This project was implemented using the Tensorflow framework.
Tensorflow 2.2.0, Python 3.7, cuda 10.2, cudnn 7.6.5
-
You will first need to request access for MIMIC-III, Single-level Clinical Classification Software(CCS) for ICD-9-CM, and Multi-level CCS.
-
Use "process_mimic.py" to process MIMIC-III dataset and generate a suitable training dataset.
-
Use "build_trees.py" to build files that contain the ancestor information of each medical code.
-
Use "process_treeseq.py" to generate knowledge sequences for model input. Use "process_label.py" to generate the labels.
-
Use "process_gcnlabel.py" to generate the labels required by the GCN embedding module, and use "process_adjacencylist.py" to generate the adjacency matrix required by the GCN embedding module.
-
Use "train.py" to generate GCN Embedding in the "gcn_embedding" folder. Before to execute this algorithm, it is necessary to install these required packages shown in the file named ' requirements.txt '. You can also install all the required packages by just using one command :
$ pip install -r requirements.txt -
Run DMKAP by using "run.py".