is it labeled with a red circle by hand and then measured by human (then it is area instead of volume)? so we should expect error on these data?
", "messages": [ { "id": "102499", "postDate": "12/23/2015 04:13:35", "content": "is it labeled with a red circle by hand and then measured by human (then it is area instead of volume)? so we should expect error on these data?
", "rawMarkdown": "is it labeled with a red circle by hand and then measured by human (then it is area instead of volume)? so we should expect error on these data?", "votes": null }, { "id": "102546", "postDate": "12/23/2015 15:01:18", "content": "Hi woshialex,
\n\nA cardiologist draws contours (circles) at end systole and end diastole for each slice in the short axis stack, and a program is used to calculate the volumes.
\n\nThe video walk-through of the MRIs may be helpful (posted on the Resources page, and linked here).
", "rawMarkdown": "Hi woshialex,\r\n\r\nA cardiologist draws contours (circles) at end systole and end diastole for each slice in the short axis stack, and a program is used to calculate the volumes.\r\n\r\nThe video walk-through of the MRIs may be helpful (posted on the [Resources][1] page, and linked [here][2]).\r\n\r\n\r\n [1]: https://www.kaggle.com/c/second-annual-data-science-bowl/details/resources\r\n [2]: https://youtu.be/dFu_5T0ODrM", "votes": null }, { "id": "102547", "postDate": "12/23/2015 15:01:26", "content": "Yes, they are hand-drawn circles which are combined into a volume. We show the process in the tutorial video. As with any real data set, you should expect errors both large and small, systematic and spurious, believable and unbelievable.
", "rawMarkdown": "Yes, they are hand-drawn circles which are combined into a volume. We show the process in the [tutorial video][1]. As with any real data set, you should expect errors both large and small, systematic and spurious, believable and unbelievable.\r\n\r\n [1]: https://youtu.be/dFu_5T0ODrM", "votes": null }, { "id": "103253", "postDate": "12/30/2015 19:17:29", "content": "I would like to confirm if each volume in the training dataset corresponds to each sax folder/patient or is it the total volume comprising all folders/patient ?
", "rawMarkdown": "I would like to confirm if each volume in the training dataset corresponds to each sax folder/patient or is it the total volume comprising all folders/patient ?", "votes": null }, { "id": "103514", "postDate": "01/03/2016 22:36:14", "content": "In the tutorial video I don't think anything is mentioned about the long axis measurement of the LV and the influence of that on the calculated volumes. Is that taken into account and if yes, how? Or is it f.e. covered by assuming that the apex - slice reaches a zero surface on systole?
", "rawMarkdown": "In the tutorial video I don't think anything is mentioned about the long axis measurement of the LV and the influence of that on the calculated volumes. Is that taken into account and if yes, how? Or is it f.e. covered by assuming that the apex - slice reaches a zero surface on systole?", "votes": null }, { "id": "103527", "postDate": "01/04/2016 00:31:36", "content": "@DiekWillenWonnick, I am not sure that I understand your question. The long axis views of the LV just provide a different view of the ventricle compared to the short axis views, but it is a view of the same ventricle. So your question about whether it is taken into account is a bit hard to interpret. It is another view of the same volume. The volume does not change depending on how many views you have of the volume. The long axis views are often helpful in determining the location of the valve plane, i.e., where the atrium ends and the ventricle begins.
", "rawMarkdown": "DiekWillenWonnick, I am not sure that I understand your question. The long axis views of the LV just provide a different view of the ventricle compared to the short axis views, but it is a view of the same ventricle. So your question about whether it is taken into account is a bit hard to interpret. It is another view of the same volume. The volume does not change depending on how many views you have of the volume. The long axis views are often helpful in determining the location of the valve plane, i.e., where the atrium ends and the ventricle begins.", "votes": null }, { "id": "103549", "postDate": "01/04/2016 08:41:31", "content": "Hi Michael. Thanks for you explanation. Wrong interpretation of the volume calculation by me. The long-axis size (L in below picture ) is derived from the choice of the slices. It makes the right slice selection of the valve in systole and in diastole extra critical. (and also, but probably less, for the 'apex-slice').
