MOSCARD - Multimodal Opportunistic Screening for Cardiovascular Adverse events with Causal Reasoning and...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Major Adverse Cardiovascular Events (MACE) are a leading global cause of mortality.
• Opportunistic screening using routine health data can identify individuals at risk.
• Current risk assessment models have limitations due to sampling bias and single-modality constraints.

Purpose of the Study:

• To propose a novel predictive modeling framework, MOSCARD, for opportunistic cardiovascular risk estimation.
• To integrate multimodal data (CXR and ECG) with causal reasoning for enhanced risk assessment.
• To mitigate bias and confounders in opportunistic cardiovascular risk prediction.

Main Methods:

• Developed MOSCARD, a multimodal causal reasoning framework with co-attention.
• Implemented multimodal alignment of Chest X-rays (CXR) with electrocardiogram (ECG) guidance.
• Utilized a dual back-propagation graph for deconfounding biases.

Main Results:

• MOSCARD demonstrated superior performance compared to single-modality and state-of-the-art models.
• Achieved Area Under the Curve (AUC) scores of 0.75 (internal), 0.83 (shift data), and 0.71 (external MIMIC).
• The model effectively aligned CXR and ECG data while mitigating confounding factors.

Conclusions:

• The proposed MOSCARD framework offers a cost-effective approach to opportunistic cardiovascular risk screening.
• Integrating multimodal data and causal reasoning improves the accuracy of identifying at-risk individuals.
• Early intervention enabled by this screening can improve patient outcomes and reduce health disparities.