| Title: | TrustCardio: A Multimodal Large Language Model with Chain-of-Thought Explainability for Personalized Vascular Age Prediction via Edge-Enabled Wearable Devices |
| Journal: | IEEE Transactions on Consumer Electronics |
| Published: | 14 Jul 2026 |
| DOI: | https://doi.org/10.1109/tce.2026.3712612 |
Publication 20364
WARNING: the interactive features of this website use CSS3, which your browser does not support. To use the full features of this website, please update your browser.
Abstract
Next-generation connected healthcare relies on consumer electronics platforms for real-time, customized health monitoring through wearable devices. Existing methods are hindered by inability to provide chain-of-thought explanations, non-personalized models, and reliance on centralized systems limiting responsiveness and data privacy. Our proposed solution, TrustCardio, is a multimodal approach to predicting vascular age with chain-of-thought explainability for edge-based wearables. Rather than a general-purpose multimodal LLM, TrustCardio is a language-model-augmented framework coupling a signal encoder with a biomedical-LM-initialized reasoning decoder. TrustCardio comprises three core innovations: (i) an individualized multimodal pipeline fusing photoplethysmography (PPG) with demographic factors through cross-attention techniques; (ii) a hierarchical chain-of-thought explainability solution generating reasoning aligned with clinicians' understanding of vascular age; and (iii) optimized deployment on edge devices. TrustCardio has been validated on the UK Biobank (N = 98, 672), MIMIC-III Waveform Database (N = 10, 282), and PulseDB (N = 5, 361) datasets, achieving a mean absolute error (MAE) of 6.54 years (9.0% better than previous best methods), 91.3% clinician agreement, and 38.7% reduction in inference latency on resource-constrained edge devices. These results demonstrate the feasibility of deploying clinically meaningful vascular age estimation (trained against the chronological age of comorbidity-free individuals as a surrogate) on consumer wearable devices. TrustCardio advances explainability and individualization in connected healthcare.</p>
14 Authors
- Ji Li
- Jun Zhang
- Qunchai Ran
- Chuan Yu
- Xiaoyu Zhou
- Mingju He
- Keyue Han
- Wei Zhang
- Fei Yang
- Shi Peng
- Zhe Cao
- Hui Shi
- Shujuan Xu
- Bihua Xia
Enabling scientific discoveries that improve human health