Open kaphleamrit2 opened 3 years ago
Hi Amrit,
We are training CNN models in TensorFlow 2 (TF2) and Pytorch, and an XGBoost model using radiomics features. Source code is mostly in notebooks.
Training source code in TF2 main file is tf2_transfer_learning_gamma.ipynb. Fluence map files are read from a TensorFlow dataset that is created using source code tf2_create_tfrecords.ipynb. tf2_create_tfrecords.ipynb reads RT Image DICOM files and generates a training/validation partition in a TF dataset. Predictions for validation maps are computed with the notebook tf2_evaluate.ipynb.
Notebook pytorch-fluence-maps.ipynb trains models using Pytorch; predictions for the validation maps are computed in this same notebook.
And notebook rtPyRadiomicsExample.ipynb computes radiomic features from image fluence maps using the pyradiomics library. At the end of this notebook an XGBoost model is trained using cross-validation.
Most hyperparameters models are defined with variables named ARG_NNN (ex.: ARG_NETWORK, ARG_EPOCHS_WO_IMPROVEMENT, etc.) at the beginning for training notebook files.
Regards, Pablo
Hi Pablo,
Thank you for providing the description of the project. It helps a lot to understand the workflow of the project.
Thanks. Amrit
Hi Pablojrios,
I am interested in this project. There are lots of stuff here and mixed up. I cannot figure it out where to start. So, could you please give an instruction on how to use these code stepwise to predict the gamma index from fluence map so that other can be benefited from your work.
Thanks a lot for this nice work.
Amrit