\n\n", "rawMarkdown": "Hi Michael. Thanks for you explanation. Wrong interpretation of the volume calculation by me. The long-axis size (L in below picture ) is derived from the choice of the slices. It makes the right slice selection of the valve in systole and in diastole extra critical. (and also, but probably less, for the 'apex-slice'). \r\n\r\n[Simpson method][1]\r\n [1]: http://www.winfocus.org/_/rsrc/1272864420255/e-learning/echo/left_ventricle/Winfocus_CCECho_LV_Fig13.png?width=60%25", "votes": null }, { "id": "103572", "postDate": "01/04/2016 13:21:09", "content": "Hi Dirk,
\n\nYes, the selection of the valve plane in systole and diastole is where big errors are sometimes incurred. The long axis views can help with this, but it must be used carefully since it was acquired in a different breath-hold than the sax slices. The patients are usually asked to hold their breath at end expiration, but it is not perfect and variations occur from breath-hold to breath-hold. Nevertheless, the long axis views can help guide the location of the valve plane. Clinicians also look for tissue in the sax views that "belong" to the ventricular muscle (since the atria are more fibrous) and that helps them in determining the plane too.
\n\nIn any case, the placement of the valve plane can cause large variations because the cross section of the ventricle at the base is large. The apex is less critical just because the cross section is so much smaller so any errors made there are much less significant, even if the mechanism is similar.
\n\nHope this helps,\nMichael
", "rawMarkdown": "Hi Dirk,\r\n\r\nYes, the selection of the valve plane in systole and diastole is where big errors are sometimes incurred. The long axis views can help with this, but it must be used carefully since it was acquired in a different breath-hold than the sax slices. The patients are usually asked to hold their breath at end expiration, but it is not perfect and variations occur from breath-hold to breath-hold. Nevertheless, the long axis views can help guide the location of the valve plane. Clinicians also look for tissue in the sax views that \"belong\" to the ventricular muscle (since the atria are more fibrous) and that helps them in determining the plane too. \r\n\r\nIn any case, the placement of the valve plane can cause large variations because the cross section of the ventricle at the base is large. The apex is less critical just because the cross section is so much smaller so any errors made there are much less significant, even if the mechanism is similar. \r\n\r\nHope this helps,\r\nMichael", "votes": null }, { "id": "103931", "postDate": "01/08/2016 00:51:47", "content": "This might be a stupid question but I seem to have trouble finding where the areas/volumes/doctor annotations of the left ventricle is in the training data. I accessed some of the dicom files and there seems to be no element that specifies when a specific image is in diastole/systole.
\n\nAnyone know where I can get that information from the training files? Thanks!
", "rawMarkdown": "This might be a stupid question but I seem to have trouble finding where the areas/volumes/doctor annotations of the left ventricle is in the training data. I accessed some of the dicom files and there seems to be no element that specifies when a specific image is in diastole/systole. \r\n\r\nAnyone know where I can get that information from the training files? Thanks!", "votes": null }, { "id": "103932", "postDate": "01/08/2016 00:53:15", "content": "@Max, There is no such annotation. That is part of the task, to identify systole and diastole.
", "rawMarkdown": "Max, There is no such annotation. That is part of the task, to identify systole and diastole.", "votes": null }, { "id": "103933", "postDate": "01/08/2016 00:59:01", "content": "@Dr. Hansen
\n\nAh I was confused. The data description stated:
\n\n"train.zip - the train set directory, contains cases where you will have the associated systolic and diastolic volumes"
\n\nI was given the impression that the images contained labeled data containing systolic and diastolic volumes for evaluation purposes.
