An EfficientNetV2-based survival benefit prediction system was developed to predict the additional survival benefit of EGFR-TKIs and ICIs in stage IV NSCLC patients.
main.py: Includes the entry function of ESBP and the code for training and validation.
input_dir_train: The path where your training dataset is stored.
input_dir_val: The path where your validation dataset is stored.
input_dir_test: The path where your test dataset is stored.
- For the reproducibility, a subdataset (anonymized) is made publicly accessible in this repository. Readers can directly use this dataset to run the source code of the ESBP in this study.
HDF5_read.py: Defines the function of CT image reading.
train_data: The input for ESBP. Examples of the input images are presented in Fig. S1 below.
target_data: The label for training and test.
keys[index]: The name of each image. All data are anonymized, and P1_1 represents the first image of the first patient.
Weight folder: Download the "best_pt_OA.pt.tar.gz*" files and use the following command to extract the well-trained ESPS model.
cat best_pt_OA.pt.tar.gz* | tar -xzv
nets folder: The definition of the ESBP network.
utils folder: Necessary functions used in the ESBP.
Test folder: The publicly accessible test data for reproducibility. Due to the size limitation, use the following command to extract the data.
cat Open_Access_Data.hdf5.tar.gz* | tar -xzv
Fig. S1. Examples of the input images of ESBP. a(1) to a(4) and a(5) to a(8) represent two different patients.
