{ "id": 240069, "title": "No BBox info but \"Negative for Pneumonia\"==0 images", "url": "/competitions/siim-covid19-detection/discussion/240069", "author_name": "", "post_date": "2021-05-18T13:16:07.037812300Z", "votes": 9, "comment_count": 16, "views": 0, "content": "

There are some ids where \"Negative for Pneumonia\"==0 without any BBox.
\nFor example, id: 50804b6043ac_image, 51759b5579bc_image, ….
\n\"Screen
\n\"Screen
\nI guess they are abnormal (pathological) but no BBox information.
\nIf BBoxes only for pneumonia, their \"Negative for Pneumonia\" should be 1.
\nIf BBoxes also for abnormalities including pneumonia, it is possible that their \"Negative for Pneumonia\"==0, but we need BBox information.
\nOr they have pneumonia, but not because of COVID-19?
\nDo you have any idea? 😂

", "messages": [ { "id": "1313243", "postDate": "05/18/2021 13:16:07", "content": "

There are some ids where \"Negative for Pneumonia\"==0 without any BBox.
\nFor example, id: 50804b6043ac_image, 51759b5579bc_image, ….
\n\"Screen
\n\"Screen
\nI guess they are abnormal (pathological) but no BBox information.
\nIf BBoxes only for pneumonia, their \"Negative for Pneumonia\" should be 1.
\nIf BBoxes also for abnormalities including pneumonia, it is possible that their \"Negative for Pneumonia\"==0, but we need BBox information.
\nOr they have pneumonia, but not because of COVID-19?
\nDo you have any idea? 😂

", "rawMarkdown": "There are some ids where \"Negative for Pneumonia\"==0 without any BBox.\nFor example, id: 50804b6043ac_image, 51759b5579bc_image, ....\n\"Screen\n\"Screen\nI guess they are abnormal (pathological) but no BBox information.\nIf BBoxes only for pneumonia, their \"Negative for Pneumonia\" should be 1.\nIf BBoxes also for abnormalities including pneumonia, it is possible that their \"Negative for Pneumonia\"==0, but we need BBox information.\nOr they have pneumonia, but not because of COVID-19?\nDo you have any idea? 😂", "votes": null }, { "id": "1313413", "postDate": "05/18/2021 14:49:04", "content": "

Yes, there are 82 non-negative studies without any bounding boxes. 81/82 areatypical and 1 is typical. It would be nice to have more explanation on this.

\n

Also, interestingly all negative studies don't have any image with a bounding box, which might mean that there are no cases of opacity which can be labeled as negative at least thats the case with our training dataset.

", "rawMarkdown": "Yes, there are 82 non-negative studies without any bounding boxes. 81/82 are`atypical` and 1 is `typical`. It would be nice to have more explanation on this.\n\nAlso, interestingly all negative studies don't have any image with a bounding box, which might mean that there are no cases of opacity which can be labeled as `negative` at least thats the case with our training dataset.", "votes": null }, { "id": "1313416", "postDate": "05/18/2021 14:51:33", "content": "

Yeah, 81+1 is the same as my EDA.
\nOne possibility is noisy label…

", "rawMarkdown": "Yeah, 81+1 is the same as my EDA.\nOne possibility is noisy label...", "votes": null }, { "id": "1313631", "postDate": "05/18/2021 16:28:34", "content": "

Based on the images, we could arguably say that these 2 cases have lung opacities.

", "rawMarkdown": "Based on the images, we could arguably say that these 2 cases have lung opacities.", "votes": null }, { "id": "1313797", "postDate": "05/18/2021 18:11:23", "content": "

Yes. Waiting for the official answer.

", "rawMarkdown": "Yes. Waiting for the official answer.", "votes": null }, { "id": "1313906", "postDate": "05/18/2021 19:33:33", "content": "

@alexandrecc What is your intuition about no finding cases? Also annotation process is not explained what are your thoughts on that? Data page states COVID-19 patients but how can we tell apart from regular COVID-19 related pneumonia vs regular pneumonia? Is this challenge indeed specific to COVID cases? Sorry for bombardment, but couldn't resist after seeing you here in the discussion :D

", "rawMarkdown": "alexandrecc What is your intuition about no finding cases? Also annotation process is not explained what are your thoughts on that? Data page states COVID-19 patients but how can we tell apart from regular COVID-19 related pneumonia vs regular pneumonia? Is this challenge indeed specific to COVID cases? Sorry for bombardment, but couldn't resist after seeing you here in the discussion :D", "votes": null }, { "id": "1313958", "postDate": "05/18/2021 20:29:10", "content": "

What is your intuition about no finding cases?
\nEither it's 1) an error in data or 2) they considered PCR + cases as positive for the competition even if there are no visible lung opacities. I hope it's 1) because 2) doesn`t make sense for a computer vision competition.

