Skip to main navigation Skip to search Skip to main content

Multimodal deep learning for structural heart disease prediction from ECG and clinical data

  • Nurudeen Ayobami Ajadi
  • , Sulaiman Omotayo Afolabi
  • , Ibrahim Oyeyemi Adenekan
  • , Afeez Oyeshola Jimoh
  • , Abayomi Olumuyiwa Ajayi
  • , Teslim Ajibola Adeniran
  • , Gboyega David Adepoju
  • , Nana Firdausi Hassan
  • , Salimot Aanu Ajadi
  • , Isqeel Adesegun Ogunsola

Research output: Contribution to journal › Article › peer-review

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 languageEnglish
Article number100703
Number of pages18
JournalSmart Health
Volume42
DOIs
StatePublished - 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