", "rawMarkdown": "Dr. Hansen\r\n\r\nAh I was confused. The data description stated: \r\n\r\n\"train.zip - the train set directory, contains cases where you will have the associated systolic and diastolic volumes\"\r\n\r\nI was given the impression that the images contained labeled data containing systolic and diastolic volumes for evaluation purposes.", "votes": null }, { "id": "103934", "postDate": "01/08/2016 01:00:37", "content": "No, the volumes are given, but it is not stated which time frames they correspond to. If the algorithm is to be automatic, it should be able to find those frames (or analyze them all and report min/max).
", "rawMarkdown": "No, the volumes are given, but it is not stated which time frames they correspond to. If the algorithm is to be automatic, it should be able to find those frames (or analyze them all and report min/max).", "votes": null }, { "id": "103936", "postDate": "01/08/2016 01:15:26", "content": "Thank you for the clarification Dr. Hansen.
", "rawMarkdown": "Thank you for the clarification Dr. Hansen.", "votes": null }, { "id": "109837", "postDate": "03/01/2016 06:04:57", "content": "@Shannon- the program that calculates the volumes, can you confirm it uses the formula given in the Fourier tutorial (Step 4)? Some cardiology department websites I've seen indicate their calculations include both slice thickness and distance between slices, which the tutorial formula does not.
\n\nSince the real challenge here is in finding the cross-sectional area of the LV, I'd hate to have good estimates for the areas and then plug them into the wrong volume formula.
", "rawMarkdown": "Shannon- the program that calculates the volumes, can you confirm it uses the formula given in the Fourier tutorial (Step 4)? Some cardiology department websites I've seen indicate their calculations include both slice thickness *and* distance between slices, which the tutorial formula does not. \r\n\r\nSince the real challenge here is in finding the cross-sectional area of the LV, I'd hate to have good estimates for the areas and then plug them into the wrong volume formula.", "votes": null }, { "id": "110000", "postDate": "03/02/2016 00:01:45", "content": "I've been wondering about this too. Put in quite some work to develop an independent algorithm, now coming to the point where the way the volume is calculated from the crossections starts to matter a lot. But what is it? A convex hull, a nearest neighbor interpolation, a cone volume, a B-spline surface, etc.? It would have been really helpful to have provided some sort of tool to upload the crossections and thus standardize the volume calculation across participants.
\n\n[quote=Riaz;109837]
\n\n@Shannon- the program that calculates the volumes, can you confirm it uses the formula given in the Fourier tutorial (Step 4)? Some cardiology department websites I've seen indicate their calculations include both slice thickness and distance between slices, which the tutorial formula does not.
\n\nSince the real challenge here is in finding the cross-sectional area of the LV, I'd hate to have good estimates for the areas and then plug them into the wrong volume formula.
\n\n[/quote]
", "rawMarkdown": "I've been wondering about this too. Put in quite some work to develop an independent algorithm, now coming to the point where the way the volume is calculated from the crossections starts to matter a lot. But what is it? A convex hull, a nearest neighbor interpolation, a cone volume, a B-spline surface, etc.? It would have been really helpful to have provided some sort of tool to upload the crossections and thus standardize the volume calculation across participants.\r\n\r\n[quote=Riaz;109837]\r\n\r\n@Shannon- the program that calculates the volumes, can you confirm it uses the formula given in the Fourier tutorial (Step 4)? Some cardiology department websites I've seen indicate their calculations include both slice thickness *and* distance between slices, which the tutorial formula does not. \r\n\r\nSince the real challenge here is in finding the cross-sectional area of the LV, I'd hate to have good estimates for the areas and then plug them into the wrong volume formula.\r\n\r\n[/quote]", "votes": null } ], "comments": [ { "id": 102546, "author_name": "shannonlantzy", "author_url": "", "post_date": "12/23/2015 15:01:18", "content": "Hi woshialex,
\n\nA cardiologist draws contours (circles) at end systole and end diastole for each slice in the short axis stack, and a program is used to calculate the volumes.
