![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F405318%2F2b5b339c3d62b46b8fcd30ee7268bf0f%2F__results___3_0.png?generation=1609180695053338&alt=media\")
Following previous lung contour dataset (as in https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/207183) I also decided to share the model predictions of tracheal bifurcation points. These bifurcation points are used to determine if ETT is normal/borderline/abnormal. The ETT tube point should be 3cm+ above bifurcation point to be considered normal.
\n
The dataset can be found at:
\nhttps://www.kaggle.com/raddar/ranzcr-clip-tracheal-bifurcation
Simple exploratory notebook:
\nhttps://www.kaggle.com/raddar/simple-ett-bifurcation-visualization
\n^ In the notebook I discovered that there are wrong labels in the training set - be careful!
Following previous lung contour dataset (as in https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/207183) I also decided to share the model predictions of tracheal bifurcation points. These bifurcation points are used to determine if ETT is normal/borderline/abnormal. The ETT tube point should be 3cm+ above bifurcation point to be considered normal.
\n
The dataset can be found at:
\nhttps://www.kaggle.com/raddar/ranzcr-clip-tracheal-bifurcation
Simple exploratory notebook:
\nhttps://www.kaggle.com/raddar/simple-ett-bifurcation-visualization
\n^ In the notebook I discovered that there are wrong labels in the training set - be careful!
Some extra info about errors in ETT - Abnormal class:
https://www.kaggle.com/raddar/errors-in-ett-abnormal-labels
", "rawMarkdown": "Some extra info about errors in `ETT - Abnormal` class:\n\nhttps://www.kaggle.com/raddar/errors-in-ett-abnormal-labels", "votes": null }, { "id": "1134820", "postDate": "01/01/2021 15:51:34", "content": "Dear @raddar,
\nthe idea of detecting the bifurcation is the correct way to determine the abnormal position of the ETT, however
\non the image provided here (and all the images in the ett-bifurcation-visualisation notebook) the red circle does not denote the bifurcation itself, it is positioned above the bifurcation. I suppose they denote the point already ~ 3 cm above the bifurcation.
The difference between borderline and abnormal is not only the position of the tip: by questionable position of the tip of ETT we look at the ventillation of both lungs! Is the affected side not hypoventillated ( == \"white\") then it may be considered borderline.
\nThis is one of the shortcommings of the present challenge, because the question is - do we draw a therapeutic decision of out diagnosis, and in both abnormal and borderline positions we suggest the repositioning of the ETT, so it could have been one category.
", "rawMarkdown": "Dear @raddar,\n\nthe idea of detecting the bifurcation is the correct way to determine the abnormal position of the ETT, however \non the image provided here (and all the images in the ett-bifurcation-visualisation notebook) the red circle does not denote the bifurcation itself, it is positioned above the bifurcation. I suppose they denote the point already ~ 3 cm above the bifurcation.\n\nThe difference between borderline and abnormal is not only the position of the tip: by questionable position of the tip of ETT we look at the ventillation of both lungs! Is the affected side not hypoventillated ( == \"white\") then it may be considered borderline.\n\nThis is one of the shortcommings of the present challenge, because the question is - do we draw a therapeutic decision of out diagnosis, and in both abnormal and borderline positions we suggest the repositioning of the ETT, so it could have been one category.", "votes": null }, { "id": "1134849", "postDate": "01/01/2021 16:23:43", "content": "Thank you for your comments.
\nI agree with your observations - the examples I have shown was for illustration purposes. Also, these are model predictions (not radiologist annotations) - and I agree that they are not perfect (like any other model predictions). The reason for model failing is that the model was trained on DICOM data and we have a single-view jpg's which makes the task harder.
\nAs for borderline, the competition was defined by organizers as:
\nThe borderline category includes lines that would ideally require some repositioning but would in most cases still function adequately in their current position.
Great work!So,how should we use this data?We need add it to the competition data 'train_annotations.csv'?Thanks!
