I'm having a hard time answering this myself from looking at visualizations of the data, and was wondering if anybody could chime in.
\n\nThe data description page mentions that "Each slice is acquired on a separate breath hold. This is important since the registration from slice to slice is expected to be imperfect."
\n\nVisually, for a given study, it looks like each series of frames has been captured (or ordered/shifted) such that the nth (say, the 5th) frame of each slice represents an image capture at an "equivalent time" with respect to the systole-diastole cadence.
\n\nIs the description essentially stating that this is not necessarily the case?
\n\nDoes this question make sense? I hope so!
", "messages": [ { "id": "101822", "postDate": "12/17/2015 09:08:17", "content": "I'm having a hard time answering this myself from looking at visualizations of the data, and was wondering if anybody could chime in.
\n\nThe data description page mentions that "Each slice is acquired on a separate breath hold. This is important since the registration from slice to slice is expected to be imperfect."
\n\nVisually, for a given study, it looks like each series of frames has been captured (or ordered/shifted) such that the nth (say, the 5th) frame of each slice represents an image capture at an "equivalent time" with respect to the systole-diastole cadence.
\n\nIs the description essentially stating that this is not necessarily the case?
\n\nDoes this question make sense? I hope so!
", "rawMarkdown": "I'm having a hard time answering this myself from looking at visualizations of the data, and was wondering if anybody could chime in.\r\n\r\nThe data description page mentions that \"Each slice is acquired on a separate breath hold. This is important since the registration from slice to slice is expected to be imperfect.\"\r\n\r\nVisually, for a given study, it looks like each series of frames has been captured (or ordered/shifted) such that the nth (say, the 5th) frame of *each* slice represents an image capture at an \"equivalent time\" with respect to the systole-diastole cadence.\r\n\r\nIs the description essentially stating that this is not necessarily the case?\r\n\r\nDoes this question make sense? I hope so!", "votes": null }, { "id": "101845", "postDate": "12/17/2015 13:10:06", "content": "Hi Martin. Yes, you are correct. The ideal is a perfectly coherent series, but since each slice is captured on a different breath hold, the ideal will not always be met.
", "rawMarkdown": "Hi Martin. Yes, you are correct. The ideal is a perfectly coherent series, but since each slice is captured on a different breath hold, the ideal will not always be met.", "votes": null }, { "id": "103553", "postDate": "01/04/2016 08:59:47", "content": "[quote=Shannon;101845]
\n\nHi Martin. Yes, you are correct. The ideal is a perfectly coherent series, but since each slice is captured on a different breath hold, the ideal will not always be met.
\n\n[/quote]
\n\nSo, does it mean that every slice have 30 images and any two n-th images in 2 different slices (corresponding to the same heart) are taken in different time moments?
\n\nWhere can we find the order of the slices and the distance between neighbor slices?
", "rawMarkdown": "[quote=Shannon;101845]\r\n\r\nHi Martin. Yes, you are correct. The ideal is a perfectly coherent series, but since each slice is captured on a different breath hold, the ideal will not always be met. \r\n\r\n[/quote]\r\n\r\nSo, does it mean that every slice have 30 images and any two n-th images in 2 different slices (corresponding to the same heart) are taken in different time moments?\r\n\r\nWhere can we find the order of the slices and the distance between neighbor slices?", "votes": null }, { "id": "103557", "postDate": "01/04/2016 09:52:21", "content": "Also, I would like to ask considering every slice is taken in different breath hold, does the n-th image (in series of 30) correspond to the same n-th phase of heart bit?
", "rawMarkdown": "Also, I would like to ask considering every slice is taken in different breath hold, does the n-th image (in series of 30) correspond to the same n-th phase of heart bit?", "votes": null }, { "id": "103574", "postDate": "01/04/2016 13:30:13", "content": "A few comments:
\n\nUsusally 30 images are reconstructed for each cardiac cycle. The data were acquired over multiple physical heart-beats and reconstructed on a pseudo-time axis, where the timing of each individual data element is interpolated based on an algorithm that tries to take into account that there is variation from beat to beat.
There are also variations from breath-hold to breath-hold, of course. In practice, it is not guaranteed that frame 15 in one breath-hold was acquired at exactly the same time from the R wave in the ECG signal. But we assume that it was. Errors are incurred due to this assumption, but they are believed to be small.
Keep in mind that when the length of the cardiac cycle changes, the cardiac function and volumes are likely to change a bit too. From that perspective, using multiple breath-holds that each use data from multiple cardiac cycles is imperfect, but it is the best that we can do.
To sum all of this up. Small errors are made due to heart rate variation during acquisition (in each slice and between slices). These errors are thought to be small and in standard clinical practice (and in this competition), they are ignored.
