I understand I cannot demand things here, but it would be good to have a few consecutive scans in the test data. The reason for it is that you usually need a sequence to understand the pattern….
\nE.g. if I say guess number in this sequence: “1, 3, 5, 7…”. Clearly it’s 9. But what if I say guess number in this sequence: “13”.
\nYou will not be able to guess it because there is no pattern. Same here - you will not be able to make an accurate prediction by having only one scan. It’s important to understand the dynamics.
", "messages": [ { "id": "974325", "postDate": "08/17/2020 22:16:42", "content": "I understand I cannot demand things here, but it would be good to have a few consecutive scans in the test data. The reason for it is that you usually need a sequence to understand the pattern….
\nE.g. if I say guess number in this sequence: “1, 3, 5, 7…”. Clearly it’s 9. But what if I say guess number in this sequence: “13”.
\nYou will not be able to guess it because there is no pattern. Same here - you will not be able to make an accurate prediction by having only one scan. It’s important to understand the dynamics.
", "rawMarkdown": "I understand I cannot demand things here, but it would be good to have a few consecutive scans in the test data. The reason for it is that you usually need a sequence to understand the pattern....\n\nE.g. if I say guess number in this sequence: “1, 3, 5, 7...”. Clearly it’s 9. But what if I say guess number in this sequence: “13”. \n\nYou will not be able to guess it because there is no pattern. Same here - you will not be able to make an accurate prediction by having only one scan. It’s important to understand the dynamics.", "votes": null }, { "id": "974334", "postDate": "08/17/2020 22:40:57", "content": "Any thoughts on this?
", "rawMarkdown": "Any thoughts on this?", "votes": null }, { "id": "974335", "postDate": "08/17/2020 22:42:02", "content": "Do you mean within the single CT study provided? Or are you saying you wish there were additional studies over the course of the weeks?
", "rawMarkdown": "Do you mean within the single CT study provided? Or are you saying you wish there were additional studies over the course of the weeks?", "votes": null }, { "id": "974343", "postDate": "08/17/2020 23:08:56", "content": "Yes, exactly.
", "rawMarkdown": "Yes, exactly.", "votes": null }, { "id": "974355", "postDate": "08/17/2020 23:37:05", "content": "In clinical practice you would not expect a lot of repeat CT scans for a patient. So, it makes sense that you only have one CT scan to base your prediction off of. Hopefully we can use CT scans from multiple patients to determine the course of the disease.
", "rawMarkdown": "In clinical practice you would not expect a lot of repeat CT scans for a patient. So, it makes sense that you only have one CT scan to base your prediction off of. Hopefully we can use CT scans from multiple patients to determine the course of the disease.", "votes": null }, { "id": "974358", "postDate": "08/17/2020 23:42:00", "content": "I think to assume that more CTs would provide the answer is an over-simplification of the problem. The target measure doesn't correlate directly with what you'd see on imaging. The airspace that you might see on imaging can't be directly calculated because of other factors such as dead space (air in the lungs that is not involved in gas exchange), the fact that the imaging is basically static while the lungs themselves are dynamic and can inflate (or collapse) due to multiple factors with time such as underlying pathology that affects the compliance (elasticity) of the lungs, atelectasis (collapsing of the alveola due to plugging of the airways), pulmonary toilette, etc.
\nThere are some basic things that we're not given in this competition, like the FEV1 values. Usually you calculate the FEV1/FVC ratios to determine the pattern of disease (such as whether this is a restrictive or obstructive type pattern). It seems implausible that the competition organizers don't have that data, so I'm guessing that they felt it might be a source of data leak, and perhaps are hoping that this competition can serve as a means of squeezing as much of that information from the scans they have provided.
\nAs some of the others have pointed out, there are very expensive medications that can potentially help, but not cure this disease, so it'd be nice to find out who might need it sooner than later. Also, it'd be nice to get as much information from the initial CT scan since they can also be very expensive (thousands of dollars depending on what type of insurance coverage you have, at least here in the US).
", "rawMarkdown": "I think to assume that more CTs would provide the answer is an over-simplification of the problem. The target measure doesn't correlate directly with what you'd see on imaging. The airspace that you might see on imaging can't be directly calculated because of other factors such as dead space (air in the lungs that is not involved in gas exchange), the fact that the imaging is basically static while the lungs themselves are dynamic and can inflate (or collapse) due to multiple factors with time such as underlying pathology that affects the compliance (elasticity) of the lungs, atelectasis (collapsing of the alveola due to plugging of the airways), pulmonary toilette, etc.\n\nThere are some basic things that we're not given in this competition, like the FEV1 values. Usually you calculate the FEV1/FVC ratios to determine the pattern of disease (such as whether this is a restrictive or obstructive type pattern). It seems implausible that the competition organizers don't have that data, so I'm guessing that they felt it might be a source of data leak, and perhaps are hoping that this competition can serve as a means of squeezing as much of that information from the scans they have provided. \n\nAs some of the others have pointed out, there are very expensive medications that can potentially help, but not cure this disease, so it'd be nice to find out who might need it sooner than later. Also, it'd be nice to get as much information from the initial CT scan since they can also be very expensive (thousands of dollars depending on what type of insurance coverage you have, at least here in the US).", "votes": null }, { "id": "974387", "postDate": "08/18/2020 00:03:20", "content": "Oh, sorry for confusion, I actually meant the FVC measurement, however, the CT scan would be amazing too.
