Hi there, I noticed that sometimes the T1wCE dicoms have an odd shape that seems too consistent to be a coincidence.
Most folders of dicom images can be interpreted as a 3d array with the given dimensions:
\nCOLSROWSThis will usually yield a shape of (X, X, Z) where X is a single dimension (i.e. square images) and Z is the depth. This pattern is observable in almost all of the examples… with some exceptions.
The most common exception is when we get a shape of (X, Y, Z) where X and Y are different dimensions (i.e. rectangle images) and Z is the depth. This seems acceptable.
However, there is sometimes another exception to this pattern found within T1wCE scans. In these scans, the pattern is (X, Z, Z) where X is a single dimension (width) and Z is both the depth and the height. I thought at first this was a coincidence and the images were just rectangular with the height coincidentally being equal to the depth…. but it happens quite frequently with T1wCE scans and seems to almost never happen for any other types of scans.
Are these scans rotated?
\nI tried visualizing … but I'm not the best at understanding 3-dimensional structures intuitively and couldn't figure it out.
", "messages": [ { "id": "1398059", "postDate": "07/23/2021 17:48:57", "content": "Hi there, I noticed that sometimes the T1wCE dicoms have an odd shape that seems too consistent to be a coincidence.
Most folders of dicom images can be interpreted as a 3d array with the given dimensions:
\nCOLSROWSThis will usually yield a shape of (X, X, Z) where X is a single dimension (i.e. square images) and Z is the depth. This pattern is observable in almost all of the examples… with some exceptions.
The most common exception is when we get a shape of (X, Y, Z) where X and Y are different dimensions (i.e. rectangle images) and Z is the depth. This seems acceptable.
However, there is sometimes another exception to this pattern found within T1wCE scans. In these scans, the pattern is (X, Z, Z) where X is a single dimension (width) and Z is both the depth and the height. I thought at first this was a coincidence and the images were just rectangular with the height coincidentally being equal to the depth…. but it happens quite frequently with T1wCE scans and seems to almost never happen for any other types of scans.
Are these scans rotated?
\nI tried visualizing … but I'm not the best at understanding 3-dimensional structures intuitively and couldn't figure it out.
", "rawMarkdown": "Hi there, I noticed that sometimes the **`T1wCE`** dicoms have an odd shape that seems too consistent to be a coincidence.\n\nMost folders of dicom images can be interpreted as a 3d array with the given dimensions:\n- Width = dicom property **`COLS`**\n- Height = dicom property **`ROWS`**\n- Depth = number of dicom files\n\nThis will usually yield a shape of **`(X, X, Z)`** where **`X`** is a single dimension (i.e. square images) and **`Z`** is the depth. This pattern is observable in almost all of the examples... with some exceptions.\n\nThe most common exception is when we get a shape of **`(X, Y, Z)`** where **`X`** and **`Y`** are different dimensions (i.e. rectangle images) and **`Z`** is the depth. This seems acceptable.\n\nHowever, there is sometimes another exception to this pattern found within **`T1wCE`** scans. In these scans, the pattern is **`(X, Z, Z)`** where **`X`** is a single dimension (width) and **`Z`** is both the depth and the height. I thought at first this was a coincidence and the images were just rectangular with the height coincidentally being equal to the depth.... but it happens quite frequently with **`T1wCE`** scans and seems to almost never happen for any other types of scans.\n\nIt's kind of a coincidence. Some of the acquisition matrices MR images are scanned as are smaller than you would expect. A lot of them are 256x192 (or smaller), with a pixel spacing of 1 .. or really close to it (~0.9). That means one pixel is equal to 1mm in the real world. So, If I wanted to do a multiplanar reconstruction at 1mm thickness, with no space between slices, I would need to export 256 images .. which coincidently is the width of the matrix (depending on which plane is being reconstructed).
", "rawMarkdown": "It's kind of a coincidence. Some of the acquisition matrices MR images are scanned as are smaller than you would expect. A lot of them are 256x192 (or smaller), with a pixel spacing of 1 .. or really close to it (~0.9). That means one pixel is equal to 1mm in the real world. So, If I wanted to do a multiplanar reconstruction at 1mm thickness, with no space between slices, I would need to export 256 images .. which coincidently is the width of the matrix (depending on which plane is being reconstructed).", "votes": null }, { "id": "1398258", "postDate": "07/23/2021 23:29:55", "content": "The interesting thing about this is that when the image count matches the dimensions .. (256, 512), you can stack them into a 3D array and reformat into other 2D planes (coronal to sagittal etc), or 3D, and get a reasonably accurate reformation without calculating slice position from DICOM tags.