\n\nThe video walk-through of the MRIs may be helpful (posted on the Resources page, and linked here).
", "votes": null, "replies": [] }, { "id": 102547, "author_name": "wcukierski", "author_url": "", "post_date": "12/23/2015 15:01:26", "content": "Yes, they are hand-drawn circles which are combined into a volume. We show the process in the tutorial video. As with any real data set, you should expect errors both large and small, systematic and spurious, believable and unbelievable.
", "votes": null, "replies": [] }, { "id": 103253, "author_name": "liveflow", "author_url": "", "post_date": "12/30/2015 19:17:29", "content": "I would like to confirm if each volume in the training dataset corresponds to each sax folder/patient or is it the total volume comprising all folders/patient ?
", "votes": null, "replies": [] }, { "id": 103514, "author_name": "diwiwo", "author_url": "", "post_date": "01/03/2016 22:36:14", "content": "In the tutorial video I don't think anything is mentioned about the long axis measurement of the LV and the influence of that on the calculated volumes. Is that taken into account and if yes, how? Or is it f.e. covered by assuming that the apex - slice reaches a zero surface on systole?
", "votes": null, "replies": [] }, { "id": 103527, "author_name": "michaelhansen", "author_url": "", "post_date": "01/04/2016 00:31:36", "content": "@DiekWillenWonnick, I am not sure that I understand your question. The long axis views of the LV just provide a different view of the ventricle compared to the short axis views, but it is a view of the same ventricle. So your question about whether it is taken into account is a bit hard to interpret. It is another view of the same volume. The volume does not change depending on how many views you have of the volume. The long axis views are often helpful in determining the location of the valve plane, i.e., where the atrium ends and the ventricle begins.
", "votes": null, "replies": [] }, { "id": 103549, "author_name": "diwiwo", "author_url": "", "post_date": "01/04/2016 08:41:31", "content": "Hi Michael. Thanks for you explanation. Wrong interpretation of the volume calculation by me. The long-axis size (L in below picture ) is derived from the choice of the slices. It makes the right slice selection of the valve in systole and in diastole extra critical. (and also, but probably less, for the 'apex-slice').
\n\n", "votes": null, "replies": [] }, { "id": 103572, "author_name": "michaelhansen", "author_url": "", "post_date": "01/04/2016 13:21:09", "content": "Hi Dirk,
\n\nYes, the selection of the valve plane in systole and diastole is where big errors are sometimes incurred. The long axis views can help with this, but it must be used carefully since it was acquired in a different breath-hold than the sax slices. The patients are usually asked to hold their breath at end expiration, but it is not perfect and variations occur from breath-hold to breath-hold. Nevertheless, the long axis views can help guide the location of the valve plane. Clinicians also look for tissue in the sax views that "belong" to the ventricular muscle (since the atria are more fibrous) and that helps them in determining the plane too.
\n\nIn any case, the placement of the valve plane can cause large variations because the cross section of the ventricle at the base is large. The apex is less critical just because the cross section is so much smaller so any errors made there are much less significant, even if the mechanism is similar.
\n\nHope this helps,\nMichael
", "votes": null, "replies": [] }, { "id": 103931, "author_name": "freytheviking", "author_url": "", "post_date": "01/08/2016 00:51:47", "content": "This might be a stupid question but I seem to have trouble finding where the areas/volumes/doctor annotations of the left ventricle is in the training data. I accessed some of the dicom files and there seems to be no element that specifies when a specific image is in diastole/systole.
\n\nAnyone know where I can get that information from the training files? Thanks!
", "votes": null, "replies": [] }, { "id": 103932, "author_name": "michaelhansen", "author_url": "", "post_date": "01/08/2016 00:53:15", "content": "@Max, There is no such annotation. That is part of the task, to identify systole and diastole.
