Prospective clinical impact of MARCUS

Determine whether deploying MARCUS (Multimodal Autonomous Reasoning and Chat for Ultrasound and Signals), an agentic multimodal vision-language system for interpreting electrocardiograms, echocardiograms, and cardiac magnetic resonance imaging, improves patient outcomes, reduces diagnostic errors, and shortens time to treatment when assessed in prospective clinical trials.

Background

MARCUS is presented as an agentic, multimodal vision-LLM that performs end-to-end interpretation of ECG, echocardiography, and CMR and demonstrates strong retrospective performance across internal and external datasets. However, the study design is retrospective and focuses on benchmarked questions rather than real-time clinical deployment.

The authors explicitly note that it remains unclear whether the use of MARCUS translates into measurable clinical benefits such as improved outcomes, fewer diagnostic errors, or faster treatment initiation. They state that this must be established through prospective trials, highlighting a key unresolved question for clinical validation and adoption.