\nIn fact, it's the only time you can simply stack jpegs to reformat them properly.
", "rawMarkdown": "The interesting thing about this is that when the image count matches the dimensions .. (256, 512), you can stack them into a 3D array and reformat into other 2D planes (coronal to sagittal etc), or 3D, and get a reasonably accurate reformation without calculating slice position from DICOM tags. \n\nIn fact, it's the only time you can simply stack jpegs to reformat them properly.", "votes": null }, { "id": "1400052", "postDate": "07/26/2021 01:51:16", "content": "Fascinating, thanks David. Many high yield contributions to this comp so far.
", "rawMarkdown": "Fascinating, thanks David. Many high yield contributions to this comp so far.", "votes": null }, { "id": "1401400", "postDate": "07/27/2021 09:40:39", "content": "Check https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FSLeyes for the visualization of MRIs, it's my favorite
", "rawMarkdown": "Check https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FSLeyes for the visualization of MRIs, it's my favorite", "votes": null }, { "id": "1401661", "postDate": "07/27/2021 13:26:57", "content": "If you want to visualise images in video format, this would be helpful.
", "rawMarkdown": "If you want to visualise images in video format, [this](https://www.kaggle.com/abhijeetptl5/rsna-miccai-video) would be helpful.", "votes": null }, { "id": "1406033", "postDate": "07/31/2021 12:26:07", "content": "Agree with this statement for @davidbroberts !
", "rawMarkdown": "Agree with this statement for @davidbroberts !", "votes": null } ], "comments": [ { "id": 1398211, "author_name": "davidbroberts", "author_url": "", "post_date": "07/23/2021 21:15:43", "content": "It's kind of a coincidence. Some of the acquisition matrices MR images are scanned as are smaller than you would expect. A lot of them are 256x192 (or smaller), with a pixel spacing of 1 .. or really close to it (~0.9). That means one pixel is equal to 1mm in the real world. So, If I wanted to do a multiplanar reconstruction at 1mm thickness, with no space between slices, I would need to export 256 images .. which coincidently is the width of the matrix (depending on which plane is being reconstructed).
", "votes": null, "replies": [] }, { "id": 1398258, "author_name": "davidbroberts", "author_url": "", "post_date": "07/23/2021 23:29:55", "content": "The interesting thing about this is that when the image count matches the dimensions .. (256, 512), you can stack them into a 3D array and reformat into other 2D planes (coronal to sagittal etc), or 3D, and get a reasonably accurate reformation without calculating slice position from DICOM tags.
\nIn fact, it's the only time you can simply stack jpegs to reformat them properly.
", "votes": null, "replies": [ { "id": 1400052, "author_name": "reubenschmidt", "author_url": "", "post_date": "07/26/2021 01:51:16", "content": "Fascinating, thanks David. Many high yield contributions to this comp so far.
", "votes": null, "replies": [] }, { "id": 1406033, "author_name": "elenaeb", "author_url": "", "post_date": "07/31/2021 12:26:07", "content": "Agree with this statement for @davidbroberts !
", "votes": null, "replies": [] } ] }, { "id": 1401400, "author_name": "ranafago", "author_url": "", "post_date": "07/27/2021 09:40:39", "content": "Check https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FSLeyes for the visualization of MRIs, it's my favorite
", "votes": null, "replies": [] }, { "id": 1401661, "author_name": "abhijeetptl5", "author_url": "", "post_date": "07/27/2021 13:26:57", "content": "If you want to visualise images in video format, this would be helpful.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1398059": "Hi there, I noticed that sometimes the **`T1wCE`** dicoms have an odd shape that seems too consistent to be a coincidence.\n\nMost folders of dicom images can be interpreted as a 3d array with the given dimensions:\n- Width = dicom property **`COLS`**\n- Height = dicom property **`ROWS`**\n- Depth = number of dicom files\n\nThis will usually yield a shape of **`(X, X, Z)`** where **`X`** is a single dimension (i.e. square images) and **`Z`** is the depth. This pattern is observable in almost all of the examples... with some exceptions.\n\nThe most common exception is when we get a shape of **`(X, Y, Z)`** where **`X`** and **`Y`** are different dimensions (i.e. rectangle images) and **`Z`** is the depth. This seems acceptable.\n\nHowever, there is sometimes another exception to this pattern found within **`T1wCE`** scans. In these scans, the pattern is **`(X, Z, Z)`** where **`X`** is a single dimension (width) and **`Z`** is both the depth and the height. I thought at first this was a coincidence and the images were just rectangular with the height coincidentally being equal to the depth.... but it happens quite frequently with **`T1wCE`** scans and seems to almost never happen for any other types of scans.\n\n