Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Causal Representation Learning with Observation Grouping for CXR Classification

Created by
  • Haebom

Author

Rajat Rasal, Avinash Kori, Ben Glocker

Outline

In this paper, we present a method to improve the generalization performance and robustness of disease classification by learning identifiable causal representations in medical images. In particular, we introduce an end-to-end framework to learn identifiable representations by grouping observations in chest X line images and enhancing invariance to race, gender, and image view. Experimental results show that the causal representations learned through grouping improve the generalization performance and robustness in various classification tasks.

Takeaways, Limitations

•
Takeaways:
◦
We highlight the importance of causal representation learning in medical image analysis and suggest practical applications.
◦
We show that grouping techniques can learn more fair and generalizable models by controlling for potential confounders such as race, gender, and image view.
◦
We present an effective method to improve generalization performance and robustness in various disease classification tasks.
•
Limitations:
◦
The performance of the presented framework may depend on a specific dataset.
◦
Further research is needed on optimizing grouping strategies and exploring different grouping methods.
◦
Further validation of generalizability to other medical imaging modalities or disease types is needed.
👍
Made with Slashpage