Key result
DT4ECG achieves ~99% accuracy in identity recognition and ~90% in activity classification from ECG data.
Why the study?
ECG signals contain information about both user identity and physical activity, but individual physiological differences complicate activity inference.
Does the DT4ECG multi-task deep learning framework improve joint human identity recognition and human activity classification accuracy from ECG signals compared to standard deep learning models in healthy volunteers?
Does the DT4ECG multi-task deep learning framework improve joint human identity recognition and human activity classification accuracy from ECG signals compared to standard deep learning models in healthy volunteers?
Combining ECG-based biometric identity recognition and human activity classification into a single multi-task deep learning framework enables simultaneous user authentication and physiological activity monitoring with high accuracy.
May enable simultaneous ECG-based authentication and activity monitoring; leaves open prospective validation before clinical adoption.
Electrocardiogram (ECG) signals contain information about both user identity and physical activity, but individual physiological differences complicate activity inference. We propose DT4ECG, a subject-aware multi-task framework for joint identity recognition (IDR) and human activity recognition (HAR) among enrolled users. A Multi-Scale Residual Convolution (MSRC) backbone extracts complementary waveform and temporal features, while a Subject Adaptive Fusion Module (SAFM) uses learned identity-related representations to calibrate activity features. Evaluation on two public datasets and one self-built dataset achieves IDR macro F1 of 98.59--99.96\% and HAR macro F1 of 95.40--96.82\%. Ablation results show consistent HAR improvements from SAFM across the three datasets, and recorded-noise tests characterize sensitivity to common ECG artifacts. The two-channel model contains 0.361 million parameters. These findings demonstrate the value of treating subject-related physiology as context for activity recognition. By linking activity estimates to user attribution within a single ECG framework, this work provides a basis for integrated physiological monitoring among enrolled users in shared-device settings.
No takes yet. Share an insight, caveat, or question.
You et al. (2026) studied Healthy volunteers undergoing ECG-based identity and physical activity monitoring (n=15). DT4ECG dual-task deep learning framework (1D-ResNet with Sequence Channel Attention and GradNorm) vs. Baseline deep learning models (LSTM, MLP, 1D-CNN, 1D-MobileNet, 1D-EfficientNet, 1D-ResNet, 1D-DenseNet) was evaluated on Classification accuracy for human identity recognition (ID) and human activity recognition (HAR). The DT4ECG framework achieved 99.12% accuracy in identity recognition and 90.11% accuracy in activity classification on the self-built ECG dataset.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: