HiCu-ICD

This repo contains code for our MLHC 2022 paper HiCu: Leveraging Hierarchy for Curriculum Learning in Automated ICD Coding.

Setup

Install the following packages to run the code in this repository:

• gensim==4.1.2
• nltk==3.5
• numpy==1.18.1
• pandas==1.0.0
• scikit_learn==1.1.1
• scipy==1.4.1
• torch==1.7.1
• tqdm==4.62.3
• transformers==4.5.1

pip install -r requirements.txt

Data Preprocessing

We use MIMIC-III for model training and evaluation. We use the same data preprocessing code as MultiResCNN. To set up the dataset, place the MIMIC-III files into /data as shown below:

data
|   D_ICD_DIAGNOSES.csv
|   D_ICD_PROCEDURES.csv
└───mimic3/
|   |   NOTEEVENTS.csv
|   |   DIAGNOSES_ICD.csv
|   |   PROCEDURES_ICD.csv
|   |   train_full_hadm_ids.csv
|   |   train_50_hadm_ids.csv
|   |   dev_full_hadm_ids.csv
|   |   dev_50_hadm_ids.csv
|   |   test_full_hadm_ids.csv
|   |   test_50_hadm_ids.csv

The *_hadm_ids.csv files can be found here.

After setting up the files, run the following command to preprocess the data:

python preprocess_mimic3.py

Training

1. See files under /runs for training configs for MultiResCNN and RAC models.
2. For LAAT (Bi-LSTM) models, switch to LAAT branch and use the training configs in the root folder.

Acknowledgement

A large portion of the code in this repository is borrowed from foxlf823/Multi-Filter-Residual-Convolutional-Neural-Network . Thanks to their great work.