{ "id": 220872, "title": "Bounding box annotations of 2.3k tracheostomy tubes", "url": "/competitions/ranzcr-clip-catheter-line-classification/discussion/220872", "author_name": "", "post_date": "2021-02-19T23:10:29.544478400Z", "votes": 37, "comment_count": 15, "views": 0, "content": "

Hi all,

\n

i'm a radiologist and I hope some of you find the dataset i made useful:
\ni segmented 2231 tracheostomy tubes from x-rays of the dataset of the current challenge.

\n

You can find them here:
\nhttps://www.kaggle.com/sandorkonya/23k-tracheostomy-tube-annotated-on-chest-xray
\n(note: do not use dots in the dataset name… :P )

\n

… 5k trachea bifurcation annotations are also coming, so stay tuned.

", "messages": [ { "id": "1211044", "postDate": "02/19/2021 23:10:29", "content": "

Hi all,

\n

i'm a radiologist and I hope some of you find the dataset i made useful:
\ni segmented 2231 tracheostomy tubes from x-rays of the dataset of the current challenge.

\n

You can find them here:
\nhttps://www.kaggle.com/sandorkonya/23k-tracheostomy-tube-annotated-on-chest-xray
\n(note: do not use dots in the dataset name… :P )

\n

… 5k trachea bifurcation annotations are also coming, so stay tuned.

", "rawMarkdown": "Hi all, \n\ni'm a radiologist and I hope some of you find the dataset i made useful:\ni segmented 2231 tracheostomy tubes from x-rays of the dataset of the current challenge.\n\nYou can find them here:\nhttps://www.kaggle.com/sandorkonya/23k-tracheostomy-tube-annotated-on-chest-xray\n(note: do not use dots in the dataset name... :P )\n\n\n... 5k trachea bifurcation annotations are also coming, so stay tuned.", "votes": null }, { "id": "1211052", "postDate": "02/19/2021 23:26:12", "content": "

Thanks doc.

", "rawMarkdown": "Thanks doc.", "votes": null }, { "id": "1211060", "postDate": "02/19/2021 23:36:02", "content": "

Amazing. Thanks a lot!

", "rawMarkdown": "Amazing. Thanks a lot!", "votes": null }, { "id": "1211244", "postDate": "02/20/2021 05:12:35", "content": "

We really appreciate it. Thank you very much!

", "rawMarkdown": "We really appreciate it. Thank you very much!", "votes": null }, { "id": "1211334", "postDate": "02/20/2021 06:25:43", "content": "

i think kaggle should introduce a scheme to give credits to dataset creator like you if the final top winning solution uses your work.

", "rawMarkdown": "i think kaggle should introduce a scheme to give credits to dataset creator like you if the final top winning solution uses your work.", "votes": null }, { "id": "1211383", "postDate": "02/20/2021 07:05:29", "content": "

@hengck23 ,

\n

ideally such data contributes to the team's success and gets published afterwards.
\nMy teammembers got other priorities, do not further participate in the present competition and I did not want the work to get completely lost.
\nNevertheless an upvote on the data gets me somewhat further with kaggle's perks =)

", "rawMarkdown": "hengck23 ,\n\nideally such data contributes to the team's success and gets published afterwards. \nMy teammembers got other priorities, do not further participate in the present competition and I did not want the work to get completely lost.\nNevertheless an upvote on the data gets me somewhat further with kaggle's perks =)", "votes": null }, { "id": "1211386", "postDate": "02/20/2021 07:06:14", "content": "

@nguyenbadung,
\nmore coming ;)

", "rawMarkdown": "nguyenbadung, \nmore coming ;)", "votes": null }, { "id": "1211708", "postDate": "02/20/2021 13:19:51", "content": "

Just my 2 cents here - and please do not take it too personally - your work is really appreciated. I think these have little value to the competition.

\n

Tracheostomy tubes cannot(?) be borderline/abnormal (as train labels confirm that based on your annotations - all ETT - Normal). So any model would pick them up easily - if tracheostomy => 1.0 probability for ETT - Normal category.

\n

To confirm this, I took one of my model predictions. sampled 2k non ETT - Normal images as negatives for your tracheostomy cases. Model AUC = 0.999 on this stratified sample.

