1. Prepare these data to generate DL input in the "raw" folder
(1) Brain mask
(2) AIF mask
(3) Input: 4D DSC-MRI
(4) Ground truth: Derived parameters (generated via the RAPID)
2. Scripts for generating DL input
Run from_mat_to_npy.py
a. You need to run this script at least twice. 
(1) Generate normal and lesion ROIs (for training and validation sets).
(2) Generate WB (for testing set)
b. You can define your input type, ROIs ratio, and patch steps via the commands. Example is as the following.
Setting: Train/Val set, OP_NM ratio 1:2, patch steps 3
Command: python3 from_mat_to_npy.py --save_path=ST_pad_InRawHighDCE_GtRaw_KtransTh/ --output=’OP_NM_CAhigh’ --ratio 1 2 --patch_steps 3

3. Run ST-Net model
(1). How to run the model: open a terminal, conda activate your envieonment, and run "python train.py", the results will be saved in exp folder
(2). config.toml: set hyperparameters
	- ckpt_path: If you want to train the modek from scratch, please set this value as 0. If you want to load checkpoint, please indicate the checkpoint folder.
	- fold_excel: Generate an .xlsx file to set the training or testing subjects.
	- curr_fold: Indicate which fold you are going to run
	- Augmentation: If you want to use augmentation, please set this value as 1; otherwise, please set it as 0.
(3). utils/dataset_ST.py: dataloader
(4). The deep learning architecture is in model folder and the results are saved in exp folder
(5). plot_loss.py: open a terminal and run $ python plot_loss.py --model_path='{your model folder name in the exp folder}'
(6). How to run the testing using pretrained weight: When you train the model, a config.toml will save in the result folder and will be used in the testing. Run $ python test.py --model_path='{your model folder name in the exp folder}'. Terminal will also show the performance for each testing data.
(7). If you have already run the test.py and only want to print out the testing performance, run $ python print_metrics.py