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.

Towards Cardiac MRI Foundation Models: Comprehensive Visual-Tabular Representations for Whole-Heart Assessment and Beyond

Created by
  • Haebom

Author

Yundi Zhang, Paul Hager, Che Liu, Suprosanna Shit, Chen Chen, Daniel Rueckert, Jiazhen Pan

Outline

In this paper, we present a foundational model, ViTa, that integrates cardiac magnetic resonance imaging (CMR) and patient-level health factors to enable a comprehensive understanding of cardiac health and an accurate interpretation of individual disease risk. Using data from 42,000 UK Biobank participants, we integrate 3D+T cine stacks of short- and long-axis images with detailed tabular patient-level factors. This multimodal paradigm supports multiple subtasks, including cardiac phenotype and physiological feature prediction, segmentation, and cardiac and metabolic disease classification, within a single, unified framework. By learning a shared latent representation that connects rich image features with patient context, we demonstrate the potential to go beyond existing task-specific models to provide a general, patient-specific understanding of cardiac health, enhancing clinical utility and scalability of cardiac analytics.

Takeaways, Limitations

•
Takeaways:
◦
Introducing a new model, ViTa, that provides a comprehensive understanding of cardiac health by integrating CMR images with various patient-level information.
◦
A single, integrated framework capable of performing diverse tasks including cardiac phenotype prediction, segmentation, and disease classification.
◦
Goes beyond existing task-specific models to provide patient-specific understanding of cardiac health, suggesting potential for improved clinical utility and scalability.
◦
Validation of model performance and generalization ability using a large dataset from the UK Biobank.
•
Limitations:
◦
This paper does not present specific numerical results on the performance of the ViTa model.
◦
Further validation of the model's generalization performance is needed, particularly a comparative analysis of its performance on the UK Biobank dataset and other datasets.
◦
Further research is needed on the interpretability and reliability of the model. The explainability of the predicted results needs to be improved.
👍
Made with Slashpage