", "rawMarkdown": "Just my 2 cents here - and please do not take it too personally - your work is really appreciated. I think these have little value to the competition.\n\nTracheostomy tubes cannot(?) be borderline/abnormal (as train labels confirm that based on your annotations - all `ETT - Normal`). So any model would pick them up easily - if tracheostomy => 1.0 probability for `ETT - Normal` category.\n\nTo confirm this, I took one of my model predictions. sampled 2k non `ETT - Normal` images as negatives for your tracheostomy cases. Model AUC = 0.999 on this stratified sample.", "votes": null }, { "id": "1211757", "postDate": "02/20/2021 14:13:19", "content": "

@raddar ,

\n

hi there, i observed exactly the same: in this dataset there was no abnormally placed thracheostomy tube… which on the other hand would have been pretty hard to detect if we cannot identify a tracheostomy tube vs. ETT.
\nFYI a tracheostomy tube could be misplaced to, but it causes pretty rapidly clinically relavant symptomes so it's malposition is w/o imaging pretty obvious =)

\n

This is why i annotated them for further investigation to see whether any of them is abnormal at all… and - as you - i did not find any in the dataset… so your assumption: if tracheostomy tube ~= \"ETT Normal\" label is probably correct.

\n

There were cases however (with only partially captured ETT's or wire-strengthened ETT's that have been mislabelled as tracheostomy tube, but they were even all \"ETT normal\" anyway.

", "rawMarkdown": "raddar ,\n\nhi there, i observed exactly the same: in this dataset there was no abnormally placed thracheostomy tube... which on the other hand would have been pretty hard to detect if we cannot identify a tracheostomy tube vs. ETT.\nFYI a tracheostomy tube could be misplaced to, but it causes pretty rapidly clinically relavant symptomes so it's malposition is w/o imaging pretty obvious =)\n\nThis is why i annotated them for further investigation to see whether any of them is abnormal at all... and - as you - i did not find any in the dataset... so your assumption: if tracheostomy tube ~= \"ETT Normal\" label is probably correct.\n\nThere were cases however (with only partially captured ETT's or wire-strengthened ETT's that have been mislabelled as tracheostomy tube, but they were even all \"ETT normal\" anyway.", "votes": null }, { "id": "1212019", "postDate": "02/20/2021 19:26:25", "content": "

I did some quick pseudolabeling on remaining train set, these were picked up:

\n

['1.2.826.0.1.3680043.8.498.10691109321215449169858921729737657552',
\n '1.2.826.0.1.3680043.8.498.67765185042322560730445282680235021957',
\n '1.2.826.0.1.3680043.8.498.10128199840925023408402703824969146572',
\n '1.2.826.0.1.3680043.8.498.66720904138960949145882479059576919204',
\n '1.2.826.0.1.3680043.8.498.99506083249595014359124859275968802835',
\n '1.2.826.0.1.3680043.8.498.11947414383094977021120146998340264825',
\n '1.2.826.0.1.3680043.8.498.12521684042177560843225639265615222030',
\n '1.2.826.0.1.3680043.8.498.10403643318820518980119415879496953467',
\n '1.2.826.0.1.3680043.8.498.13380144744220166212663042237373006283',
\n '1.2.826.0.1.3680043.8.498.22544742954181425156936634481475702500',
\n '1.2.826.0.1.3680043.8.498.18386757238555595487710768413991842410',
\n '1.2.826.0.1.3680043.8.498.89731243445631162862494675265766306590',
\n '1.2.826.0.1.3680043.8.498.74610493264621702314300635578142075251',
\n '1.2.826.0.1.3680043.8.498.84939523501306668183596702990832330453',
\n '1.2.826.0.1.3680043.8.498.78250685323360307738948154683455025022',
\n '1.2.826.0.1.3680043.8.498.26040311428295892195247941419133654159',
\n '1.2.826.0.1.3680043.8.498.26025410563246460470961780945557113068',
\n '1.2.826.0.1.3680043.8.498.43291508294149348778091794222222641436',
\n '1.2.826.0.1.3680043.8.498.33308596717182494805755647707550375551',
\n '1.2.826.0.1.3680043.8.498.71070306150092575551403902997688327271',
\n '1.2.826.0.1.3680043.8.498.47425394958331616747477785636847932948',
\n '1.2.826.0.1.3680043.8.498.51453968865050617560669858565163944718',
\n '1.2.826.0.1.3680043.8.498.49105589657274382470535120468435810820']

\n

These do contain Borderline. Could you confirm if these are also tracheostomies?

