![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F987995%2Fff2b3a20911101b9fab84446feab3657%2Flumbar1.png?generation=1716552324536375&alt=media\")
For those who are unfamiliar with the structure of the MRI datasets i would like to show what are the connections between the studies and the series of a study:
\nFollowing figure contains a volume made of a sagittal series and one transversal series:
\n
The position information where voxels (3d equivalent of image pixels) are situated are coded in the dicom (source) files.
\nThis means, every y,x coordinate pair is situated somewhere within this volume depicted.
", "messages": [ { "id": 2833825, "postDate": "2024-05-24T12:15:03.297Z", "content": "For those who are unfamiliar with the structure of the MRI datasets i would like to show what are the connections between the studies and the series of a study:
\nFollowing figure contains a volume made of a sagittal series and one transversal series:
\n
The position information where voxels (3d equivalent of image pixels) are situated are coded in the dicom (source) files.
\nThis means, every y,x coordinate pair is situated somewhere within this volume depicted.
", "rawMarkdown": "For those who are unfamiliar with the structure of the MRI datasets i would like to show what are the connections between the **studies and the series of a study**:\n\n- one study is composed of multiple series that are aquired within one session.\n- series are set of images, usually aquired in one run (in this competition this is the case for sagittal but not neceserily for axial series!)\n- the images of each serie depict a given volume of the spine - different series might **overlap each other** but **not necesarily completely**\n\nFollowing figure contains a volume made of a sagittal series and one transversal series:\n\n\n\nThe position information where voxels (3d equivalent of image pixels) are situated are coded in the dicom (source) files.\n\nThis means, every y,x coordinate pair is situated somewhere within this volume depicted.", "votes": 93 }, { "id": 2834708, "postDate": "2024-05-25T00:17:58.060Z", "content": "Thank you for sharing this insightful information. I'm really glad you're here to provide such valuable context.
\nI have a couple of questions, if that's alright:
\nIs the reason there are many similar images due to the need for ensuring that at least one captures the necessary details clearly? In other words, are multiple images taken to account for possible variations in quality and to increase the chances of obtaining a diagnostically useful image?
Regarding image quality, would you agree that in the test set, image 12 can be considered high quality because it clearly shows all spine levels? Conversely, image 22 appears to be of lower quality as it does not distinctly display all spine levels. Are there specific criteria you use to assess the quality of these images?
Thank you again for your contribution. Your expertise is very much appreciated.
\nstudy_id = 44036939
\nseries_id = 2828203845
![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F19436996%2F9bc1eb1223b5086c76c920f86ee83009%2FScreen%20Shot%202024-05-25%20at%2010.07.45%20am.png?generation=1716595834288042&alt=media\")
![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F19436996%2Fcd1734de3c19fec42f6aabfad8037365%2FScreen%20Shot%202024-05-25%20at%2010.08.38%20am.png?generation=1716595868258281&alt=media\")
\n\nThank you for sharing this insightful information. I'm really glad you're here to provide such valuable context.
\n
You're welcome, this is only the first one, there will be much more!
\n\n\nIs the reason there are many similar images due to the need for ensuring that at least one captures the necessary details clearly? In other words, are multiple images taken to account for possible variations in quality and to increase the chances of obtaining a diagnostically useful image?
\n
We have to capture images from the whole region - in this case lumbar spine. This includes the body parts around the spine (like lumbar muscles, aorta, partially the renal system and so on).
\nWe aquire slices that depict one thin slice in plane of the given region, so consecutive images are \"next to each other\" in space. You're completely right, this is because we have to capture every part in order to be able to see where exactly the problem is, so we can determine the cause of the problem.
\n\nRegarding image quality, would you agree that in the test set, image 12 can be considered high quality because it clearly shows all spine levels? Conversely, image 22 appears to be of lower quality as it does not distinctly display all spine levels. Are there specific criteria you use to assess the quality of these images?
\n
You're right, on instance 12 one can see more vertebrae bodies and intervertebral spaces, however in this sequence (T1) the spinal coloumn stenosis is (for human eye!) not good assessable. The neuroforamen stenosis is better assessable, which means that on instance 22 the neuroforamen L4-L5 and L5-S1 on the right side (red circle) are assessable.
