DigestAI news desk
Research updated

Latent-Attention Masked Autoencoders Boost Multimodal Cardiac AI on 1.2M Hospital Stays

A new self‑supervised model called Latent‑Attention Masked Autoencoders (LAMAE) learns patient‑level representations from multiple cardiac data streams—ECG, echo, chest X‑ray, and clinical variables—by exchanging information in a shared latent‑attention module. Trained on more than 1.2 million MIMIC‑IV hospital stays, LAMAE outperforms modality‑specific pre‑training and strong contrastive or…

1 source primary source

Key points

  • LAMAE learns joint representations from ECG, echo, X‑ray, and clinical data via a shared latent‑attention module.
  • Pre‑trained on 1.2 M MIMIC‑IV stays, it beats modality‑specific and contrastive baselines on mortality, coding, and LOS tasks.
  • Model remains effective with missing modalities, enabling flexible deployment in real‑world hospitals.

The study highlights the advantage of modeling both intra‑modal structure and inter‑modal relationships, offering a more robust and transferable foundation for clinical AI. By integrating the evidence clinicians naturally combine, LAMAE could streamline multi‑modal diagnostic pipelines and reduce the need for separate models per data type.

Read the original at arXiv cs.AI · by Andrea Agostini, Simon B\"ohi, Moritz Vandenhirtz, Samuel Ruiperez-Campillo, Max Kr\"ahenmann, Silke M\"uhlstedt, Irene Cannistraci, Ece \"Ozkan Elsen, Julia E. Vogt, Thomas M. Sutter primary source Open source ↗
Topics · follow one to build your own front page

The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us.

Comments

via GitHub Discussions

More in Research

All →

Related stories