", "rawMarkdown": "I did some quick pseudolabeling on remaining train set, these were picked up:\n\n['1.2.826.0.1.3680043.8.498.10691109321215449169858921729737657552',\n '1.2.826.0.1.3680043.8.498.67765185042322560730445282680235021957',\n '1.2.826.0.1.3680043.8.498.10128199840925023408402703824969146572',\n '1.2.826.0.1.3680043.8.498.66720904138960949145882479059576919204',\n '1.2.826.0.1.3680043.8.498.99506083249595014359124859275968802835',\n '1.2.826.0.1.3680043.8.498.11947414383094977021120146998340264825',\n '1.2.826.0.1.3680043.8.498.12521684042177560843225639265615222030',\n '1.2.826.0.1.3680043.8.498.10403643318820518980119415879496953467',\n '1.2.826.0.1.3680043.8.498.13380144744220166212663042237373006283',\n '1.2.826.0.1.3680043.8.498.22544742954181425156936634481475702500',\n '1.2.826.0.1.3680043.8.498.18386757238555595487710768413991842410',\n '1.2.826.0.1.3680043.8.498.89731243445631162862494675265766306590',\n '1.2.826.0.1.3680043.8.498.74610493264621702314300635578142075251',\n '1.2.826.0.1.3680043.8.498.84939523501306668183596702990832330453',\n '1.2.826.0.1.3680043.8.498.78250685323360307738948154683455025022',\n '1.2.826.0.1.3680043.8.498.26040311428295892195247941419133654159',\n '1.2.826.0.1.3680043.8.498.26025410563246460470961780945557113068',\n '1.2.826.0.1.3680043.8.498.43291508294149348778091794222222641436',\n '1.2.826.0.1.3680043.8.498.33308596717182494805755647707550375551',\n '1.2.826.0.1.3680043.8.498.71070306150092575551403902997688327271',\n '1.2.826.0.1.3680043.8.498.47425394958331616747477785636847932948',\n '1.2.826.0.1.3680043.8.498.51453968865050617560669858565163944718',\n '1.2.826.0.1.3680043.8.498.49105589657274382470535120468435810820']\n\n\nThese do contain Borderline. Could you confirm if these are also tracheostomies?", "votes": null }, { "id": "1212050", "postDate": "02/20/2021 20:25:33", "content": "

Not all of them are tracheostomy tubes:

\n

1.2.826.0.1.3680043.8.498.67765185042322560730445282680235021957, this is TT (there is a tube in projection of the right main bronchus to but extracorporal =) very tricky)
\n1.2.826.0.1.3680043.8.498.10128199840925023408402703824969146572, this is the case i was talking about --> probably ETT with reinforcements or special reinforced TT partially caught
\n1.2.826.0.1.3680043.8.498.84939523501306668183596702990832330453
\n --> again ETT with reinforcements or special reinforced TT

\n

The others are all TT from probably different vendors (due to differences in appearance).

", "rawMarkdown": "Not all of them are tracheostomy tubes:\n\n1.2.826.0.1.3680043.8.498.67765185042322560730445282680235021957, this is TT (there is a tube in projection of the right main bronchus to but extracorporal =) very tricky)\n1.2.826.0.1.3680043.8.498.10128199840925023408402703824969146572, this is the case i was talking about --> probably ETT with reinforcements or special reinforced TT partially caught\n1.2.826.0.1.3680043.8.498.84939523501306668183596702990832330453\n --> again ETT with reinforcements or special reinforced TT\n\nThe others are all TT from probably different vendors (due to differences in appearance).", "votes": null }, { "id": "1212701", "postDate": "02/21/2021 13:49:17", "content": "

Amazing!
\nBut how can we use it?

", "rawMarkdown": "Amazing!\nBut how can we use it?", "votes": null }, { "id": "1212743", "postDate": "02/21/2021 14:34:47", "content": "

@zekunn,
\nwell, it seems that every TT is labelled as \"ETT - Normal\" in the dataset, so if there is a TT on the image, it cannot be \"Abnormal\" - at least on the images in this competition.

", "rawMarkdown": "zekunn,\nwell, it seems that every TT is labelled as \"ETT - Normal\" in the dataset, so if there is a TT on the image, it cannot be \"Abnormal\" - at least on the images in this competition.", "votes": null }, { "id": "1212745", "postDate": "02/21/2021 14:38:28", "content": "

Thanks, we can label them!

", "rawMarkdown": "Thanks, we can label them!", "votes": null }, { "id": "1214382", "postDate": "02/22/2021 19:48:51", "content": "

Thank you!

", "rawMarkdown": "Thank you!", "votes": null }, { "id": "1218904", "postDate": "02/26/2021 09:19:40", "content": "

Great work. Can you share any kernel which uses this dataset? Thanks in advance.

", "rawMarkdown": "Great work. Can you share any kernel which uses this dataset? Thanks in advance.", "votes": null } ], "comments": [ { "id": 1211052, "author_name": "underwearfitting", "author_url": "", "post_date": "02/19/2021 23:26:12", "content": "

Thanks doc.

", "votes": null, "replies": [] }, { "id": 1211060, "author_name": "virilo", "author_url": "", "post_date": "02/19/2021 23:36:02", "content": "

Amazing. Thanks a lot!

", "votes": null, "replies": [] }, { "id": 1211244, "author_name": "nguyenbadung", "author_url": "", "post_date": "02/20/2021 05:12:35", "content": "

We really appreciate it. Thank you very much!

", "votes": null, "replies": [ { "id": 1211386, "author_name": "sandorkonya", "author_url": "", "post_date": "02/20/2021 07:06:14", "content": "

@nguyenbadung,
\nmore coming ;)

", "votes": null, "replies": [] } ] }, { "id": 1211334, "author_name": "hengck23", "author_url": "", "post_date": "02/20/2021 06:25:43", "content": "

i think kaggle should introduce a scheme to give credits to dataset creator like you if the final top winning solution uses your work.

", "votes": null, "replies": [ { "id": 1211383, "author_name": "sandorkonya", "author_url": "", "post_date": "02/20/2021 07:05:29", "content": "

@hengck23 ,

\n

ideally such data contributes to the team's success and gets published afterwards.
\nMy teammembers got other priorities, do not further participate in the present competition and I did not want the work to get completely lost.
\nNevertheless an upvote on the data gets me somewhat further with kaggle's perks =)

", "votes": null, "replies": [] } ] }, { "id": 1211708, "author_name": "raddar", "author_url": "", "post_date": "02/20/2021 13:19:51", "content": "

Just my 2 cents here - and please do not take it too personally - your work is really appreciated. I think these have little value to the competition.

\n

Tracheostomy tubes cannot(?) be borderline/abnormal (as train labels confirm that based on your annotations - all ETT - Normal). So any model would pick them up easily - if tracheostomy => 1.0 probability for ETT - Normal category.

\n

To confirm this, I took one of my model predictions. sampled 2k non ETT - Normal images as negatives for your tracheostomy cases. Model AUC = 0.999 on this stratified sample.

", "votes": null, "replies": [ { "id": 1211757, "author_name": "sandorkonya", "author_url": "", "post_date": "02/20/2021 14:13:19", "content": "

@raddar ,

\n

hi there, i observed exactly the same: in this dataset there was no abnormally placed thracheostomy tube… which on the other hand would have been pretty hard to detect if we cannot identify a tracheostomy tube vs. ETT.
\nFYI a tracheostomy tube could be misplaced to, but it causes pretty rapidly clinically relavant symptomes so it's malposition is w/o imaging pretty obvious =)

\n

This is why i annotated them for further investigation to see whether any of them is abnormal at all… and - as you - i did not find any in the dataset… so your assumption: if tracheostomy tube ~= \"ETT Normal\" label is probably correct.

\n

There were cases however (with only partially captured ETT's or wire-strengthened ETT's that have been mislabelled as tracheostomy tube, but they were even all \"ETT normal\" anyway.

", "votes": null, "replies": [] }, { "id": 1212019, "author_name": "raddar", "author_url": "", "post_date": "02/20/2021 19:26:25", "content": "

I did some quick pseudolabeling on remaining train set, these were picked up:

\n

['1.2.826.0.1.3680043.8.498.10691109321215449169858921729737657552',
\n '1.2.826.0.1.3680043.8.498.67765185042322560730445282680235021957',
\n '1.2.826.0.1.3680043.8.498.10128199840925023408402703824969146572',
\n '1.2.826.0.1.3680043.8.498.66720904138960949145882479059576919204',
\n '1.2.826.0.1.3680043.8.498.99506083249595014359124859275968802835',
\n '1.2.826.0.1.3680043.8.498.11947414383094977021120146998340264825',
\n '1.2.826.0.1.3680043.8.498.12521684042177560843225639265615222030',
\n '1.2.826.0.1.3680043.8.498.10403643318820518980119415879496953467',
\n '1.2.826.0.1.3680043.8.498.13380144744220166212663042237373006283',
\n '1.2.826.0.1.3680043.8.498.22544742954181425156936634481475702500',
\n '1.2.826.0.1.3680043.8.498.18386757238555595487710768413991842410',
\n '1.2.826.0.1.3680043.8.498.89731243445631162862494675265766306590',
\n '1.2.826.0.1.3680043.8.498.74610493264621702314300635578142075251',
\n '1.2.826.0.1.3680043.8.498.84939523501306668183596702990832330453',
\n '1.2.826.0.1.3680043.8.498.78250685323360307738948154683455025022',
\n '1.2.826.0.1.3680043.8.498.26040311428295892195247941419133654159',
\n '1.2.826.0.1.3680043.8.498.26025410563246460470961780945557113068',
\n '1.2.826.0.1.3680043.8.498.43291508294149348778091794222222641436',
\n '1.2.826.0.1.3680043.8.498.33308596717182494805755647707550375551',
\n '1.2.826.0.1.3680043.8.498.71070306150092575551403902997688327271',
\n '1.2.826.0.1.3680043.8.498.47425394958331616747477785636847932948',
\n '1.2.826.0.1.3680043.8.498.51453968865050617560669858565163944718',
\n '1.2.826.0.1.3680043.8.498.49105589657274382470535120468435810820']

\n

These do contain Borderline. Could you confirm if these are also tracheostomies?

", "votes": null, "replies": [] }, { "id": 1212050, "author_name": "sandorkonya", "author_url": "", "post_date": "02/20/2021 20:25:33", "content": "

Not all of them are tracheostomy tubes:

\n

1.2.826.0.1.3680043.8.498.67765185042322560730445282680235021957, this is TT (there is a tube in projection of the right main bronchus to but extracorporal =) very tricky)
\n1.2.826.0.1.3680043.8.498.10128199840925023408402703824969146572, this is the case i was talking about --> probably ETT with reinforcements or special reinforced TT partially caught
\n1.2.826.0.1.3680043.8.498.84939523501306668183596702990832330453
\n --> again ETT with reinforcements or special reinforced TT

\n

The others are all TT from probably different vendors (due to differences in appearance).

", "votes": null, "replies": [] } ] }, { "id": 1212701, "author_name": "zekunn", "author_url": "", "post_date": "02/21/2021 13:49:17", "content": "

Amazing!
\nBut how can we use it?

", "votes": null, "replies": [ { "id": 1212743, "author_name": "sandorkonya", "author_url": "", "post_date": "02/21/2021 14:34:47", "content": "

@zekunn,
\nwell, it seems that every TT is labelled as \"ETT - Normal\" in the dataset, so if there is a TT on the image, it cannot be \"Abnormal\" - at least on the images in this competition.

", "votes": null, "replies": [] }, { "id": 1212745, "author_name": "zekunn", "author_url": "", "post_date": "02/21/2021 14:38:28", "content": "

Thanks, we can label them!

", "votes": null, "replies": [] } ] }, { "id": 1214382, "author_name": "atamazian", "author_url": "", "post_date": "02/22/2021 19:48:51", "content": "

Thank you!

", "votes": null, "replies": [] }, { "id": 1218904, "author_name": "durbin164", "author_url": "", "post_date": "02/26/2021 09:19:40", "content": "

Great work. Can you share any kernel which uses this dataset? Thanks in advance.