\n![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F987995%2Ff80abb5d25a8b8259f523aba188df147%2Frsna1.png?generation=1716634939788840&alt=media\")
The quality of both of the images are ok, there is no noise of motion artefact - these are some of the quality criteria we use.
\nWhat you see is that the spinal column is not always straight, this condition is called scoliosis:
\n![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F987995%2F95158d9b1364b9ed206b27a6ccf582fe%2Frsna3.png?generation=1716635650625444&alt=media\")
Parallel black lines are our slices, so the image planes next to each other, let's say each of them represents one image.
\nIn these situations the images \"more on the side\" might depict only few vertebrae bodies and corresponding neuroforamen (since the other ones are simply not in the same plane, the images \"from the middle\" might show more vertebral bodies and pars of the spinal coloumn.
\nAlso note, that i rotated the parallel black lines slightly and they are not exactly along the length axis of the body - this is quite common in these situations in the sagittal and also in the transversal (perpendicular) planes.
\nDifferent sequences show different tissue with different signal intensities - i will write a post about it soon, but it is like a puzzle… we have diferent signals from the same region over multiple sequences and we have to puzzle together what kind og tissue is giving us the signal we see.
", "rawMarkdown": ">Thank you for sharing this insightful information. I'm really glad you're here to provide such valuable context.\n\nYou're welcome, this is only the first one, there will be much more!\n\n>Is the reason there are many similar images due to the need for ensuring that at least one captures the necessary details clearly? In other words, are multiple images taken to account for possible variations in quality and to increase the chances of obtaining a diagnostically useful image?\n\nWe have to capture images from the whole region - in this case lumbar spine. This includes the body parts around the spine (like lumbar muscles, aorta, partially the renal system and so on). \nWe **aquire slices** that **depict** one **thin slice in plane** of the given region, so consecutive images are \"next to each other\" in space. You're completely right, this is because we have to capture every part in order to be able to see where exactly the problem is, so we can determine the cause of the problem.\n\n>Regarding image quality, would you agree that in the test set, image 12 can be considered high quality because it clearly shows all spine levels? Conversely, image 22 appears to be of lower quality as it does not distinctly display all spine levels. Are there specific criteria you use to assess the quality of these images?\n\nYou're right, on instance 12 one can see more vertebrae bodies and intervertebral spaces, however in this sequence (T1) the spinal coloumn stenosis is (for human eye!) not good assessable. The neuroforamen stenosis is better assessable, which means that on instance 22 the neuroforamen L4-L5 and L5-S1 on the right side (red circle) are assessable.\n\n\n\nThe quality of both of the images are ok, there is no noise of motion artefact - these are some of the quality criteria we use.\n\nWhat you see is that the spinal column is not always straight, this condition is called scoliosis:\n\n\nParallel black lines are our slices, so the image planes next to each other, let's say each of them represents one image.\n\nIn these situations the images \"more on the side\" might depict only few vertebrae bodies and corresponding neuroforamen (since the other ones are simply **not in the same plane**, the images \"from the middle\" might show more vertebral bodies and pars of the spinal coloumn. \n\nAlso note, that i rotated the parallel black lines slightly and they are not exactly along the length axis of the body - this is quite common in these situations in the sagittal and also in the transversal (perpendicular) planes.\n\nDifferent sequences show different tissue with different signal intensities - i will write a post about it soon, but it is like a puzzle... we have diferent signals from the same region over multiple sequences and we have to puzzle together what kind og tissue is giving us the signal we see. ", "votes": 18, "replies": [ { "id": 2835622, "postDate": "2024-05-25T12:28:03.650Z", "content": "Yes, how to make good use of this data is crucial; there are many methods that can be used to handle this effectively.
", "rawMarkdown": "Yes, how to make good use of this data is crucial; there are many methods that can be used to handle this effectively.\n", "votes": 4 }, { "id": 2836501, "postDate": "2024-05-26T00:59:37.820Z", "content": "\n\nAlso note, that i rotated the parallel black lines slightly and they are not exactly along the length axis of the body - this is quite common in these situations in the sagittal and also in the transversal (perpendicular) planes.
\n
This seems to answer my question about multiple series_id under the same study_id, the Scoliosis might cause this?
