Multimodal Cardiovascular Risk Profiling Using Self-Supervised Learning of Polysomnography

2025-07-11

AI Summary

Current cardiovascular disease (CVD) individualized risk assessment suffers from insufficient accuracy. Method: We propose a self-supervised representation learning framework leveraging multimodal sleep physiological signals—electroencephalography (EEG), electrocardiography (ECG), and respiration—to extract cardiovascular risk features without clinical labels. Our approach employs a contrastive embedding-driven projection scoring model trained on unlabeled polysomnography data, jointly learning cross-modal risk representations, which are then fused with the classical Framingham Risk Score to form an integrated predictive model. Contribution/Results: This work is the first to enable label-free, multimodal sleep signal–driven CVD risk stratification, uncovering modality-specific associations: EEG with incident hypertension, ECG with cardiac events, and respiratory signals with incremental risk gain. In an independent validation cohort, the integrated model achieves AUCs of 0.607–0.965, significantly outperforming conventional models and establishing a novel, interpretable, and generalizable biomarker system for early CVD warning.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multimodal Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization

Abstract

Methods: We developed a self-supervised deep learning model that extracts meaningful patterns from multi-modal signals (Electroencephalography (EEG), Electrocardiography (ECG), and respiratory signals). The model was trained on data from 4,398 participants. Projection scores were derived by contrasting embeddings from individuals with and without CVD outcomes. External validation was conducted in an independent cohort with 1,093 participants. The source code is available on https://github.com/miraclehetech/sleep-ssl. Results: The projection scores revealed distinct and clinically meaningful patterns across modalities. ECG-derived features were predictive of both prevalent and incident cardiac conditions, particularly CVD mortality. EEG-derived features were predictive of incident hypertension and CVD mortality. Respiratory signals added complementary predictive value. Combining these projection scores with the Framingham Risk Score consistently improved predictive performance, achieving area under the curve values ranging from 0.607 to 0.965 across different outcomes. Findings were robustly replicated and validated in the external testing cohort. Conclusion: Our findings demonstrate that the proposed framework can generate individualized CVD risk scores directly from PSG data. The resulting projection scores have the potential to be integrated into clinical practice, enhancing risk assessment and supporting personalized care.

Problem

• Develop self-supervised model for CVD risk using PSG data
• Predict cardiac conditions via ECG and EEG features
• Enhance Framingham Risk Score with multimodal signals

Innovation

• Self-supervised deep learning model extracts patterns
• Combines multi-modal signals for CVD risk profiling
• Integrates projection scores with Framingham Risk Score