", "votes": null, "replies": [] }, { "id": 103933, "author_name": "freytheviking", "author_url": "", "post_date": "01/08/2016 00:59:01", "content": "@Dr. Hansen
\n\nAh I was confused. The data description stated:
\n\n"train.zip - the train set directory, contains cases where you will have the associated systolic and diastolic volumes"
\n\nI was given the impression that the images contained labeled data containing systolic and diastolic volumes for evaluation purposes.
", "votes": null, "replies": [] }, { "id": 103934, "author_name": "michaelhansen", "author_url": "", "post_date": "01/08/2016 01:00:37", "content": "No, the volumes are given, but it is not stated which time frames they correspond to. If the algorithm is to be automatic, it should be able to find those frames (or analyze them all and report min/max).
", "votes": null, "replies": [] }, { "id": 103936, "author_name": "freytheviking", "author_url": "", "post_date": "01/08/2016 01:15:26", "content": "Thank you for the clarification Dr. Hansen.
", "votes": null, "replies": [] }, { "id": 109837, "author_name": "riazjahangir", "author_url": "", "post_date": "03/01/2016 06:04:57", "content": "@Shannon- the program that calculates the volumes, can you confirm it uses the formula given in the Fourier tutorial (Step 4)? Some cardiology department websites I've seen indicate their calculations include both slice thickness and distance between slices, which the tutorial formula does not.
\n\nSince the real challenge here is in finding the cross-sectional area of the LV, I'd hate to have good estimates for the areas and then plug them into the wrong volume formula.
", "votes": null, "replies": [] }, { "id": 110000, "author_name": "bendlflo", "author_url": "", "post_date": "03/02/2016 00:01:45", "content": "I've been wondering about this too. Put in quite some work to develop an independent algorithm, now coming to the point where the way the volume is calculated from the crossections starts to matter a lot. But what is it? A convex hull, a nearest neighbor interpolation, a cone volume, a B-spline surface, etc.? It would have been really helpful to have provided some sort of tool to upload the crossections and thus standardize the volume calculation across participants.
\n\n[quote=Riaz;109837]
\n\n@Shannon- the program that calculates the volumes, can you confirm it uses the formula given in the Fourier tutorial (Step 4)? Some cardiology department websites I've seen indicate their calculations include both slice thickness and distance between slices, which the tutorial formula does not.
\n\nSince the real challenge here is in finding the cross-sectional area of the LV, I'd hate to have good estimates for the areas and then plug them into the wrong volume formula.
\n\n[/quote]
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "102499": "is it labeled with a red circle by hand and then measured by human (then it is area instead of volume)? so we should expect error on these data?", "102546": "Hi woshialex,\r\n\r\nA cardiologist draws contours (circles) at end systole and end diastole for each slice in the short axis stack, and a program is used to calculate the volumes.\r\n\r\nThe video walk-through of the MRIs may be helpful (posted on the [Resources][1] page, and linked [here][2]).\r\n\r\n\r\n [1]: https://www.kaggle.com/c/second-annual-data-science-bowl/details/resources\r\n [2]: https://youtu.be/dFu_5T0ODrM", "102547": "Yes, they are hand-drawn circles which are combined into a volume. We show the process in the [tutorial video][1]. As with any real data set, you should expect errors both large and small, systematic and spurious, believable and unbelievable.\r\n\r\n [1]: https://youtu.be/dFu_5T0ODrM", "103253": "I would like to confirm if each volume in the training dataset corresponds to each sax folder/patient or is it the total volume comprising all folders/patient ?", "103514": "In the tutorial video I don't think anything is mentioned about the long axis measurement of the LV and the influence of that on the calculated volumes. Is that taken into account and if yes, how? Or is it f.e. covered by assuming that the apex - slice reaches a zero surface on systole?", "103527": "DiekWillenWonnick, I am not sure that I understand your question. The long axis views of the LV just provide a different view of the ventricle compared to the short axis views, but it is a view of the same ventricle. So your question about whether it is taken into account is a bit hard to interpret. It is another view of the