The Lead Episode 157: A Discussion of Predicting Sudden Cardiac Death in Patients With Sarcoidosis Using a Multimodal Artificial Intelligence Model-The Lead Episode 157 Bonus Video File

Video Transcription

Video Summary

This LEAD podcast episode discusses a study on predicting sudden cardiac death in cardiac sarcoidosis using a multimodal AI model, MARS-CS. The guests explain that the model combines structured electronic health record data with raw late gadolinium enhancement cardiac MRI images, analyzed through neural networks. In a cohort of 317 patients followed for about 8.5 years, the fused AI model outperformed left ventricular ejection fraction alone, reaching an AUROC of 0.86 versus about 0.77 for LVEF. It identified additional high-risk patients who might otherwise be missed by standard EF-based thresholds. The panel highlights promising explainability results, but also important limitations: single-center design, small event numbers, possible overfitting, and the need for external validation. They emphasize that AI may improve ICD decision-making in cardiac sarcoidosis, but broader adoption will require prospective testing, standardization of imaging inputs, and comparison with current guideline-based approaches.

Keywords

cardiac sarcoidosis

sudden cardiac death

multimodal AI model

MARS-CS

late gadolinium enhancement MRI

left ventricular ejection fraction