GANDALF: Graph-based transformer and Data Augmentation Active Learning Framework with interpretable features for multi-label chest Xray classification

Dwarikanath Mahapatra, Behzad Bozorgtabar, Zongyuan Ge, Mauricio Reyes

Abstract

Highlights•Graph attention transformers to incorporate the importance of different nodes.•Previous aggregation methods do not emphasize informative nodes.•A novel multi-label informativeness score to quantify importance of each sample.•The multi-label informativeness score is derived from graph attention transformers.•A novel data augmentation approach to generate new synthetic images•New informative images enforce class label preservation and redundancy avoidance.•Proposed approach outperforms SOTA methods for multi-label classification.
BibTeX
@article{mahapatra2024gandalf,
  author = {Dwarikanath Mahapatra and Behzad Bozorgtabar and Zongyuan Ge and Mauricio Reyes},
  title = {{GANDALF: Graph-based transformer and Data Augmentation Active Learning Framework with interpretable features for multi-label chest Xray classification}},
  journal = {Medical Image Analysis},
  year = {2024},
  url = {https://openreview.net/forum?id=Mqdg8G8L59},
}

Loading related content...

Figure