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

Q: What is the clinical evidence from this study?

Population: Structural heart disease (n=100000). Intervention: Temporal Convolutional Network (TCN) vs. Simple CNN, ResNet 1D-18, Light Transformer, and Hybrid models. Primary outcome: Area Under the Receiver Operating Characteristic Curve (AUC) for structural heart disease prediction.

Q: What does this research mean for the field?

Temporal Convolutional Networks consistently outperform competing deep learning architectures in predicting structural heart disease from multimodal ECG and clinical data, achieving superior predictive accuracy (AUC 0.8626) alongside computational efficiency. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Nurudeen Ayobami Ajadi; Sulaiman O Afolabi; Ibrahim Oyeyemi Adenekan; Jimoh Afeez Oyeshola; Ajayi Abayomi Olumuyiwa; Adeniran Teslim Ajibola; Adepoju Gboyega David; Hassan Nana Firdaous; Ajadi Salimot Aanu; Ogunsola Isqeel Adesegun

doi:10.1016/j.smhl.2026.100703

August 26, 2026Smart HealthOpen Access

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

View Full Paper

Ask AI

Bookmark

Share

Key result

TCN model predicts structural heart disease with an AUC of ~0.86, outperforming competing architectures.

n=100,000

Why the study?

Deep learning models for structural heart disease prediction from ECG and clinical data require evaluation of multiple architectures for robustness, efficiency, and fairness.

Does a Temporal Convolution Network (TCN) multimodal deep learning model improve structural heart disease prediction from ECG and clinical data compared to other deep learning architectures?

Key Points

To evaluate and compare five deep learning architectures for predicting structural heart disease from multimodal electrocardiogram and clinical datasets while assessing model fairness and stability.
Evaluated five deep learning architectures: Temporal Convolutional Network (TCN), Simple CNN, ResNet1d18, Light transformer, and a Hybrid model.
Assessed models across three random seeds and computed 95% bootstrap confidence intervals to measure stability and performance differences.
Implemented a masked-modelling approach to address missing data within the EchoNext datasets alongside algorithmic fairness analyses.
Temporal Convolutional Networks consistently outperformed other competing architectures, achieving statistically significant improvements with stable performance across runs.
TCN demonstrated efficient computation and stable training in predictive analysis relative to Simple CNN, ResNet1d18, Light transformer, and Hybrid architectures.
Fairness evaluations demonstrated that TCN maintained balanced, stable performance across demographic variables for healthcare deployment.

Structured PICO

Does a Temporal Convolution Network (TCN) multimodal deep learning model improve structural heart disease prediction from ECG and clinical data compared to other deep learning architectures?

P

Population

100,000 12-lead electrocardiograms from adult patients (mean age 61.2 years) retrospectively collected over a one-year interval between 2008 and 2022.

E

Exposure

Temporal Convolution Network (TCN) multimodal deep learning model using 12-lead ECG waveform and tabular clinical data

C

Comparator

Other deep learning architectures (Simple CNN-1D, ResNet 1D-18, Light transformer, and Hybrid model)

O

Outcome

Structural heart disease prediction (classification performance)

Limitations

Limited model transparency (black box nature of deep learning)
Difficulties integrating deep learning systems with electronic health records (EHRs)
Presence of missing values in tabular clinical data necessitating mask-aware modeling
Class imbalance requiring positive class weighting

Implication

A multimodal Temporal Convolution Network (TCN) effectively predicts echocardiogram-confirmed structural heart disease using 12-lead ECGs and clinical tabular data, outperforming other standard deep learning architectures.

Abstract

This research presents multimodal deep learning for structural heart disease prediction. We evaluated multiple deep learning architectures, including TCN, Simple CNN, ResNet1d18, Light transformer and Hybrid model. The models were examined across the three seeds to ensure robustness, and bootstrap confidence interval is used to measure performance differences. TCN consistently outperforms other competing architectures, achieving statistically significant improvements with stable performance across runs. Similarly in predictive analysis, TCN has efficient computation and stable training compared to all competing architectures. Our results show that TCN emphasizes fairness evaluation when developing deep learning models for healthcare applications.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Ajadi et al. (2026) studied Structural heart disease (n=100,000). Temporal Convolutional Network (TCN) vs. Simple CNN, ResNet 1D-18, Light Transformer, and Hybrid models was evaluated on Area Under the Receiver Operating Characteristic Curve (AUC) for structural heart disease prediction. The Temporal Convolutional Network (TCN) consistently outperformed competing architectures in predicting structural heart disease, achieving an AUC of 0.8626 and demonstrating stable performance and efficient computation.

synapsesocial.com/papers/6ad1159b00741cc3fba7d67f https://doi.org/10.1016/j.smhl.2026.100703

Topics

Artificial intelligence in cardiology

View Full Paper

Ask AI

Bookmark

Share

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

Key Points

Structured PICO

Limitations

Implication

Abstract

Discussion

Cite This Study

Topics

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

Key Points

Structured PICO

Limitations

Implication

Abstract

Discussion

Cite This Study

Keep reading

More related papers

More on Artificial intelligence in cardiology

More from Smart Health

Topics

Keep reading

More related papers

More on Artificial intelligence in cardiology

More from Smart Health