\n

Also annotation process is not explained what are your thoughts on that?
\nI am interested as much as you are in having more details from the host about the annotation process. Mostly about the number and expertise of annotators between train and test datasets which can create a significant bias. I am also interested to know more about how the annotators were instructed regarding the atypical and indeterminate classes.

\n

Data page states COVID-19 patients but how can we tell apart from regular COVID-19 related pneumonia vs regular pneumonia?
\nFrom a human perspective, it's not really possible to be specific. The, so called, typical appearance of Covid-19 grossly means the presence of multiple bilateral peripherical lung opacities. This is somewhat non-specific with many other pathological conditions that can mimic this presentation on chest xrays. We, as radiologists, are highly biased by the context (pandemic, clinical presentation) to conclude that a case is a probable radiological presentation of a Covid-19 pneumonia. In my practice in Canada, PCR confirmation is always needed. Chest xray in the current pandemic context is mostly used to exclude other treatable thoracic problems and to have some sense of the disease severity which is usually very well correlated with the clinical condition of the patient.

\n

Is this challenge indeed specific to COVID cases?
\nMost probably the inclusion criteria for positive cases was a PCR positive test with a chest xray from different institutions. This hypothesis could also be confirmed by the host. They probably included random control-negative cases from the same institutions.

", "rawMarkdown": "**What is your intuition about no finding cases?**\nEither it's 1) an error in data or 2) they considered PCR + cases as positive for the competition even if there are no visible lung opacities. I hope it's 1) because 2) doesn`t make sense for a computer vision competition.\n\n**Also annotation process is not explained what are your thoughts on that?**\nI am interested as much as you are in having more details from the host about the annotation process. Mostly about the number and expertise of annotators between train and test datasets which can create a significant bias. I am also interested to know more about how the annotators were instructed regarding the atypical and indeterminate classes.\n\n**Data page states COVID-19 patients but how can we tell apart from regular COVID-19 related pneumonia vs regular pneumonia?**\nFrom a human perspective, it's not really possible to be specific. The, so called, typical appearance of Covid-19 grossly means the presence of multiple bilateral peripherical lung opacities. This is somewhat non-specific with many other pathological conditions that can mimic this presentation on chest xrays. We, as radiologists, are highly biased by the context (pandemic, clinical presentation) to conclude that a case is a probable radiological presentation of a Covid-19 pneumonia. In my practice in Canada, PCR confirmation is always needed. Chest xray in the current pandemic context is mostly used to exclude other treatable thoracic problems and to have some sense of the disease severity which is usually very well correlated with the clinical condition of the patient.\n\n **Is this challenge indeed specific to COVID cases?**\nMost probably the inclusion criteria for positive cases was a PCR positive test with a chest xray from different institutions. This hypothesis could also be confirmed by the host. They probably included random control-negative cases from the same institutions.", "votes": null }, { "id": "1314240", "postDate": "05/19/2021 03:55:02", "content": "

I understood the three Appearance columns to relate to the presence of pneumonia, not to the entire image.

\n

If Negative for Pneumonia == 0 (pneumonia is present), then one of the three Appearance columns will be set to 1 to describe it.

\n

The instructions say Studies in the test set may contain more than one label., but I don't see any rows in the train data with more than one label.

\n

The sample submission file shows a row with multiple labels, each with a confidence of 1. These two seem mutually exclusive.

\n

2b95d54e4be66_study,indeterminate 1 0 0 1 1 atypical 1 0 0 1 1

\n

The model is 100% sure that it can't determine that the image is 100% atypical ?!?

\n

These should work though, unless I grossly misunderstand it ..