\n\nI hope this helps,
\n\nMichael
", "rawMarkdown": "A few comments:\r\n\r\n* Ususally 30 images are reconstructed for each cardiac cycle. The data were acquired over multiple physical heart-beats and reconstructed on a pseudo-time axis, where the timing of each individual data element is interpolated based on an algorithm that tries to take into account that there is variation from beat to beat. \r\n\r\n* There are also variations from breath-hold to breath-hold, of course. In practice, it is not guaranteed that frame 15 in one breath-hold was acquired at exactly the same time from the R wave in the ECG signal. But we assume that it was. Errors are incurred due to this assumption, but they are believed to be small. \r\n\r\n* Keep in mind that when the length of the cardiac cycle changes, the cardiac function and volumes are likely to change a bit too. From that perspective, using multiple breath-holds that each use data from multiple cardiac cycles is imperfect, but it is the best that we can do.\r\n\r\nTo sum all of this up. Small errors are made due to heart rate variation during acquisition (in each slice and between slices). These errors are thought to be small and in standard clinical practice (and in this competition), they are ignored. \r\n\r\nI hope this helps,\r\n\r\nMichael", "votes": null }, { "id": "103579", "postDate": "01/04/2016 14:20:46", "content": "[quote=Kanan Mammadli;103553]\nWhere can we find the order of the slices and the distance between neighbor slices?\n[/quote]
\n\nThanks for clarification that there is no 2 images taken at the same time.\nWhat about the distance between the slices?
", "rawMarkdown": "[quote=Kanan Mammadli;103553]\r\nWhere can we find the order of the slices and the distance between neighbor slices?\r\n[/quote]\r\n\r\nThanks for clarification that there is no 2 images taken at the same time.\r\nWhat about the distance between the slices?", "votes": null }, { "id": "103598", "postDate": "01/04/2016 16:31:36", "content": "I think the order and the slice distance can be derived from the metadata field: SliceLocation. It also shows that frequently Slices are repeated (probably due to irregularities in the image). Here an example (in attachment) in study 49 : (study, path, SliceLocation, difference, PatientsAge)
", "rawMarkdown": "I think the order and the slice distance can be derived from the metadata field: SliceLocation. It also shows that frequently Slices are repeated (probably due to irregularities in the image). Here an example (in attachment) in study 49 : (study, path, SliceLocation, difference, PatientsAge)", "votes": null }, { "id": "103609", "postDate": "01/04/2016 18:25:58", "content": "@DirkWillemWonnink thanks. I was just about to write that.
\n\nAlso take a look at the tutorial:
\n\nhttps://www.kaggle.com/c/second-annual-data-science-bowl/details/deep-learning-tutorial
\n\nThey use the SliceLocation field in there.
\n\nHope this helps,\nMichael
", "rawMarkdown": "DirkWillemWonnink thanks. I was just about to write that. \r\n\r\nAlso take a look at the tutorial:\r\n\r\nhttps://www.kaggle.com/c/second-annual-data-science-bowl/details/deep-learning-tutorial\r\n\r\nThey use the SliceLocation field in there.\r\n\r\nHope this helps,\r\nMichael", "votes": null } ], "comments": [ { "id": 101845, "author_name": "shannonlantzy", "author_url": "", "post_date": "12/17/2015 13:10:06", "content": "Hi Martin. Yes, you are correct. The ideal is a perfectly coherent series, but since each slice is captured on a different breath hold, the ideal will not always be met.
", "votes": null, "replies": [] }, { "id": 103553, "author_name": "kananmammadli", "author_url": "", "post_date": "01/04/2016 08:59:47", "content": "[quote=Shannon;101845]
\n\nHi Martin. Yes, you are correct. The ideal is a perfectly coherent series, but since each slice is captured on a different breath hold, the ideal will not always be met.
\n\n[/quote]
\n\nSo, does it mean that every slice have 30 images and any two n-th images in 2 different slices (corresponding to the same heart) are taken in different time moments?
\n\nWhere can we find the order of the slices and the distance between neighbor slices?
", "votes": null, "replies": [] }, { "id": 103557, "author_name": "kananmammadli", "author_url": "", "post_date": "01/04/2016 09:52:21", "content": "Also, I would like to ask considering every slice is taken in different breath hold, does the n-th image (in series of 30) correspond to the same n-th phase of heart bit?
", "votes": null, "replies": [] }, { "id": 103574, "author_name": "michaelhansen", "author_url": "", "post_date": "01/04/2016 13:30:13", "content": "A few comments:
\n\nUsusally 30 images are reconstructed for each cardiac cycle. The data were acquired over multiple physical heart-beats and reconstructed on a pseudo-time axis, where the timing of each individual data element is interpolated based on an algorithm that tries to take into account that there is variation from beat to beat.
There are also variations from breath-hold to breath-hold, of course. In practice, it is not guaranteed that frame 15 in one breath-hold was acquired at exactly the same time from the R wave in the ECG signal. But we assume that it was. Errors are incurred due to this assumption, but they are believed to be small.
Keep in mind that when the length of the cardiac cycle changes, the cardiac function and volumes are likely to change a bit too. From that perspective, using multiple breath-holds that each use data from multiple cardiac cycles is imperfect, but it is the best that we can do.