", "rawMarkdown": "> Also note, that i rotated the parallel black lines slightly and they are not exactly along the length axis of the body - this is quite common in these situations in the sagittal and also in the transversal (perpendicular) planes.\n\nThis seems to answer my question about multiple series_id under the same study_id, the Scoliosis might cause this?", "votes": 1 }, { "id": 2836662, "postDate": "2024-05-26T04:08:46.477Z", "content": "This is great information, especially regarding the clarification that image 22 is useful. I was thinking of coming up with a function to disregard low quality images. Do you have any advice in this area?
\nThanks again for your detailed explanations it is greatly appreciated.
", "rawMarkdown": "This is great information, especially regarding the clarification that image 22 is useful. I was thinking of coming up with a function to disregard low quality images. Do you have any advice in this area?\n\nThanks again for your detailed explanations it is greatly appreciated. " }, { "id": 2836675, "postDate": "2024-05-26T04:17:06.480Z", "content": "Hi zby, do you care to share any of those methods?
\nCongrats on the 0.17 score btw, nice work.
", "rawMarkdown": "Hi zby, do you care to share any of those methods? \n\nCongrats on the 0.17 score btw, nice work. " }, { "id": 2836746, "postDate": "2024-05-26T05:25:07.750Z", "content": "The study is the collection of all series within. The number of series varies, usually the number of distinct axial series increases by scoliosis because you want to have them aligned perpendicular to the given segment, so yes, in case of scoliosis you might end up with more axial series (or even more sagittal ones but that is imho not common).
", "rawMarkdown": "The study is the collection of all series within. The number of series varies, usually the number of distinct axial series increases by scoliosis because you want to have them aligned perpendicular to the given segment, so yes, in case of scoliosis you might end up with more axial series (or even more sagittal ones but that is imho not common).", "votes": 3 }, { "id": 2836749, "postDate": "2024-05-26T05:26:04.613Z", "content": "Any option beside of registering the volumes?
", "rawMarkdown": "Any option beside of registering the volumes?" } ] } ] }, { "id": 2834653, "postDate": "2024-05-24T21:58:18.497Z", "content": "It's always great to read the professionals insights, mostly those that work in Radiology field when we are dealing with dcm files.
\nVielen dank.
Do you have any idea on using the x,y info ?
", "rawMarkdown": "Do you have any idea on using the x,y info ?", "votes": 2 }, { "id": 2839908, "postDate": "2024-05-27T19:02:05.500Z", "content": "Super insightful! Congrats.
\nI have a question: which view (sagittal vs axial) do you, as a radiologist, think is better at assessing each one of the conditions (central/lateral/foraminal stenosis)?
", "rawMarkdown": "Super insightful! Congrats.\n\nI have a question: which view (sagittal vs axial) do you, as a radiologist, think is better at assessing each one of the conditions (central/lateral/foraminal stenosis)?", "votes": 2 }, { "id": 2891755, "postDate": "2024-06-26T20:30:08.203Z", "content": "Thank you very much for your insights. This is a complex dataset with a ton of variables to understand, so your comments help greatly to frame the problem space. Having domain experts here to help is fantastic!
", "rawMarkdown": "Thank you very much for your insights. This is a complex dataset with a ton of variables to understand, so your comments help greatly to frame the problem space. Having domain experts here to help is fantastic!" }, { "id": 2884298, "postDate": "2024-06-22T12:10:21.073Z", "content": "Thank you for sharing the insights, the problem makes more sense now to me.
", "rawMarkdown": "Thank you for sharing the insights, the problem makes more sense now to me." }, { "id": 2838306, "postDate": "2024-05-27T01:35:42.790Z", "rawMarkdown": "", "isDeleted": true }, { "id": 2838336, "postDate": "2024-05-27T01:56:59.217Z", "content": "Thank you for sharing.
", "rawMarkdown": "Thank you for sharing.", "votes": 1 }, { "id": 2871244, "postDate": "2024-06-14T06:04:41.033Z", "content": "Thanks for sharing!
", "rawMarkdown": "Thanks for sharing!" } ], "comments": [ { "id": 2834708, "author_name": "MrNosaj", "author_url": "", "post_date": "2024-05-25T00:17:58.060000", "content": "Thank you for sharing this insightful information. I'm really glad you're here to provide such valuable context.
\nI have a couple of questions, if that's alright:
\nIs the reason there are many similar images due to the need for ensuring that at least one captures the necessary details clearly? In other words, are multiple images taken to account for possible variations in quality and to increase the chances of obtaining a diagnostically useful image?