\n

negative 1 0 0 1 1 typical 1 0 0 1 or negative 0 0 0 1 1 atypical 1 0 0 0 1 1

", "rawMarkdown": "I understood the three Appearance columns to relate to the presence of pneumonia, not to the entire image.\n\nIf `Negative for Pneumonia == 0 ` (pneumonia is present), then one of the three Appearance columns will be set to 1 to describe it.\n\nThe instructions say `Studies in the test set may contain more than one label.`, but I don't see any rows in the train data with more than one label.\n\nThe sample submission file shows a row with multiple labels, each with a confidence of 1. These two seem mutually exclusive. \n\n`2b95d54e4be66_study,indeterminate 1 0 0 1 1 atypical 1 0 0 1 1`\n\nThe model is 100% sure that it can't determine that the image is 100% atypical ?!?\n\nThese should work though, unless I grossly misunderstand it ..\n\n`negative 1 0 0 1 1 typical 1 0 0 1` or `negative 0 0 0 1 1 atypical 1 0 0 0 1 1`", "votes": null }, { "id": "1314440", "postDate": "05/19/2021 07:08:29", "content": "

For train, one label per image as far as I know.
\nConfidence: 1.0 is just an example.
\nBecause the metric is mAP, we should considerate every possible label even if its confidence is low.

", "rawMarkdown": "For train, one label per image as far as I know.\nConfidence: 1.0 is just an example.\nBecause the metric is mAP, we should considerate every possible label even if its confidence is low.", "votes": null }, { "id": "1314451", "postDate": "05/19/2021 07:15:12", "content": "

So much appreciate for expert's opinion.
\nI'm also interested in annotation process because there are same images in the train data with different annotations as I discussed in other thread.

", "rawMarkdown": "So much appreciate for expert's opinion.\nI'm also interested in annotation process because there are same images in the train data with different annotations as I discussed in other thread.", "votes": null }, { "id": "1314562", "postDate": "05/19/2021 08:33:36", "content": "

I think there are some study that has taken multiple images. And if the negative for pneumonia =0 (means there is pneumonia), there is at least 1 image in the study has bbox, not necessary all images has bbox, but at least 1. If negative =1 (there is no pneumonia), all the images in the study will have no bbox.

", "rawMarkdown": "I think there are some study that has taken multiple images. And if the negative for pneumonia =0 (means there is pneumonia), there is at least 1 image in the study has bbox, not necessary all images has bbox, but at least 1. If negative =1 (there is no pneumonia), all the images in the study will have no bbox.", "votes": null }, { "id": "1314606", "postDate": "05/19/2021 09:07:42", "content": "

I wish at least one BBox for the case, but nothing according to the csv.

", "rawMarkdown": "I wish at least one BBox for the case, but nothing according to the csv.", "votes": null }, { "id": "1315047", "postDate": "05/19/2021 14:17:50", "content": "

This explanation from the host on the annotation grading methodology should help answer your questions here.

", "rawMarkdown": "[This explanation from the host](https://www.kaggle.com/c/siim-covid19-detection/discussion/240250) on the annotation grading methodology should help answer your questions here.", "votes": null }, { "id": "1315052", "postDate": "05/19/2021 14:24:37", "content": "

Thank you!
\nNow I fully understand.

", "rawMarkdown": "Thank you!\nNow I fully understand.", "votes": null }, { "id": "1321483", "postDate": "05/24/2021 17:53:06", "content": "

The annotators were asked to only place bounding boxes on lung opacities or consolidations, but not on pleural effusions or pneumothoraces. In the bottom image above, there are bilateral pleural effusions, so no bounding boxes were created. For the top image, \"indeterminate appearance\" is likely appropriate since I see right lung opacities as well (although would need to look at the original full resolution image to best categorize this one).

", "rawMarkdown": "The annotators were asked to only place bounding boxes on lung opacities or consolidations, but not on pleural effusions or pneumothoraces. In the bottom image above, there are bilateral pleural effusions, so no bounding boxes were created. For the top image, \"indeterminate appearance\" is likely appropriate since I see right lung opacities as well (although would need to look at the original full resolution image to best categorize this one).", "votes": null }, { "id": "1321490", "postDate": "05/24/2021 17:57:27", "content": "

Alex, thank you for pointing that out. I agree, for the top image, \"indeterminate appearance\" is likely appropriate since I see right lung opacities as well (although would need to look at the original full resolution image to best categorize this one). For the bottom image, \"atypical appearance\" seems most appropriate for pleural effusions being present.

", "rawMarkdown": "Alex, thank you for pointing that out. I agree, for the top image, \"indeterminate appearance\" is likely appropriate since I see right lung opacities as well (although would need to look at the original full resolution image to best categorize this one). For the bottom image, \"atypical appearance\" seems most appropriate for pleural effusions being present.", "votes": null }, { "id": "1348692", "postDate": "06/14/2021 07:29:19", "content": "

@paras42 by lung opacity you mean opacity caused by Covid infection not by any other viral or bacterial ?