To sum all of this up. Small errors are made due to heart rate variation during acquisition (in each slice and between slices). These errors are thought to be small and in standard clinical practice (and in this competition), they are ignored.
\n\nI hope this helps,
\n\nMichael
", "votes": null, "replies": [] }, { "id": 103579, "author_name": "kananmammadli", "author_url": "", "post_date": "01/04/2016 14:20:46", "content": "[quote=Kanan Mammadli;103553]\nWhere can we find the order of the slices and the distance between neighbor slices?\n[/quote]
\n\nThanks for clarification that there is no 2 images taken at the same time.\nWhat about the distance between the slices?
", "votes": null, "replies": [] }, { "id": 103598, "author_name": "diwiwo", "author_url": "", "post_date": "01/04/2016 16:31:36", "content": "I think the order and the slice distance can be derived from the metadata field: SliceLocation. It also shows that frequently Slices are repeated (probably due to irregularities in the image). Here an example (in attachment) in study 49 : (study, path, SliceLocation, difference, PatientsAge)
", "votes": null, "replies": [] }, { "id": 103609, "author_name": "michaelhansen", "author_url": "", "post_date": "01/04/2016 18:25:58", "content": "@DirkWillemWonnink thanks. I was just about to write that.
\n\nAlso take a look at the tutorial:
\n\nhttps://www.kaggle.com/c/second-annual-data-science-bowl/details/deep-learning-tutorial
\n\nThey use the SliceLocation field in there.
\n\nHope this helps,\nMichael
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "101822": "I'm having a hard time answering this myself from looking at visualizations of the data, and was wondering if anybody could chime in.\r\n\r\nThe data description page mentions that \"Each slice is acquired on a separate breath hold. This is important since the registration from slice to slice is expected to be imperfect.\"\r\n\r\nVisually, for a given study, it looks like each series of frames has been captured (or ordered/shifted) such that the nth (say, the 5th) frame of *each* slice represents an image capture at an \"equivalent time\" with respect to the systole-diastole cadence.\r\n\r\nIs the description essentially stating that this is not necessarily the case?\r\n\r\nDoes this question make sense? I hope so!", "101845": "Hi Martin. Yes, you are correct. The ideal is a perfectly coherent series, but since each slice is captured on a different breath hold, the ideal will not always be met.", "103553": "[quote=Shannon;101845]\r\n\r\nHi Martin. Yes, you are correct. The ideal is a perfectly coherent series, but since each slice is captured on a different breath hold, the ideal will not always be met. \r\n\r\n[/quote]\r\n\r\nSo, does it mean that every slice have 30 images and any two n-th images in 2 different slices (corresponding to the same heart) are taken in different time moments?\r\n\r\nWhere can we find the order of the slices and the distance between neighbor slices?", "103557": "Also, I would like to ask considering every slice is taken in different breath hold, does the n-th image (in series of 30) correspond to the same n-th phase of heart bit?", "103574": "A few comments:\r\n\r\n* Ususally 30 images are reconstructed for each cardiac cycle. The data were acquired over multiple physical heart-beats and reconstructed on a pseudo-time axis, where the timing of each individual data element is interpolated based on an algorithm that tries to take into account that there is variation from beat to beat. \r\n\r\n* There are also variations from breath-hold to breath-hold, of course. In practice, it is not guaranteed that frame 15 in one breath-hold was acquired at exactly the same time from the R wave in the ECG signal. But we assume that it was. Errors are incurred due to this assumption, but they are believed to be small. \r\n\r\n* Keep in mind that when the length of the cardiac cycle changes, the cardiac function and volumes are likely to change a bit too. From that perspective, using multiple breath-holds that each use data from multiple cardiac cycles is imperfect, but it is the best that we can do.\r\n\r\nTo sum all of this up. Small errors are made due to heart rate variation during acquisition (in each slice and between slices). These errors are thought to be small and in standard clinical practice (and in this competition), they are ignored. \r\n\r\nI hope this helps,\r\n\r\nMichael", "103579": "[quote=Kanan Mammadli;103553]\r\nWhere can we find the order of the slices and the distance between neighbor slices?\r\n[/quote]\r\n\r\nThanks for clarification that there is no 2 images taken at the same time.\r\nWhat about the distance between the slices?", "103598": "I think the order and the slice distance can be derived from the metadata field: SliceLocation. It also shows that frequently Slices are repeated (probably due to irregularities in the image). Here an example (in attachment) in study 49 : (study, path, SliceLocation, difference, PatientsAge)", "103609": "DirkWillemWonnink thanks. I was just about to write that. \r\n\r\nAlso take a look at the tutorial:\r\n\r\nhttps://www.kaggle.com/c/second-annual-data-science-bowl/details/deep-learning-tutorial\r\n\r\nThey use the SliceLocation field in there.\r\n\r\nHope this helps,\r\nMichael" }, "source": "meta" }