Regarding image quality, would you agree that in the test set, image 12 can be considered high quality because it clearly shows all spine levels? Conversely, image 22 appears to be of lower quality as it does not distinctly display all spine levels. Are there specific criteria you use to assess the quality of these images?
Thank you again for your contribution. Your expertise is very much appreciated.
\nstudy_id = 44036939
\nseries_id = 2828203845
\n\nThank you for sharing this insightful information. I'm really glad you're here to provide such valuable context.
\n
You're welcome, this is only the first one, there will be much more!
\n\n\nIs the reason there are many similar images due to the need for ensuring that at least one captures the necessary details clearly? In other words, are multiple images taken to account for possible variations in quality and to increase the chances of obtaining a diagnostically useful image?
\n
We have to capture images from the whole region - in this case lumbar spine. This includes the body parts around the spine (like lumbar muscles, aorta, partially the renal system and so on).
\nWe aquire slices that depict one thin slice in plane of the given region, so consecutive images are \"next to each other\" in space. You're completely right, this is because we have to capture every part in order to be able to see where exactly the problem is, so we can determine the cause of the problem.
\n\nRegarding image quality, would you agree that in the test set, image 12 can be considered high quality because it clearly shows all spine levels? Conversely, image 22 appears to be of lower quality as it does not distinctly display all spine levels. Are there specific criteria you use to assess the quality of these images?
\n
You're right, on instance 12 one can see more vertebrae bodies and intervertebral spaces, however in this sequence (T1) the spinal coloumn stenosis is (for human eye!) not good assessable. The neuroforamen stenosis is better assessable, which means that on instance 22 the neuroforamen L4-L5 and L5-S1 on the right side (red circle) are assessable.
\n
The quality of both of the images are ok, there is no noise of motion artefact - these are some of the quality criteria we use.
\nWhat you see is that the spinal column is not always straight, this condition is called scoliosis:
\n
Parallel black lines are our slices, so the image planes next to each other, let's say each of them represents one image.
\nIn these situations the images \"more on the side\" might depict only few vertebrae bodies and corresponding neuroforamen (since the other ones are simply not in the same plane, the images \"from the middle\" might show more vertebral bodies and pars of the spinal coloumn.
\nAlso note, that i rotated the parallel black lines slightly and they are not exactly along the length axis of the body - this is quite common in these situations in the sagittal and also in the transversal (perpendicular) planes.
\nDifferent sequences show different tissue with different signal intensities - i will write a post about it soon, but it is like a puzzle… we have diferent signals from the same region over multiple sequences and we have to puzzle together what kind og tissue is giving us the signal we see.
", "votes": 18, "replies": [ { "id": 2835622, "author_name": "", "author_url": "", "post_date": "2024-05-25T12:28:03.650000", "content": "Yes, how to make good use of this data is crucial; there are many methods that can be used to handle this effectively.
", "votes": 4, "replies": [] }, { "id": 2836501, "author_name": "xbxbxb", "author_url": "", "post_date": "2024-05-26T00:59:37.820000", "content": "\n\nAlso note, that i rotated the parallel black lines slightly and they are not exactly along the length axis of the body - this is quite common in these situations in the sagittal and also in the transversal (perpendicular) planes.
\n
This seems to answer my question about multiple series_id under the same study_id, the Scoliosis might cause this?
", "votes": 1, "replies": [] }, { "id": 2836662, "author_name": "MrNosaj", "author_url": "", "post_date": "2024-05-26T04:08:46.477000", "content": "This is great information, especially regarding the clarification that image 22 is useful. I was thinking of coming up with a function to disregard low quality images. Do you have any advice in this area?
\nThanks again for your detailed explanations it is greatly appreciated.
", "votes": 0, "replies": [] }, { "id": 2836675, "author_name": "MrNosaj", "author_url": "", "post_date": "2024-05-26T04:17:06.480000", "content": "Hi zby, do you care to share any of those methods?
\nCongrats on the 0.17 score btw, nice work.
", "votes": 0, "replies": [] }, { "id": 2836746, "author_name": "dr. Konya", "author_url": "", "post_date": "2024-05-26T05:25:07.750000", "content": "The study is the collection of all series within. The number of series varies, usually the number of distinct axial series increases by scoliosis because you want to have them aligned perpendicular to the given segment, so yes, in case of scoliosis you might end up with more axial series (or even more sagittal ones but that is imho not common).
