I shared this kernel that coverts a folder of DICOM images to a single NifiTI file with dcmstack, in the hope that it might help others. Then I followed this method to train a 3d model and create a submission.csv.
\nI converted to NifiTI because I was struggling to come up with a method to standardize the shape of a 3D image (specifically the depth). I saw this method, but decide to try the NifiTi path.
", "messages": [ { "id": "1483559", "postDate": "08/20/2021 17:45:35", "content": "I shared this kernel that coverts a folder of DICOM images to a single NifiTI file with dcmstack, in the hope that it might help others. Then I followed this method to train a 3d model and create a submission.csv.
\nI converted to NifiTI because I was struggling to come up with a method to standardize the shape of a 3D image (specifically the depth). I saw this method, but decide to try the NifiTi path.
", "rawMarkdown": "I shared this [kernel](https://www.kaggle.com/ohbewise/dicom-to-nifti) that coverts a folder of DICOM images to a single NifiTI file with [dcmstack](https://dcmstack.readthedocs.io/en/v0.6.1/), in the hope that it might help others. Then I followed [this method](https://keras.io/examples/vision/3D_image_classification/) to train a 3d model and create a submission.csv.\n\nI converted to NifiTI because I was struggling to come up with a method to standardize the shape of a 3D image (specifically the depth). I saw [this method](https://www.kaggle.com/sentdex/first-pass-through-data-w-3d-convnet), but decide to try the NifiTi path.", "votes": null }, { "id": "1488280", "postDate": "08/24/2021 07:51:10", "content": "Hello. Most likely your method is not working well?
", "rawMarkdown": "Hello. Most likely your method is not working well?", "votes": null }, { "id": "1528409", "postDate": "09/29/2021 16:03:53", "content": "Dcmstack does a great job of converting DICOM to NifTi. What is done with the NifTi files after that is left to the user. There really isn't enough data to training a new model well, but transfer learning should work better.
", "rawMarkdown": "Dcmstack does a great job of converting DICOM to NifTi. What is done with the NifTi files after that is left to the user. There really isn't enough data to training a new model well, but transfer learning should work better.", "votes": null } ], "comments": [ { "id": 1488280, "author_name": "zaakciiru", "author_url": "", "post_date": "08/24/2021 07:51:10", "content": "Hello. Most likely your method is not working well?
", "votes": null, "replies": [ { "id": 1528409, "author_name": "ohbewise", "author_url": "", "post_date": "09/29/2021 16:03:53", "content": "Dcmstack does a great job of converting DICOM to NifTi. What is done with the NifTi files after that is left to the user. There really isn't enough data to training a new model well, but transfer learning should work better.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1483559": "I shared this [kernel](https://www.kaggle.com/ohbewise/dicom-to-nifti) that coverts a folder of DICOM images to a single NifiTI file with [dcmstack](https://dcmstack.readthedocs.io/en/v0.6.1/), in the hope that it might help others. Then I followed [this method](https://keras.io/examples/vision/3D_image_classification/) to train a 3d model and create a submission.csv.\n\nI converted to NifiTI because I was struggling to come up with a method to standardize the shape of a 3D image (specifically the depth). I saw [this method](https://www.kaggle.com/sentdex/first-pass-through-data-w-3d-convnet), but decide to try the NifiTi path.", "1488280": "Hello. Most likely your method is not working well?", "1528409": "Dcmstack does a great job of converting DICOM to NifTi. What is done with the NifTi files after that is left to the user. There really isn't enough data to training a new model well, but transfer learning should work better." }, "source": "meta" }