article.md

We propose a new way of explainability for zero-shot diagnosis prediction in the clinical domain. Instead of directly predicting a diagnosis, we prompt the model to classify the existence of descriptive observations, which a radiologist would look for on an X-Ray scan, and use the descriptor probabilities to estimate the likelihood of a diagnosis, making our model explainable by design. For this we leverage BioVil, a pretrained CLIP model for X-rays and apply contrastive observation-based prompting. We evaluate Xplainer on two chest X-ray datasets, CheXpert and ChestX-ray14, and demonstrate its effectiveness in improving the performance and explainability of zero-shot diagnosis. Authors: Chantal Pellegrini, Matthias Keicher, Ege Özsoy, Petra Jiraskova, Rickmer Braren, Nassir Navab

License: MIT

Where to send questions or comments about the model: Open an issue on Xplainer repo.

Intended Use: This model is intended to be used solely for (I) future research on visual-language processing and (II) reproducibility of the experimental results reported in the reference paper.

Primary intended uses/users: Vision-Language and CAD researchers

Citation

@article{pellegrini2023xplainer,
  title={Xplainer: From X-Ray Observations to Explainable Zero-Shot Diagnosis},
  author={Pellegrini, Chantal and Keicher, Matthias and {\"O}zsoy, Ege and Jiraskova, Petra and Braren, Rickmer and Navab, Nassir},
  journal={arXiv preprint arXiv:2303.13391},
  year={2023}
}