post_date": "2022-10-09T21:31:17.570000", "content": "Hi Barbaros, if my late submission score which could have been a second best score on the leadeboard. Will it contribute towards the goal of this competition in any way?
", "votes": 0, "replies": [] }, { "id": 1959330, "author_name": "Ab_Rafey", "author_url": "", "post_date": "2022-09-28T05:04:56.667000", "content": "Hi everyone.. I was just going through the data and wanted to know, how everyone processed and created model for this large amount of data just in case you don't have a personal GPU with you.
", "votes": 0, "replies": [ { "id": 1960614, "author_name": "tdiceman", "author_url": "", "post_date": "2022-09-28T16:52:25.557000", "content": "As far as I have understood there are two ways for reducing computational complexity caused by dataset size:
\n1) Downscale whole slide images - which will likely result in loss of training signal but you can build and test end to end models faster.
\n2) Tile creation and selection - in this method you create tiles from images and have criteria for selecting tiles with maximum signal, and then train an MIL based method which is more complicated but may do better (or not, but there's a better chance). Tile creation and selection is key, this will reduce your training time based on pre-selected best guess part of the data.
I have only tried method two and struggled throughout - so that you know how much salt to take these statements with :)
\nAll the best.
", "votes": 2, "replies": [] }, { "id": 1961210, "author_name": "Ab_Rafey", "author_url": "", "post_date": "2022-09-29T04:40:59.490000", "content": "Thanks for the detailed explanation.. I will surely experiment with tile creation and selection.
", "votes": 1, "replies": [] } ] }, { "id": 1951816, "author_name": "MITEL-UNIUD", "author_url": "", "post_date": "2022-09-23T09:04:21.520000", "content": "From the paper, it seems all slides were acquired with the same scanner, however the magnification is not specified: was it 20x or 40x? The TIFF files have an improper DPI.
", "votes": 0, "replies": [] }, { "id": 1936821, "author_name": "Koyo horiuchi", "author_url": "", "post_date": "2022-09-13T05:06:44.240000", "content": "Hello @barbaroserdal ,
\nLet me ask a question about private test data.
\nIs there only \"CE\" and \"LAA\" labels for private test data? In other words, would there be no labels like \"other\" or \"unknown\"?
", "votes": 0, "replies": [] }, { "id": 1928042, "author_name": "Arian Ghasemi", "author_url": "", "post_date": "2022-09-06T07:41:43.267000", "content": "Hi guys
\nI released a dataset that contains resized images from this competition(5GB) . Check it out link
Thank you for your organising. This competition is so interesting and I enjoy it !
", "votes": 0, "replies": [] }, { "id": 1872069, "author_name": "Pathik Ghugare", "author_url": "", "post_date": "2022-07-26T16:45:00.097000", "content": "I had some difficulties making submissions,
\ncan someone please check and lemme know?
\nhttps://www.kaggle.com/competitions/mayo-clinic-strip-ai/discussion/339590
Thank you
", "votes": 0, "replies": [] }, { "id": 1860512, "author_name": "ArturRa", "author_url": "", "post_date": "2022-07-18T11:17:08.713000", "content": "I am really going to enjoy this competition. I hope to learn a lot and make a good solution. Thank you for organising!
", "votes": 0, "replies": [] }, { "id": 1869158, "author_name": "", "author_url": "", "post_date": "2022-07-24T14:36:10.173000", "content": "", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "1846085": "Greetings from the Center for Augmented Intelligence in Imaging (CAII) of the Department of Radiology, Mayo Clinic, Florida\n\nWe welcome you to participate in the “Stroke Thromboembolism Registery of Imaging and Pathology (STRIP) AI Challenge”!\n\nEach year in the United States, over 700,000 individuals experience an ischemic stroke caused by a blood clot blocking an artery to the brain. A second stroke (23% of total events are recurrent) worsens the chances of the patient’s survival. However, subsequent strokes may be mitigated if the source of the thrombotic material can be determined at the time of the first event, thereby directing preventive management of future clot formation and its embolization to brain arteries. \n\nThis competition is about predicting the origins of blood clot resulting in ischemic stroke using Digital Pathology Slides (aka Whole Slide Images (WSI)) of the thrombotic material extracted mechanically during acute neurovascular procedures. Collaborative efforts across 18 institutions, led by Waleed Brinjikji, MD of the Mayo Clinic Rochester allowed the compilation of this unique dataset (https://pubmed.ncbi.nlm.nih.gov/33722963/). \n \nUnlike traditional images, WSIs present many additional challenges due to their high resolution and size. In addition, based on the source MSI scanner hardware, they may be saved in different file formats (e.g., .svs, .ndpi, etc. ). Hence, in order to ease some of the potential pitfalls, we have converted and standardized all collected data into the TIFF format.\n\nYou could potentially help many patients by developing and sharing your classification of blood clot origins. We look forward to interacting with you towards this goal. \n\nWe hope you will enjoy participating in this competition!\n\nCAII Team, Mayo Clinic Florida \n\nHosts: \nMayo Clinic\nBarbaros Selnur Erdal (@Barbaros), Associate Professor of Radiology, Department of Radiology, Mayo Clinic Florida\nMohammed Sobhi Jabal (@Sobhi Jabal), Research Fellow, Department of Radiology, Mayo Clinic Rochester\n*Special thanks to Dr. Waleed Brinjikji, Dr. David F. Kallmes, Dr. Ramanathan Kadirvel, and Dr. Richard D. White.