", "rawMarkdown": "paras42 by lung opacity you mean opacity caused by Covid infection not by any other viral or bacterial ?", "votes": null } ], "comments": [ { "id": 1313413, "author_name": "keremt", "author_url": "", "post_date": "05/18/2021 14:49:04", "content": "

Yes, there are 82 non-negative studies without any bounding boxes. 81/82 areatypical and 1 is typical. It would be nice to have more explanation on this.

\n

Also, interestingly all negative studies don't have any image with a bounding box, which might mean that there are no cases of opacity which can be labeled as negative at least thats the case with our training dataset.

", "votes": null, "replies": [ { "id": 1313416, "author_name": "drtausamaru", "author_url": "", "post_date": "05/18/2021 14:51:33", "content": "

Yeah, 81+1 is the same as my EDA.
\nOne possibility is noisy label…

", "votes": null, "replies": [] } ] }, { "id": 1313631, "author_name": "alexandrecc", "author_url": "", "post_date": "05/18/2021 16:28:34", "content": "

Based on the images, we could arguably say that these 2 cases have lung opacities.

", "votes": null, "replies": [ { "id": 1313797, "author_name": "drtausamaru", "author_url": "", "post_date": "05/18/2021 18:11:23", "content": "

Yes. Waiting for the official answer.

", "votes": null, "replies": [] }, { "id": 1313906, "author_name": "keremt", "author_url": "", "post_date": "05/18/2021 19:33:33", "content": "

@alexandrecc What is your intuition about no finding cases? Also annotation process is not explained what are your thoughts on that? Data page states COVID-19 patients but how can we tell apart from regular COVID-19 related pneumonia vs regular pneumonia? Is this challenge indeed specific to COVID cases? Sorry for bombardment, but couldn't resist after seeing you here in the discussion :D

", "votes": null, "replies": [ { "id": 1313958, "author_name": "alexandrecc", "author_url": "", "post_date": "05/18/2021 20:29:10", "content": "

What is your intuition about no finding cases?
\nEither it's 1) an error in data or 2) they considered PCR + cases as positive for the competition even if there are no visible lung opacities. I hope it's 1) because 2) doesn`t make sense for a computer vision competition.

\n

Also annotation process is not explained what are your thoughts on that?
\nI am interested as much as you are in having more details from the host about the annotation process. Mostly about the number and expertise of annotators between train and test datasets which can create a significant bias. I am also interested to know more about how the annotators were instructed regarding the atypical and indeterminate classes.

\n

Data page states COVID-19 patients but how can we tell apart from regular COVID-19 related pneumonia vs regular pneumonia?
\nFrom a human perspective, it's not really possible to be specific. The, so called, typical appearance of Covid-19 grossly means the presence of multiple bilateral peripherical lung opacities. This is somewhat non-specific with many other pathological conditions that can mimic this presentation on chest xrays. We, as radiologists, are highly biased by the context (pandemic, clinical presentation) to conclude that a case is a probable radiological presentation of a Covid-19 pneumonia. In my practice in Canada, PCR confirmation is always needed. Chest xray in the current pandemic context is mostly used to exclude other treatable thoracic problems and to have some sense of the disease severity which is usually very well correlated with the clinical condition of the patient.

\n

Is this challenge indeed specific to COVID cases?
\nMost probably the inclusion criteria for positive cases was a PCR positive test with a chest xray from different institutions. This hypothesis could also be confirmed by the host. They probably included random control-negative cases from the same institutions.

", "votes": null, "replies": [] } ] }, { "id": 1314451, "author_name": "drtausamaru", "author_url": "", "post_date": "05/19/2021 07:15:12", "content": "

So much appreciate for expert's opinion.
\nI'm also interested in annotation process because there are same images in the train data with different annotations as I discussed in other thread.

", "votes": null, "replies": [] }, { "id": 1321490, "author_name": "paras42", "author_url": "", "post_date": "05/24/2021 17:57:27", "content": "

Alex, thank you for pointing that out. I agree, for the top image, \"indeterminate appearance\" is likely appropriate since I see right lung opacities as well (although would need to look at the original full resolution image to best categorize this one). For the bottom image, \"atypical appearance\" seems most appropriate for pleural effusions being present.

