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.

A Hybrid Fully Convolutional CNN-Transformer Model for Inherently Interpretable Disease Detection from Retinal Fundus Images

Created by
  • Haebom

Author

Kerol Djoumessi, Samuel Ofosu Mensah, Philipp Berens

Outline

This paper proposes an interpretable hybrid model for medical image analysis that combines the local feature extraction capabilities of CNNs with the global dependency capture capabilities of ViT. To overcome the interpretability challenges of existing hybrid models, we developed a fully convolutional CNN-Transformer architecture that considered interpretability from the design stage and applied it to retinal disease detection. The proposed model outperforms existing black-box and interpretable models in predictive performance and generates class-specific sparse evidence maps in a single pass. The code is available on GitHub.

Takeaways, Limitations

•
Takeaways:
◦
We present an effective implementation of an interpretable hybrid CNN-Transformer model for medical image analysis.
◦
Achieving both superior predictive performance and interpretability compared to existing black box models.
◦
Transparently exposes the model's decision-making process by generating class-specific sparse evidence maps in a single pass.
◦
Ensure reproducibility and scalability through open code.
•
Limitations:
◦
The performance of the proposed model is limited to a specific medical image analysis task (retinal disease detection).
◦
Generalization performance to other types of medical images or diseases requires further study.
◦
Quantitative assessment of the interpretation accuracy of the evidence map may be lacking.
👍
Made with Slashpage