", "rawMarkdown": "Great work!So,how should we use this data?We need add it to the competition data 'train_annotations.csv'?Thanks!", "votes": null } ], "comments": [ { "id": 1130164, "author_name": "raddar", "author_url": "", "post_date": "12/28/2020 19:24:03", "content": "Some extra info about errors in ETT - Abnormal class:
https://www.kaggle.com/raddar/errors-in-ett-abnormal-labels
", "votes": null, "replies": [] }, { "id": 1134820, "author_name": "sandorkonya", "author_url": "", "post_date": "01/01/2021 15:51:34", "content": "Dear @raddar,
\nthe idea of detecting the bifurcation is the correct way to determine the abnormal position of the ETT, however
\non the image provided here (and all the images in the ett-bifurcation-visualisation notebook) the red circle does not denote the bifurcation itself, it is positioned above the bifurcation. I suppose they denote the point already ~ 3 cm above the bifurcation.
The difference between borderline and abnormal is not only the position of the tip: by questionable position of the tip of ETT we look at the ventillation of both lungs! Is the affected side not hypoventillated ( == \"white\") then it may be considered borderline.
\nThis is one of the shortcommings of the present challenge, because the question is - do we draw a therapeutic decision of out diagnosis, and in both abnormal and borderline positions we suggest the repositioning of the ETT, so it could have been one category.
", "votes": null, "replies": [ { "id": 1134849, "author_name": "raddar", "author_url": "", "post_date": "01/01/2021 16:23:43", "content": "Thank you for your comments.
\nI agree with your observations - the examples I have shown was for illustration purposes. Also, these are model predictions (not radiologist annotations) - and I agree that they are not perfect (like any other model predictions). The reason for model failing is that the model was trained on DICOM data and we have a single-view jpg's which makes the task harder.
\nAs for borderline, the competition was defined by organizers as:
\nThe borderline category includes lines that would ideally require some repositioning but would in most cases still function adequately in their current position.
Great work!So,how should we use this data?We need add it to the competition data 'train_annotations.csv'?Thanks!
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1130106": "Following previous lung contour dataset (as in https://www.kaggle.com/c/ranzcr-clip-catheter-line-classification/discussion/207183) I also decided to share the model predictions of tracheal bifurcation points. These bifurcation points are used to determine if ETT is normal/borderline/abnormal. The ETT tube point should be 3cm+ above bifurcation point to be considered normal.\n\n\n\nThe dataset can be found at:\nhttps://www.kaggle.com/raddar/ranzcr-clip-tracheal-bifurcation\n\nSimple exploratory notebook:\nhttps://www.kaggle.com/raddar/simple-ett-bifurcation-visualization\n^ In the notebook I discovered that there are wrong labels in the training set - be careful!", "1130164": "Some extra info about errors in `ETT - Abnormal` class:\n\nhttps://www.kaggle.com/raddar/errors-in-ett-abnormal-labels", "1134820": "Dear @raddar,\n\nthe idea of detecting the bifurcation is the correct way to determine the abnormal position of the ETT, however \non the image provided here (and all the images in the ett-bifurcation-visualisation notebook) the red circle does not denote the bifurcation itself, it is positioned above the bifurcation. I suppose they denote the point already ~ 3 cm above the bifurcation.\n\nThe difference between borderline and abnormal is not only the position of the tip: by questionable position of the tip of ETT we look at the ventillation of both lungs! Is the affected side not hypoventillated ( == \"white\") then it may be considered borderline.\n\nThis is one of the shortcommings of the present challenge, because the question is - do we draw a therapeutic decision of out diagnosis, and in both abnormal and borderline positions we suggest the repositioning of the ETT, so it could have been one category.", "1134849": "Thank you for your comments.\n\nI agree with your observations - the examples I have shown was for illustration purposes. Also, these are model predictions (not radiologist annotations) - and I agree that they are not perfect (like any other model predictions). The reason for model failing is that the model was trained on DICOM data and we have a single-view jpg's which makes the task harder.\n\nAs for borderline, the competition was defined by organizers as:\n\n`The borderline category includes lines that would ideally require some repositioning but would in most cases still function adequately in their current position.`", "1164639": "Great work!So,how should we use this data?We need add it to the competition data 'train_annotations.csv'?Thanks!" }, "source": "meta" }