{ "id": 272876, "title": "A RSNA MRI Solution: From DICOM To Submission", "url": "/competitions/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/272876", "author_name": "Laynr", "post_date": "2021-09-17T22:34:31.444000", "votes": 8, "comment_count": 0, "views": 0, "content": "

I posted A RSNA MRI Solution: From DICOM To Submission, a complete solution kernel for the 2021 RSNA-MICCAI Brain Tumor Radiogenomic Classification Kaggle competition. This is definitely not a winning solution, but is offered as an example of an end-to-end solution from DICOM to submission. It uses TorchIO for data manipulation, NiBabel for reading NifiTi images, and Tensorflow to train a 3d Constitutional Neural Network from a keras.io example.

\n

Predictions are only slightly better than a coin flip, but the notebook is broken out into logical tasks in the following standalone notebooks so they can be easily swapped out and improved.

\n
    \n
  1. DICOM to normalized NifiTi with TorchIO
    2. \n
  2. NifiTi to Split Dataset with NiBabel
    3. \n
  3. Dataset to Model with Tensorflow
    4. \n
  4. Model Prediction to Submission
    5. \n
\n

And again, the link to the complete notebook: A RSNA MRI Solution: From DICOM To Submission

\n

Note with demo=False the complete notebook errors out with Your notebook tried to allocate more memory than is available but when ran in the steps above it complete successfully. Hope this is helpful to others struggling with any step along the process.

", "messages": [ { "id": 1516104, "postDate": "2021-09-17T22:34:31.443Z", "content": "

I posted A RSNA MRI Solution: From DICOM To Submission, a complete solution kernel for the 2021 RSNA-MICCAI Brain Tumor Radiogenomic Classification Kaggle competition. This is definitely not a winning solution, but is offered as an example of an end-to-end solution from DICOM to submission. It uses TorchIO for data manipulation, NiBabel for reading NifiTi images, and Tensorflow to train a 3d Constitutional Neural Network from a keras.io example.

\n

Predictions are only slightly better than a coin flip, but the notebook is broken out into logical tasks in the following standalone notebooks so they can be easily swapped out and improved.

\n
    \n
  1. DICOM to normalized NifiTi with TorchIO
    2. \n
  2. NifiTi to Split Dataset with NiBabel
    3. \n
  3. Dataset to Model with Tensorflow
    4. \n
  4. Model Prediction to Submission
    5. \n
\n

And again, the link to the complete notebook: A RSNA MRI Solution: From DICOM To Submission

\n

Note with demo=False the complete notebook errors out with Your notebook tried to allocate more memory than is available but when ran in the steps above it complete successfully. Hope this is helpful to others struggling with any step along the process.

", "rawMarkdown": "I posted [A RSNA MRI Solution: From DICOM To Submission](https://www.kaggle.com/ohbewise/a-rsna-mri-solution-from-dicom-to-submission), a *complete* solution kernel for the 2021 RSNA-MICCAI Brain Tumor Radiogenomic Classification Kaggle competition. This is definitely not a winning solution, but is offered as an example of an end-to-end solution from DICOM to submission. It uses **TorchIO** for data manipulation, **NiBabel** for reading NifiTi images, and **Tensorflow** to train a 3d Constitutional Neural Network from a **keras.io** example.\n\nPredictions are only slightly better than a coin flip, but the notebook is broken out into logical tasks in the following standalone notebooks so they can be easily swapped out and improved.\n\n1. [DICOM to normalized NifiTi with TorchIO](https://www.kaggle.com/ohbewise/dicom-to-normalized-nifiti-with-torchio)\n1. [NifiTi to Split Dataset with NiBabel](https://www.kaggle.com/ohbewise/nifiti-to-split-dataset-with-nibabel)\n1. [Dataset to Model with Tensorflow](https://www.kaggle.com/ohbewise/dataset-to-model-with-tensorflow)\n1. [Model Prediction to Submission](https://www.kaggle.com/ohbewise/model-prediction-to-submission)\n\nAnd again, the link to the complete notebook: [A RSNA MRI Solution: From DICOM To Submission](https://www.kaggle.com/ohbewise/a-rsna-mri-solution-from-dicom-to-submission)\n\nNote with `demo=False` the complete notebook errors out with `Your notebook tried to allocate more memory than is available` but when ran in the steps above it complete successfully. Hope this is helpful to others struggling with any step along the process.", "votes": 8 } ], "comments": [], "raw_markdown_by_id": { "1516104": "I posted [A RSNA MRI Solution: From DICOM To Submission](https://www.kaggle.com/ohbewise/a-rsna-mri-solution-from-dicom-to-submission), a *complete* solution kernel for the 2021 RSNA-MICCAI Brain Tumor Radiogenomic Classification Kaggle competition. This is definitely not a winning solution, but is offered as an example of an end-to-end solution from DICOM to submission. It uses **TorchIO** for data manipulation, **NiBabel** for reading NifiTi images, and **Tensorflow** to train a 3d Constitutional Neural Network from a **keras.io** example.\n\nPredictions are only slightly better than a coin flip, but the notebook is broken out into logical tasks in the following standalone notebooks so they can be easily swapped out and improved.\n\n1. [DICOM to normalized NifiTi with TorchIO](https://www.kaggle.com/ohbewise/dicom-to-normalized-nifiti-with-torchio)\n1. [NifiTi to Split Dataset with NiBabel](https://www.kaggle.com/ohbewise/nifiti-to-split-dataset-with-nibabel)\n1. [Dataset to Model with Tensorflow](https://www.kaggle.com/ohbewise/dataset-to-model-with-tensorflow)\n1. [Model Prediction to Submission](https://www.kaggle.com/ohbewise/model-prediction-to-submission)\n\nAnd again, the link to the complete notebook: [A RSNA MRI Solution: From DICOM To Submission](https://www.kaggle.com/ohbewise/a-rsna-mri-solution-from-dicom-to-submission)\n\nNote with `demo=False` the complete notebook errors out with `Your notebook tried to allocate more memory than is available` but when ran in the steps above it complete successfully. Hope this is helpful to others struggling with any step along the process." } }