Gustavo Garzón, Santiago Gómez, Fabio Martínez
There exists a directory named model where you have to implement the desired architecture. Afterwards, a file named after train_$model_name.py should be created in the root of the project.
python run.py --config=configs/run.py:pix2pix,/home/sangohe/projects/lesion-aware-translation/data/APIS_synth-1_0_3_0_10_0_shuffledIn the examples directory are stored the data and masks for patients with ischemic stroke lesions. Below is table with the details of each patient and the information about the lesion. Note: the slice index number is for the 1mm$^3$ resampled volumes, the numbers in parenthesis are for the original modalities.
| Patient ID | Difficulty | Slice index | Observations |
|---|---|---|---|
| train_026 | Easy | 45 (7) | The NCCT presents an old lesion (easy difficulty) but the acute lesion is not visible (hard difficulty) |
| train_035 | Middle | 45 (7) | The NCCT presents a subtle hypoattenuation (middle difficulty) |
| train_044 | Hard | 99 (16) | The lesion is clearly visible in the ADC image but not on NCCT (hard difficulty) |
| train_058 | Easy | 104 (17) | Control patient, the model should not generate any lesion (easy difficulty) |
| test_022 | Easy | 69 (11) | The lesion is clearly visible in the NCCT (easy difficulty) |
| test_037 | Middle | 105 (17) | The lesion in the NCCT is located at a superior with a subtle hypoattentuation (middle difficulty) |
pip install runx pip install -qq git+https://github.com/keras-team/keras-cv@master
- Add more details about the run
- Add details about how to reproduce env
- Finish the
predict_and_savefunction and add the code to the train_*.py