MAGMA goes Medi: Multimodal Augmentation through Adapter-based Fine-tuning for Radiology Report Generation

Data processing

The scripts used for data preprocessing are:

• DATA_combined_prepare.py
• DATA_iuxray_pepare.py
• DATA_mimix_prepare.py

Training

The training of the two training experiment are were done with the following: *train.py Configs for 1. Training: configs/MAGMA_medi_biomedlm_mimic.yml Configs for 2. Training: configs/MAGMA_medi_biomedlm_mimic.ym

Inference

Script used:

• medimagma_inference.py

Evaluation

Scripts used:

• preprocess_predictions.py
• all_timesteps_evaluate.py

Plots

All materials and code can be found in the plots/ folder.