Abstract
This research presents multimodal deep learning (DL) for structural heart disease prediction. We introduced novel model architecture with dual features encoders for extracting relevant features from both waveform data and clinical features in analyzing the EchoNext dataset. We evaluated multiple deep learning architectures trained with our proposed Multi-Fusion architecture, including TCN, Simple CNN, ResNet1d18, Light transformer and Hybrid model and compare our results with the existing results. The models were examined across the multiple seeds to ensure robustness, and bootstrap confidence interval is used to measure significant performance. From our results, the TCN model consistently outperforms other competing architectures with an AUC of 86.02% and PRAUC of 82.36% demonstrating a better performance than the pioneer method. Similarly in predictive analysis, TCN has efficient computation and stable training compared to all competing architectures. TCN model also outperforms other competing models in fairness evaluation across the race and sex groups.
| Original language | English |
|---|---|
| Article number | 100703 |
| Number of pages | 18 |
| Journal | Smart Health |
| Volume | 42 |
| DOIs | |
| State | Published - Dec 2026 |
Bibliographical note
Publisher Copyright:© 2026 Elsevier Inc. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
Funding
The authors received no Funding for this work. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
ASJC Scopus Subject Areas
- Medicine (miscellaneous)
- Information Systems
- Health Informatics
- Computer Science Applications
- Health Information Management
Keywords
- Bootstrap
- Deep learning
- ECG
- Echo
- Multimodal
- Waveform
Fingerprint
Dive into the research topics of 'Multimodal deep learning for structural heart disease prediction from ECG and clinical data'. Together they form a unique fingerprint.Cite this
- APA
- Standard
- Harvard
- Vancouver
- Author
- BIBTEX
- RIS