\n
![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F581084%2F04fe7eda502fa07c4b2aefecea9b2525%2FbrainImageSet.jpg?generation=1753982626094076&alt=media\")
The dcm images seem to be in random order within each dcm folder - not ordered by location in the person's head. For example, I tried sorting by:
\ngetattr(dcm_img, 'ImagePositionPatient', 0)[2] and image number.
\nBoth seem to give image orders for random positions in the head.
\nThis has implications for 3d models and might then require image-by-image 2d analysis.
\nIs there a way to sort the images by position in the patient's head?
The dcm images seem to be in random order within each dcm folder - not ordered by location in the person's head. For example, I tried sorting by:
\ngetattr(dcm_img, 'ImagePositionPatient', 0)[2] and image number.
\nBoth seem to give image orders for random positions in the head.
\nThis has implications for 3d models and might then require image-by-image 2d analysis.
\nIs there a way to sort the images by position in the patient's head?
I think ImagePositionPatient should work well, since its the corresponding metadata for that slice's coordiantes. I use instance number as a backup just in case, but ImagePositionPatient should generally work. This is the function I wrote that worked for me.
():\n \n dcm_files = [os.path.join(series_path, f) f os.listdir(series_path) f.endswith()]\n slices = [pydicom.dcmread(f) f dcm_files]\n\n :\n slices.sort(key= s: (s.ImagePositionPatient[]))\n (AttributeError, IndexError):\n slices.sort(key= s: (s.InstanceNumber))\n slices\n@umarali1 - that seems to work - thank you. Have you run into any image sets where the \"except\" block executes?
", "rawMarkdown": "@umarali1 - that seems to work - thank you. Have you run into any image sets where the \"except\" block executes?", "replies": [ { "id": 3259032, "postDate": "2025-07-31T18:07:54.287Z", "content": "Nope, just checked and it seems like the except clause doesn't execute; at least for the particular SeriesInstanceUID I am looking at.
For me, it's failing in 322 images, for example:
\n'/kaggle/input/rsna-intracranial-aneurysm-detection/series/1.2.826.0.1.3.8.498./1.2.826.0.1.3.8.498..dcm'\nI found InstanceNumber to be helpful in sorting them. I only checked on a handful but on those it seemed to make sense.
Yes, you can sort the individual images within a series using InstanceNumber or ImagePositionPatient. This will allow you to create 3d image arrays.
Alternatives are:
\n1) Use alternative software with DICOM series reading functionality (e.g. simpleITK ImageSeriesReader()).
\n2) Use DICOM conversion software like dcm2niix (free binaries available with web search) to automatically convert to NIfTI, after which nibabel can be used to load the 3D image array.
@devinanzelmo, I tried InstanceNumber but at least to my eyes the order seems repetitive in some way with a random length, not sorted purely by brain position. Here is an image from the first few pictures from the first aneurysm sorted by ImageNumber. Thanks for your comment.
\n
Agree this does not look correctly sorted. Can you try sorting by ImagePositionPatient and/or provide a code snippet to replicate your findings?
Also, providing a single out of frame orthogonal slice (e.g. mid coronal or mid sagittal) will clearly show any sorting errors. For example using np.transpose() on the 3D array. This can be useful for quick reference.
I think ImagePositionPatient should work well, since its the corresponding metadata for that slice's coordiantes. I use instance number as a backup just in case, but ImagePositionPatient should generally work. This is the function I wrote that worked for me.
():\n \n dcm_files = [os.path.join(series_path, f) f os.listdir(series_path) f.endswith()]\n slices = [pydicom.dcmread(f) f dcm_files]\n\n :\n slices.sort(key= s: (s.ImagePositionPatient[]))\n (AttributeError, IndexError):\n slices.sort(key= s: (s.InstanceNumber))\n slices\n@umarali1 - that seems to work - thank you. Have you run into any image sets where the \"except\" block executes?
", "votes": 0, "replies": [ { "id": 3259032, "author_name": "Umar", "author_url": "", "post_date": "2025-07-31T18:07:54.287000", "content": "Nope, just checked and it seems like the except clause doesn't execute; at least for the particular SeriesInstanceUID I am looking at.
For me, it's failing in 322 images, for example:
\n'/kaggle/input/rsna-intracranial-aneurysm-detection/series/1.2.826.0.1.3.8.498./1.2.826.0.1.3.8.498..dcm'\nI found InstanceNumber to be helpful in sorting them. I only checked on a handful but on those it seemed to make sense.
Yes, you can sort the individual images within a series using InstanceNumber or ImagePositionPatient. This will allow you to create 3d image arrays.
Alternatives are:
\n1) Use alternative software with DICOM series reading functionality (e.g. simpleITK ImageSeriesReader()).
\n2) Use DICOM conversion software like dcm2niix (free binaries available with web search) to automatically convert to NIfTI, after which nibabel can be used to load the 3D image array.
@devinanzelmo, I tried InstanceNumber but at least to my eyes the order seems repetitive in some way with a random length, not sorted purely by brain position. Here is an image from the first few pictures from the first aneurysm sorted by ImageNumber. Thanks for your comment.
\n
Agree this does not look correctly sorted. Can you try sorting by ImagePositionPatient and/or provide a code snippet to replicate your findings?
Also, providing a single out of frame orthogonal slice (e.g. mid coronal or mid sagittal) will clearly show any sorting errors. For example using np.transpose() on the 3D array. This can be useful for quick reference.