", "votes": null, "replies": [] } ] }, { "id": 1314240, "author_name": "davidbroberts", "author_url": "", "post_date": "05/19/2021 03:55:02", "content": "

I understood the three Appearance columns to relate to the presence of pneumonia, not to the entire image.

\n

If Negative for Pneumonia == 0 (pneumonia is present), then one of the three Appearance columns will be set to 1 to describe it.

\n

The instructions say Studies in the test set may contain more than one label., but I don't see any rows in the train data with more than one label.

\n

The sample submission file shows a row with multiple labels, each with a confidence of 1. These two seem mutually exclusive.

\n

2b95d54e4be66_study,indeterminate 1 0 0 1 1 atypical 1 0 0 1 1

\n

The model is 100% sure that it can't determine that the image is 100% atypical ?!?

\n

These should work though, unless I grossly misunderstand it ..

\n

negative 1 0 0 1 1 typical 1 0 0 1 or negative 0 0 0 1 1 atypical 1 0 0 0 1 1

", "votes": null, "replies": [ { "id": 1314440, "author_name": "drtausamaru", "author_url": "", "post_date": "05/19/2021 07:08:29", "content": "

For train, one label per image as far as I know.
\nConfidence: 1.0 is just an example.
\nBecause the metric is mAP, we should considerate every possible label even if its confidence is low.

", "votes": null, "replies": [] } ] }, { "id": 1314562, "author_name": "leezhixiong", "author_url": "", "post_date": "05/19/2021 08:33:36", "content": "

I think there are some study that has taken multiple images. And if the negative for pneumonia =0 (means there is pneumonia), there is at least 1 image in the study has bbox, not necessary all images has bbox, but at least 1. If negative =1 (there is no pneumonia), all the images in the study will have no bbox.

", "votes": null, "replies": [ { "id": 1314606, "author_name": "drtausamaru", "author_url": "", "post_date": "05/19/2021 09:07:42", "content": "

I wish at least one BBox for the case, but nothing according to the csv.

", "votes": null, "replies": [] } ] }, { "id": 1315047, "author_name": "juliaelliott", "author_url": "", "post_date": "05/19/2021 14:17:50", "content": "

This explanation from the host on the annotation grading methodology should help answer your questions here.

", "votes": null, "replies": [ { "id": 1315052, "author_name": "drtausamaru", "author_url": "", "post_date": "05/19/2021 14:24:37", "content": "

Thank you!
\nNow I fully understand.

", "votes": null, "replies": [] } ] }, { "id": 1321483, "author_name": "paras42", "author_url": "", "post_date": "05/24/2021 17:53:06", "content": "

The annotators were asked to only place bounding boxes on lung opacities or consolidations, but not on pleural effusions or pneumothoraces. In the bottom image above, there are bilateral pleural effusions, so no bounding boxes were created. For the top image, \"indeterminate appearance\" is likely appropriate since I see right lung opacities as well (although would need to look at the original full resolution image to best categorize this one).

", "votes": null, "replies": [ { "id": 1348692, "author_name": "jaideepvalani", "author_url": "", "post_date": "06/14/2021 07:29:19", "content": "

@paras42 by lung opacity you mean opacity caused by Covid infection not by any other viral or bacterial ?

