[공지사항]을 빙자한 안부와 근황 
Show more

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.

CXR-TFT: Multi-Modal Temporal Fusion Transformer for Predicting Chest X-ray Trajectories

Created by
  • Haebom

Author

Mehak Arora, Ayman Ali, Kaiyuan Wu, Carolyn Davis, Takashi Shimazui, Mahmoud Alwakeel, Victor Moas, Philip Yang, Annette Esper, Rishikesan Kamaleswaran

Outline

In this paper, we propose a multimodal framework, CXR-TFT, for predicting clinical outcomes in intensive care unit (ICU) patients. CXR-TFT predicts changes in CXR findings in critically ill patients by integrating temporally irregularly acquired chest X-ray images (CXRs), radiology reports, and high-frequency clinical data such as vital signs, laboratory results, and respiratory flow charts. The latent vectors extracted from the image encoder are combined with temporally consistent clinical data through temporal interpolation, and the CXR latent vectors are predicted hourly by a transformer model conditioned on previous latent vectors and clinical measurements. In a retrospective study of 20,000 ICU patients, CXR-TFT has been shown to predict abnormal CXR findings with high accuracy up to 12 hours before radiological presentation. This has significant potential for improving the management of time-sensitive diseases such as acute respiratory distress syndrome, where early intervention is crucial and diagnosis is often delayed.

Takeaways, Limitations

•
Takeaways:
◦
We present a novel method to predict the clinical course of critically ill patients by integrating temporally irregular CXR data and diverse clinical data.
◦
Contributes to improving clinical outcomes by enabling early diagnosis and intervention of time-sensitive diseases such as acute respiratory distress syndrome.
◦
Bringing temporal resolution to CXR analysis, providing actionable insights into the ‘whole patient’.
◦
Demonstrated predictive power of abnormal CXR findings up to 12 hours in advance.
•
Limitations:
◦
Based on a retrospective study, verification through prospective studies is needed.
◦
Further studies are needed to evaluate the model's generalization performance and applicability to different patient populations.
◦
Further analysis is needed on the possible errors and their impact during time interpolation.
◦
Applicability evaluation for various medical institutions and datasets is required.
👍
Made with Slashpage