From 56cc31b86e82e78791c2915af74463d9fd50cbb4 Mon Sep 17 00:00:00 2001 From: taliq Date: Thu, 9 Jun 2022 09:09:27 +0900 Subject: [PATCH] Update README.md --- README.md | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-) diff --git a/README.md b/README.md index 851efa2..05d8e54 100644 --- a/README.md +++ b/README.md @@ -5,7 +5,7 @@ Graph deep learning on whole slide image predicts the context-aware prognostic p * To install the dependencies for this project, see the "requirements.yaml" * Tested on Nvidia TESLA V100 x 2 with CUDA 11.1 -## Processing whole slide image (WSI) into superpatch-graph +## Step 1: Processing whole slide image (WSI) into superpatch-graph #### What is the superpatch-graph? * Superpatch-graph is the compressed representation of whole slide image into graph structure in memory efficient manner. * Run the ./Superpatch_network_construction/supernode_generation.py @@ -16,13 +16,13 @@ Graph deep learning on whole slide image predicts the context-aware prognostic p * Node position information in "_node_location_list.csv" * Superpatch aggregated dictionary in "_artifact_sophis_final.csv" -## Training TEA-graph using superpatch-graph +## Step 2: Training TEA-graph using superpatch-graph * Users can predict the prognosis of entire host with tumor environment-associated context analysis using deep graph learning (TEA-graph) * Run the ./main.py with appropriate hyperparameters * Users can simply run the above script with pre-defined parameters and datasets * Or, users can use their own dataset preprocessed by "supernode_generation" script -## Visualization of IG (Integrated gradients) value on WSI +## Step 3: Visualization of IG (Integrated gradients) value on WSI * Users can visualize the IG value which is highly correlated with risk value of each region in WSI * Also, we provide subgraph-level contextual pathological feature extraction * Run the ./IG_attention_feature_cal_main.py with same parameters you used for training your own TEA-graph model @@ -38,9 +38,15 @@ Graph deep learning on whole slide image predicts the context-aware prognostic p * "IG_again" directory is also created inside each patient's folder * "_IG_TME_subgraph.csv" indicates the each IG group's subgraph +## Step 4: Biomarker discovery +* Users can extract the contextual biomarker using the calculated IG values and extracted feature at the previous step +* Run the ./Context_marker_discovery_main.py with approprate directory path +* Users can obtain the several candidate pathology images with visualized graph for contextual biomarker + + ## Acknowledgments * http://github.com/mahmoodlab/Patch-GCN * http://github.com/lukemelas/EfficientNet-PyTorch * http://github.com/pyg-team/pytorch_geometric -BiNEL (http://binel.snu.ac.kr) - This code is made available under the MIT License and is available for non-commercial academic purposes \ No newline at end of file +BiNEL (http://binel.snu.ac.kr) - This code is made available under the MIT License and is available for non-commercial academic purposes