SALAD: Score-based Adaptive Learning via Auxiliary Disease Classifier for Radiology Report Generation

Getting Started

Installation

1. Prepare the code and the environment

1. Git clone our repository and create a new conda environment.

conda create -n SALAD python=3.9 -y
conda activate SALAD

2. Install the requirements.

pip install -r requirements.txt

3. Modify the code of the Transformers library for score-based adaptive learning.

In transformers/models/llama/modeling_llama.py,
replace loss_fct = CrossEntropyLoss() with loss_fct = nn.CrossEntropyLoss(reduction='none')

2. Prepare the training dataset

IU-Xray

MIMIC-CXR

CheXpert Plus

Training

bash scripts/run_iu.sh

bash scripts/run_mimic.sh

bash scripts/run_chexpert.sh

Testing

bash scripts/test_iu.sh

bash scripts/test_mimic.sh

bash scripts/test_chexpert.sh

Acknowledgement

• R2GenGPT

License

This repository is under BSD 3-Clause License.