A biologically interpretable integrative deep learning model that integrates PAthological images and GEnomic data
- Patch-wise pre-trained CNN
- Two-stage Aggeragation
- Integration of aggregated pathological images and genomic data
- patch_extraction : it extracts valid patchs from WSI by removing background and stains.
- Datagenerator.py : it makes the data ready for Keras image data loaders. (As data is large we used keras dataloaders for loading data)
- PAGE_net_pretrain : code for pretraining and saving the pretrained model.
- Aggergation.py : it generates the aggregated score and saves in csv.
- DataMatching.py : it splits the aggregated data in train, test, and validation
Run.py: to train the model with the inputs from train.csv. Hyperparmeters are optimized by grid search automatically with validation.csv. C-index is used to evaluate the model performance with test.csv.
#Package Requirments ###Preprocessing
- PyHistopathology package
- Keras
- Tensorflow
- Keract