\n\nThank you to our collaborators at Kaggle\nAshley (@Ashley Chow)\nRyan (@Ryan Holbrook)\n", "1847228": "Thanks for participating!", "1990878": "I also tried this. \n", "1848793": "Hi @barbaroserdal - \n\nThis may be a dumb question, but can doctors do this task? \nIn other words, are there physiological indications of stroke etiology/origin?\n\nAre there papers that discuss this that you can link to?", "1931550": "Just had done another medical competition. So really glad to be able to work on another one :)", "1859152": "Hi I am a doctor and I want to join some group here where we can learn to build the algorithm. I can be of any medical related help if required. I will be happy if any group of expert data scientists can chip me in. Many thanks ", "1848246": "Thanks for setting up this competition! I have a few queries on the background information on the dataset e.g. IRB status. Is there a contact email address where I can send these queries? @barbaroserdal maybe? (I'm unable to message you here on Kaggle unfortunately)", "1848156": "Hi, I am new to Kaggle. I have access to a GPU. Is it allowed in the rules to download the data and train the model in my computer (to see how the model performs) and then upload the notebook to do the same thing within Kaggle? This is because GPU resurces in Kaggle are limited.", "1889065": "Frankly, a poor effort at the \"converted and standardized\" all data into tiff format. It is unmanageable data at the size being 300GB+. I and im sure many other google colab users cannot even download it. You should've offered a subset of some lower res slides - contrary to popular beflied deep learning does not always require a lot of data to get decent results, especially with pretrained models. Poorly managed competition.", "1868700": "Hello @barbaroserdal,\nThank you for organizing this competition, I am new to Kaggle and already learning so much! I have a question on the data set and how these tissue slides where stained originally. I have read that using Hematoxylin and eosin (H&E) for staining is the most common way to stain tissues. Were these tissues treated with H & E for staining? across different centers?\n\nThank you much!", "1861343": "Hey thanks for putting up this competition.\n\nI have a question, regarding the image generation in its different centers. \nAs per I understood, the images were generated in eleven different centers (with different file formats, different equipments, models, vendors, parameters, and so on.. ). \nIt would be very important to understand a bit of how the images were generated, and to figure out how they differ from one center to another. This has a huge impact on the model structure and strategy. For instance, centers 11, 4 and 7 account for more than 60% of the images generated, thus, this could create a bias on the model, if such particularities are not tackled properly. \n\nI have posted this question in the forum, too, please take a look. \n https://www.kaggle.com/competitions/mayo-clinic-strip-ai/discussion/337811\n\nHope to hear from you.\n\nBest regards,\nFelipe", "1855961": "Hi @barbaroserdal can people working at Mayo take part?", "1847108": "Thanks for organising this competition. Hope to learn a lot!", "1865564": "Mr. Barbaros, of the 700,000 individuals that experience ischemic Stroke, how many of them are females and Males? From my observations more attention is giving to males than females when it comes to stroke awareness. The actually cause of a stoke cannot be determined. This is what is known as a \"cryptogenic\" stroke or a stroke of unknown cause. It's estimated that 25-30% of ischemic strokes are cryptogenic or unexpected. (Medtronic) \nSince strokes are unexpected lets folks pray to God for good health, eat healthily, and exercise once or twice a week. My advice is; for males one pushup a day, and eat one banana or a grip fruit, for females 15 munities walk in the park or 15 jumping jack will do.", "1862812": "The best solution to preventing Stroke caused by a Blood clot is to consume about three Grapefruit a day, Florida is a Tropical State a Grapefruit plantation should be planted to address this Stroke issue. ", "2001058": "Still fighting ...........!", "1979967": "Hi Barbaros, if my late submission score which could have been a second best score on the leadeboard. Will it contribute towards the goal of this competition in any way?", "1959330": "Hi everyone.. I was just going through the data and wanted to know, how everyone processed and created model for this large amount of data just in case you don't have a personal GPU with you.", "1951816": "From the paper, it seems all slides were acquired with the same scanner, however the magnification is not specified: was it 20x or 40x? The TIFF files have an improper DPI. ", "1936821": "Hello @barbaroserdal ,\n\nLet me ask a question about private test data.\n\nIs there only \"CE\" and \"LAA\" labels for private test data? In other words, would there be no labels like \"other\" or \"unknown\"?", "1928042": "Hi guys\nI released a dataset that contains resized images from this competition(5GB) . Check it out [link](https://www.kaggle.com/datasets/arianghasemi/mayo-clinic-resized-5gb)", "1924323": "Thank you for your organising. This competition is so interesting and I enjoy it ! ", "1872069": "I had some difficulties making **submissions**, \ncan someone please check and lemme know?\nhttps://www.kaggle.com/competitions/mayo-clinic-strip-ai/discussion/339590\n\nThank you", "1860512": "I am really going to enjoy this competition. I hope to learn a lot and make a good solution. Thank you for organising!", "1869158": "" } }