same volume. The volume does not change depending on how many views you have of the volume. The long axis views are often helpful in determining the location of the valve plane, i.e., where the atrium ends and the ventricle begins.", "103549": "Hi Michael. Thanks for you explanation. Wrong interpretation of the volume calculation by me. The long-axis size (L in below picture ) is derived from the choice of the slices. It makes the right slice selection of the valve in systole and in diastole extra critical. (and also, but probably less, for the 'apex-slice'). \r\n\r\n[Simpson method][1]\r\n [1]: http://www.winfocus.org/_/rsrc/1272864420255/e-learning/echo/left_ventricle/Winfocus_CCECho_LV_Fig13.png?width=60%25", "103572": "Hi Dirk,\r\n\r\nYes, the selection of the valve plane in systole and diastole is where big errors are sometimes incurred. The long axis views can help with this, but it must be used carefully since it was acquired in a different breath-hold than the sax slices. The patients are usually asked to hold their breath at end expiration, but it is not perfect and variations occur from breath-hold to breath-hold. Nevertheless, the long axis views can help guide the location of the valve plane. Clinicians also look for tissue in the sax views that \"belong\" to the ventricular muscle (since the atria are more fibrous) and that helps them in determining the plane too. \r\n\r\nIn any case, the placement of the valve plane can cause large variations because the cross section of the ventricle at the base is large. The apex is less critical just because the cross section is so much smaller so any errors made there are much less significant, even if the mechanism is similar. \r\n\r\nHope this helps,\r\nMichael", "103931": "This might be a stupid question but I seem to have trouble finding where the areas/volumes/doctor annotations of the left ventricle is in the training data. I accessed some of the dicom files and there seems to be no element that specifies when a specific image is in diastole/systole. \r\n\r\nAnyone know where I can get that information from the training files? Thanks!", "103932": "Max, There is no such annotation. That is part of the task, to identify systole and diastole.", "103933": "Dr. Hansen\r\n\r\nAh I was confused. The data description stated: \r\n\r\n\"train.zip - the train set directory, contains cases where you will have the associated systolic and diastolic volumes\"\r\n\r\nI was given the impression that the images contained labeled data containing systolic and diastolic volumes for evaluation purposes.", "103934": "No, the volumes are given, but it is not stated which time frames they correspond to. If the algorithm is to be automatic, it should be able to find those frames (or analyze them all and report min/max).", "103936": "Thank you for the clarification Dr. Hansen.", "109837": "Shannon- the program that calculates the volumes, can you confirm it uses the formula given in the Fourier tutorial (Step 4)? Some cardiology department websites I've seen indicate their calculations include both slice thickness *and* distance between slices, which the tutorial formula does not. \r\n\r\nSince the real challenge here is in finding the cross-sectional area of the LV, I'd hate to have good estimates for the areas and then plug them into the wrong volume formula.", "110000": "I've been wondering about this too. Put in quite some work to develop an independent algorithm, now coming to the point where the way the volume is calculated from the crossections starts to matter a lot. But what is it? A convex hull, a nearest neighbor interpolation, a cone volume, a B-spline surface, etc.? It would have been really helpful to have provided some sort of tool to upload the crossections and thus standardize the volume calculation across participants.\r\n\r\n[quote=Riaz;109837]\r\n\r\n@Shannon- the program that calculates the volumes, can you confirm it uses the formula given in the Fourier tutorial (Step 4)? Some cardiology department websites I've seen indicate their calculations include both slice thickness *and* distance between slices, which the tutorial formula does not. \r\n\r\nSince the real challenge here is in finding the cross-sectional area of the LV, I'd hate to have good estimates for the areas and then plug them into the wrong volume formula.\r\n\r\n[/quote]" }, "source": "meta" }![Simpson method](\"http://www.winfocus.org/_/rsrc/1272864420255/e-learning/echo/left_ventricle/Winfocus_CCECho_LV_Fig13.png?width=60%25\"){kind=link}