", "votes": 3, "replies": [] }, { "id": 2836749, "author_name": "dr. Konya", "author_url": "", "post_date": "2024-05-26T05:26:04.613000", "content": "Any option beside of registering the volumes?
", "votes": 0, "replies": [] } ] } ] }, { "id": 2834653, "author_name": "Marília Prata", "author_url": "", "post_date": "2024-05-24T21:58:18.497000", "content": "It's always great to read the professionals insights, mostly those that work in Radiology field when we are dealing with dcm files.
\nVielen dank.
Do you have any idea on using the x,y info ?
", "votes": 2, "replies": [] }, { "id": 2839908, "author_name": "Bernardo", "author_url": "", "post_date": "2024-05-27T19:02:05.500000", "content": "Super insightful! Congrats.
\nI have a question: which view (sagittal vs axial) do you, as a radiologist, think is better at assessing each one of the conditions (central/lateral/foraminal stenosis)?
", "votes": 2, "replies": [] }, { "id": 2891755, "author_name": "Rob Bailey", "author_url": "", "post_date": "2024-06-26T20:30:08.203000", "content": "Thank you very much for your insights. This is a complex dataset with a ton of variables to understand, so your comments help greatly to frame the problem space. Having domain experts here to help is fantastic!
", "votes": 0, "replies": [] }, { "id": 2884298, "author_name": "Sadidul Islam", "author_url": "", "post_date": "2024-06-22T12:10:21.073000", "content": "Thank you for sharing the insights, the problem makes more sense now to me.
", "votes": 0, "replies": [] }, { "id": 2838306, "author_name": "", "author_url": "", "post_date": "2024-05-27T01:35:42.790000", "content": "", "votes": 0, "replies": [] }, { "id": 2838336, "author_name": "Victor Broering", "author_url": "", "post_date": "2024-05-27T01:56:59.217000", "content": "Thank you for sharing.
", "votes": 1, "replies": [] }, { "id": 2871244, "author_name": "Mohammad Mkanna", "author_url": "", "post_date": "2024-06-14T06:04:41.033000", "content": "Thanks for sharing!
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2833825": "For those who are unfamiliar with the structure of the MRI datasets i would like to show what are the connections between the **studies and the series of a study**:\n\n- one study is composed of multiple series that are aquired within one session.\n- series are set of images, usually aquired in one run (in this competition this is the case for sagittal but not neceserily for axial series!)\n- the images of each serie depict a given volume of the spine - different series might **overlap each other** but **not necesarily completely**\n\nFollowing figure contains a volume made of a sagittal series and one transversal series:\n\n\n\nThe position information where voxels (3d equivalent of image pixels) are situated are coded in the dicom (source) files.\n\nThis means, every y,x coordinate pair is situated somewhere within this volume depicted.", "2834708": "Thank you for sharing this insightful information. I'm really glad you're here to provide such valuable context.\n\nI have a couple of questions, if that's alright:\n\n1. Is the reason there are many similar images due to the need for ensuring that at least one captures the necessary details clearly? In other words, are multiple images taken to account for possible variations in quality and to increase the chances of obtaining a diagnostically useful image?\n\n1. Regarding image quality, would you agree that in the test set, image 12 can be considered high quality because it clearly shows all spine levels? Conversely, image 22 appears to be of lower quality as it does not distinctly display all spine levels. Are there specific criteria you use to assess the quality of these images?\n\nThank you again for your contribution. Your expertise is very much appreciated.\n\n**study_id** = 44036939\n**series_id** = 2828203845\n\n\n\n\n\n\n", "2834653": "It's always great to read the professionals insights, mostly those that work in Radiology field when we are dealing with dcm files.\nVielen dank.", "2840015": "Do you have any idea on using the x,y info ?", "2839908": "Super insightful! Congrats.\n\nI have a question: which view (sagittal vs axial) do you, as a radiologist, think is better at assessing each one of the conditions (central/lateral/foraminal stenosis)?", "2891755": "Thank you very much for your insights. This is a complex dataset with a ton of variables to understand, so your comments help greatly to frame the problem space. Having domain experts here to help is fantastic!", "2884298": "Thank you for sharing the insights, the problem makes more sense now to me.", "2838306": "", "2838336": "Thank you for sharing.", "2871244": "Thanks for sharing!" } }