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1211044": "Hi all, \n\ni'm a radiologist and I hope some of you find the dataset i made useful:\ni segmented 2231 tracheostomy tubes from x-rays of the dataset of the current challenge.\n\nYou can find them here:\nhttps://www.kaggle.com/sandorkonya/23k-tracheostomy-tube-annotated-on-chest-xray\n(note: do not use dots in the dataset name... :P )\n\n\n... 5k trachea bifurcation annotations are also coming, so stay tuned.", "1211052": "Thanks doc.", "1211060": "Amazing. Thanks a lot!", "1211244": "We really appreciate it. Thank you very much!", "1211334": "i think kaggle should introduce a scheme to give credits to dataset creator like you if the final top winning solution uses your work.", "1211383": "hengck23 ,\n\nideally such data contributes to the team's success and gets published afterwards. \nMy teammembers got other priorities, do not further participate in the present competition and I did not want the work to get completely lost.\nNevertheless an upvote on the data gets me somewhat further with kaggle's perks =)", "1211386": "nguyenbadung, \nmore coming ;)", "1211708": "Just my 2 cents here - and please do not take it too personally - your work is really appreciated. I think these have little value to the competition.\n\nTracheostomy tubes cannot(?) be borderline/abnormal (as train labels confirm that based on your annotations - all `ETT - Normal`). So any model would pick them up easily - if tracheostomy => 1.0 probability for `ETT - Normal` category.\n\nTo confirm this, I took one of my model predictions. sampled 2k non `ETT - Normal` images as negatives for your tracheostomy cases. Model AUC = 0.999 on this stratified sample.", "1211757": "raddar ,\n\nhi there, i observed exactly the same: in this dataset there was no abnormally placed thracheostomy tube... which on the other hand would have been pretty hard to detect if we cannot identify a tracheostomy tube vs. ETT.\nFYI a tracheostomy tube could be misplaced to, but it causes pretty rapidly clinically relavant symptomes so it's malposition is w/o imaging pretty obvious =)\n\nThis is why i annotated them for further investigation to see whether any of them is abnormal at all... and - as you - i did not find any in the dataset... so your assumption: if tracheostomy tube ~= \"ETT Normal\" label is probably correct.\n\nThere were cases however (with only partially captured ETT's or wire-strengthened ETT's that have been mislabelled as tracheostomy tube, but they were even all \"ETT normal\" anyway.", "1212019": "I did some quick pseudolabeling on remaining train set, these were picked up:\n\n['1.2.826.0.1.3680043.8.498.10691109321215449169858921729737657552',\n '1.2.826.0.1.3680043.8.498.67765185042322560730445282680235021957',\n '1.2.826.0.1.3680043.8.498.10128199840925023408402703824969146572',\n '1.2.826.0.1.3680043.8.498.66720904138960949145882479059576919204',\n '1.2.826.0.1.3680043.8.498.99506083249595014359124859275968802835',\n '1.2.826.0.1.3680043.8.498.11947414383094977021120146998340264825',\n '1.2.826.0.1.3680043.8.498.12521684042177560843225639265615222030',\n '1.2.826.0.1.3680043.8.498.10403643318820518980119415879496953467',\n '1.2.826.0.1.3680043.8.498.13380144744220166212663042237373006283',\n '1.2.826.0.1.3680043.8.498.22544742954181425156936634481475702500',\n '1.2.826.0.1.3680043.8.498.18386757238555595487710768413991842410',\n '1.2.826.0.1.3680043.8.498.89731243445631162862494675265766306590',\n '1.2.826.0.1.3680043.8.498.74610493264621702314300635578142075251',\n '1.2.826.0.1.3680043.8.498.84939523501306668183596702990832330453',\n '1.2.826.0.1.3680043.8.498.78250685323360307738948154683455025022',\n '1.2.826.0.1.3680043.8.498.26040311428295892195247941419133654159',\n '1.2.826.0.1.3680043.8.498.26025410563246460470961780945557113068',\n '1.2.826.0.1.3680043.8.498.43291508294149348778091794222222641436',\n '1.2.826.0.1.3680043.8.498.33308596717182494805755647707550375551',\n '1.2.826.0.1.3680043.8.498.71070306150092575551403902997688327271',\n '1.2.826.0.1.3680043.8.498.47425394958331616747477785636847932948',\n '1.2.826.0.1.3680043.8.498.51453968865050617560669858565163944718',\n '1.2.826.0.1.3680043.8.498.49105589657274382470535120468435810820']\n\n\nThese do contain Borderline. Could you confirm if these are also tracheostomies?", "1212050": "Not all of them are tracheostomy tubes:\n\n1.2.826.0.1.3680043.8.498.67765185042322560730445282680235021957, this is TT (there is a tube in projection of the right main bronchus to but extracorporal =) very tricky)\n1.2.826.0.1.3680043.8.498.10128199840925023408402703824969146572, this is the case i was talking about --> probably ETT with reinforcements or special reinforced TT partially caught\n1.2.826.0.1.3680043.8.498.84939523501306668183596702990832330453\n --> again ETT with reinforcements or special reinforced TT\n\nThe others are all TT from probably different vendors (due to differences in appearance).", "1212701": "Amazing!\nBut how can we use it?", "1212743": "zekunn,\nwell, it seems that every TT is labelled as \"ETT - Normal\" in the dataset, so if there is a TT on the image, it cannot be \"Abnormal\" - at least on the images in this competition.", "1212745": "Thanks, we can label them!", "1214382": "Thank you!", "1218904": "Great work. Can you share any kernel which uses this dataset? Thanks in advance." }, "source": "meta" }