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1313243": "There are some ids where \"Negative for Pneumonia\"==0 without any BBox.\nFor example, id: 50804b6043ac_image, 51759b5579bc_image, ....\n\"Screen\n\"Screen\nI guess they are abnormal (pathological) but no BBox information.\nIf BBoxes only for pneumonia, their \"Negative for Pneumonia\" should be 1.\nIf BBoxes also for abnormalities including pneumonia, it is possible that their \"Negative for Pneumonia\"==0, but we need BBox information.\nOr they have pneumonia, but not because of COVID-19?\nDo you have any idea? 😂", "1313413": "Yes, there are 82 non-negative studies without any bounding boxes. 81/82 are`atypical` and 1 is `typical`. It would be nice to have more explanation on this.\n\nAlso, interestingly all negative studies don't have any image with a bounding box, which might mean that there are no cases of opacity which can be labeled as `negative` at least thats the case with our training dataset.", "1313416": "Yeah, 81+1 is the same as my EDA.\nOne possibility is noisy label...", "1313631": "Based on the images, we could arguably say that these 2 cases have lung opacities.", "1313797": "Yes. Waiting for the official answer.", "1313906": "alexandrecc What is your intuition about no finding cases? Also annotation process is not explained what are your thoughts on that? Data page states COVID-19 patients but how can we tell apart from regular COVID-19 related pneumonia vs regular pneumonia? Is this challenge indeed specific to COVID cases? Sorry for bombardment, but couldn't resist after seeing you here in the discussion :D", "1313958": "**What is your intuition about no finding cases?**\nEither it's 1) an error in data or 2) they considered PCR + cases as positive for the competition even if there are no visible lung opacities. I hope it's 1) because 2) doesn`t make sense for a computer vision competition.\n\n**Also annotation process is not explained what are your thoughts on that?**\nI am interested as much as you are in having more details from the host about the annotation process. Mostly about the number and expertise of annotators between train and test datasets which can create a significant bias. I am also interested to know more about how the annotators were instructed regarding the atypical and indeterminate classes.\n\n**Data page states COVID-19 patients but how can we tell apart from regular COVID-19 related pneumonia vs regular pneumonia?**\nFrom a human perspective, it's not really possible to be specific. The, so called, typical appearance of Covid-19 grossly means the presence of multiple bilateral peripherical lung opacities. This is somewhat non-specific with many other pathological conditions that can mimic this presentation on chest xrays. We, as radiologists, are highly biased by the context (pandemic, clinical presentation) to conclude that a case is a probable radiological presentation of a Covid-19 pneumonia. In my practice in Canada, PCR confirmation is always needed. Chest xray in the current pandemic context is mostly used to exclude other treatable thoracic problems and to have some sense of the disease severity which is usually very well correlated with the clinical condition of the patient.\n\n **Is this challenge indeed specific to COVID cases?**\nMost probably the inclusion criteria for positive cases was a PCR positive test with a chest xray from different institutions. This hypothesis could also be confirmed by the host. They probably included random control-negative cases from the same institutions.", "1314240": "I understood the three Appearance columns to relate to the presence of pneumonia, not to the entire image.\n\nIf `Negative for Pneumonia == 0 ` (pneumonia is present), then one of the three Appearance columns will be set to 1 to describe it.\n\nThe instructions say `Studies in the test set may contain more than one label.`, but I don't see any rows in the train data with more than one label.\n\nThe sample submission file shows a row with multiple labels, each with a confidence of 1. These two seem mutually exclusive. \n\n`2b95d54e4be66_study,indeterminate 1 0 0 1 1 atypical 1 0 0 1 1`\n\nThe model is 100% sure that it can't determine that the image is 100% atypical ?!?\n\nThese should work though, unless I grossly misunderstand it ..\n\n`negative 1 0 0 1 1 typical 1 0 0 1` or `negative 0 0 0 1 1 atypical 1 0 0 0 1 1`", "1314440": "For train, one label per image as far as I know.\nConfidence: 1.0 is just an example.\nBecause the metric is mAP, we should considerate every possible label even if its confidence is low.", "1314451": "So much appreciate for expert's opinion.\nI'm also interested in annotation process because there are same images in the train data with different annotations as I discussed in other thread.", "1314562": "I think there are some study that has taken multiple images. And if the negative for pneumonia =0 (means there is pneumonia), there is at least 1 image in the study has bbox, not necessary all images has bbox, but at least 1. If negative =1 (there is no pneumonia), all the images in the study will have no bbox.", "1314606": "I wish at least one BBox for the case, but nothing according to the csv.", "1315047": "[This explanation from the host](https://www.kaggle.com/c/siim-covid19-detection/discussion/240250) on the annotation grading methodology should help answer your questions here.", "1315052": "Thank you!\nNow I fully understand.", "1321483": "The annotators were asked to only place bounding boxes on lung opacities or consolidations, but not on pleural effusions or pneumothoraces. In the bottom image above, there are bilateral pleural effusions, so no bounding boxes were created. For the top image, \"indeterminate appearance\" is likely appropriate since I see right lung opacities as well (although would need to look at the original full resolution image to best categorize this one).", "1321490": "Alex, thank you for pointing that out. I agree, for the top image, \"indeterminate appearance\" is likely appropriate since I see right lung opacities as well (although would need to look at the original full resolution image to best categorize this one). For the bottom image, \"atypical appearance\" seems most appropriate for pleural effusions being present.", "1348692": "paras42 by lung opacity you mean opacity caused by Covid infection not by any other viral or bacterial ?" }, "source": "meta" }