MARCUS: An agentic, multimodal vision-language model for cardiac diagnosis and management

2026-03-23 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, AI in Healthcare · Depth: Expert, quick

Summary

MARCUS (Multimodal Autonomous Reasoning and Chat for Ultrasound and Signals) is an agentic vision-language system designed for comprehensive interpretation of cardiac tests, including electrocardiograms (ECGs), echocardiograms, and cardiac magnetic resonance imaging (CMR). This system processes single-modality inputs and integrates multimodal data, addressing limitations of current AI models. MARCUS utilizes a hierarchical agentic architecture with modality-specific vision-language expert models, each featuring domain-trained visual encoders and multi-stage language model optimization, all coordinated by a multimodal orchestrator. Trained on 13.5 million images and a 1.6 million question expert-curated dataset, MARCUS achieved state-of-the-art performance, surpassing frontier models like GPT-5 Thinking and Gemini 2.5 Pro Deep Think. It demonstrated 87-91% accuracy for ECG, 67-86% for echocardiography, and 85-88% for CMR, outperforming frontier models by 34-45%. For multimodal cases, MARCUS achieved 70% accuracy, nearly tripling frontier models' 22-28%, and showed 1.7-3.0x higher free-text quality scores. The agentic design also provides resistance to "mirage reasoning."

Key takeaway

For AI Scientists developing diagnostic tools in cardiology, MARCUS demonstrates that an agentic, multimodal approach significantly improves accuracy and reasoning quality over current frontier models. You should consider adopting hierarchical agentic architectures with domain-specific visual encoders to enhance the reliability and performance of your medical AI systems, especially when integrating diverse data types. This approach can lead to more robust and clinically useful diagnostic support.

Key insights

MARCUS is an agentic, multimodal vision-language model excelling in cardiac diagnosis by integrating diverse imaging data.

Principles

• Agentic architecture enhances multimodal interpretation.
• Domain-specific encoders improve diagnostic accuracy.
• Multimodal integration outperforms single-modality analysis.

Method

MARCUS employs a hierarchical agentic architecture with modality-specific vision-language expert models, integrating domain-trained visual encoders and multi-stage language model optimization, coordinated by a multimodal orchestrator.

In practice

• Interpret ECGs, echocardiograms, and CMR images.
• Combine multiple cardiac imaging modalities.
• Reduce "mirage reasoning" in diagnostic AI.

Topics

• MARCUS
• Multimodal AI
• Cardiac Diagnosis
• Vision-Language Models
• Agentic AI

Best for: AI Scientist, AI Researcher, AI Engineer, Research Scientist

Related on AIssential

• Evaluating Multi-Agent LLM Architectures for Rare Disease Diagnosis — Multi-agent LLM architectures do not guarantee improved diagnostic accuracy; topology selection is critical.
• Agentic AI Vision System: Object Segmentation with SAM 3 and Qwen — Agentic AI systems combine VLMs and segmentation models for iterative, self-correcting visual understanding and object segmentation.
• The Cost of Language: Centroid Erasure Exposes and Exploits Modal Competition in Multimodal Language Models — Language representations in MLMs systematically overshadow vision, causing underperformance in visual tasks.
• A-MAR: Agent-based Multimodal Art Retrieval for Fine-Grained Artwork Understanding — Explicit reasoning plans improve multimodal art retrieval, enabling grounded, interpretable artwork understanding.
• From Conflict to Consensus: Boosting Medical Reasoning via Multi-Round Agentic RAG — MA-RAG (Multi-Round Agentic RAG) is introduced to enhance medical reasoning in Large Language Models (LLMs) by mitigating issues like hallucinations and outdated knowledge, which traditional Retrieval…
• Agentic Authoring of Interactive Multiview Visualizations in Genomics — Agentic LLM iteration significantly enhances genomics visualization quality, but complex agent architectures offer no further benefit.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.