\nIt says, that in test data, there is only one FVC scan.
", "rawMarkdown": "Oh, sorry for confusion, I actually meant the FVC measurement, however, the CT scan would be amazing too.\n\nIt says, that in test data, there is only one FVC scan.", "votes": null } ], "comments": [ { "id": 974334, "author_name": "andrisole92", "author_url": "", "post_date": "08/17/2020 22:40:57", "content": "Any thoughts on this?
", "votes": null, "replies": [] }, { "id": 974335, "author_name": "jjinho", "author_url": "", "post_date": "08/17/2020 22:42:02", "content": "Do you mean within the single CT study provided? Or are you saying you wish there were additional studies over the course of the weeks?
", "votes": null, "replies": [] }, { "id": 974343, "author_name": "andrisole92", "author_url": "", "post_date": "08/17/2020 23:08:56", "content": "Yes, exactly.
", "votes": null, "replies": [] }, { "id": 974355, "author_name": "richardepstein", "author_url": "", "post_date": "08/17/2020 23:37:05", "content": "In clinical practice you would not expect a lot of repeat CT scans for a patient. So, it makes sense that you only have one CT scan to base your prediction off of. Hopefully we can use CT scans from multiple patients to determine the course of the disease.
", "votes": null, "replies": [] }, { "id": 974358, "author_name": "jjinho", "author_url": "", "post_date": "08/17/2020 23:42:00", "content": "I think to assume that more CTs would provide the answer is an over-simplification of the problem. The target measure doesn't correlate directly with what you'd see on imaging. The airspace that you might see on imaging can't be directly calculated because of other factors such as dead space (air in the lungs that is not involved in gas exchange), the fact that the imaging is basically static while the lungs themselves are dynamic and can inflate (or collapse) due to multiple factors with time such as underlying pathology that affects the compliance (elasticity) of the lungs, atelectasis (collapsing of the alveola due to plugging of the airways), pulmonary toilette, etc.
\nThere are some basic things that we're not given in this competition, like the FEV1 values. Usually you calculate the FEV1/FVC ratios to determine the pattern of disease (such as whether this is a restrictive or obstructive type pattern). It seems implausible that the competition organizers don't have that data, so I'm guessing that they felt it might be a source of data leak, and perhaps are hoping that this competition can serve as a means of squeezing as much of that information from the scans they have provided.
\nAs some of the others have pointed out, there are very expensive medications that can potentially help, but not cure this disease, so it'd be nice to find out who might need it sooner than later. Also, it'd be nice to get as much information from the initial CT scan since they can also be very expensive (thousands of dollars depending on what type of insurance coverage you have, at least here in the US).
", "votes": null, "replies": [] }, { "id": 974387, "author_name": "andrisole92", "author_url": "", "post_date": "08/18/2020 00:03:20", "content": "Oh, sorry for confusion, I actually meant the FVC measurement, however, the CT scan would be amazing too.
\nIt says, that in test data, there is only one FVC scan.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "974325": "I understand I cannot demand things here, but it would be good to have a few consecutive scans in the test data. The reason for it is that you usually need a sequence to understand the pattern....\n\nE.g. if I say guess number in this sequence: “1, 3, 5, 7...”. Clearly it’s 9. But what if I say guess number in this sequence: “13”. \n\nYou will not be able to guess it because there is no pattern. Same here - you will not be able to make an accurate prediction by having only one scan. It’s important to understand the dynamics.", "974334": "Any thoughts on this?", "974335": "Do you mean within the single CT study provided? Or are you saying you wish there were additional studies over the course of the weeks?", "974343": "Yes, exactly.", "974355": "In clinical practice you would not expect a lot of repeat CT scans for a patient. So, it makes sense that you only have one CT scan to base your prediction off of. Hopefully we can use CT scans from multiple patients to determine the course of the disease.", "974358": "I think to assume that more CTs would provide the answer is an over-simplification of the problem. The target measure doesn't correlate directly with what you'd see on imaging. The airspace that you might see on imaging can't be directly calculated because of other factors such as dead space (air in the lungs that is not involved in gas exchange), the fact that the imaging is basically static while the lungs themselves are dynamic and can inflate (or collapse) due to multiple factors with time such as underlying pathology that affects the compliance (elasticity) of the lungs, atelectasis (collapsing of the alveola due to plugging of the airways), pulmonary toilette, etc.\n\nThere are some basic things that we're not given in this competition, like the FEV1 values. Usually you calculate the FEV1/FVC ratios to determine the pattern of disease (such as whether this is a restrictive or obstructive type pattern). It seems implausible that the competition organizers don't have that data, so I'm guessing that they felt it might be a source of data leak, and perhaps are hoping that this competition can serve as a means of squeezing as much of that information from the scans they have provided. \n\nAs some of the others have pointed out, there are very expensive medications that can potentially help, but not cure this disease, so it'd be nice to find out who might need it sooner than later. Also, it'd be nice to get as much information from the initial CT scan since they can also be very expensive (thousands of dollars depending on what type of insurance coverage you have, at least here in the US).", "974387": "Oh, sorry for confusion, I actually meant the FVC measurement, however, the CT scan would be amazing too.\n\nIt says, that in test data, there is only one FVC scan." }, "source": "meta" }