Related Experiment Video
Updated: Jun 16, 2026
14:09
Tilt Testing with Combined Lower Body Negative Pressure: a "Gold Standard" for Measuring Orthostatic Tolerance
Published on: March 21, 2013
Deep Learning-Based Early Prediction of Syncope Onset During Tilt Table Testing via Temporal Convolutional
Alex Wee Wong1, Wee Jian Chin1, Maw Pin Tan2
1Department of Mechatronics and Biomedical Engineering, Lee Kong Chian Faculty of Engineering and Science, Universiti Tunku Abdul Rahman, Kajang, 43000, Petaling Jaya, Selangor, Malaysia.
Annals of Biomedical Engineering
|May 19, 2026
Summary
This study developed an autoencoder-based anomaly detection method for early syncope prediction during head-up tilt table tests (HUTT). The model achieved high accuracy, enabling preemptive test termination and reducing patient discomfort.
Area of Science:
- Cardiology
- Biomedical Engineering
- Artificial Intelligence
Background:
- Head-up tilt table tests (HUTT) are lengthy and uncomfortable, often inducing syncope and associated symptoms like nausea and pallor.
- Current HUTT procedures lack methods for early syncope onset prediction, leading to prolonged patient discomfort.
Purpose of the Study:
- To develop and evaluate an autoencoding-based anomaly detection system for early syncope prediction during HUTT.
- To enable preemptive termination of HUTT, thereby minimizing patient discomfort and adverse events.
Main Methods:
- Utilized heart rate (HR), systolic blood pressure (SBP), diastolic blood pressure (DBP), and high frequency normalized RRI (Hfnu_RRI) signals.
- Processed signals into feature images for autoencoder (AE) input, coupled with an anomaly severity algorithm.
- Evaluated six AE architectures, fine-tuned the best model, and validated the syncope detector over 100 iterations.
Main Results:
- The temporal convolutional autoencoder anomaly detector (TCAAD) achieved high performance metrics: accuracy (0.9424), recall (0.9838), precision (0.9141), F1 score (0.9461).
- The TCAAD demonstrated a significant early prediction time of 523.69 seconds.
- Model performance was comparable to existing real-time prediction methods.
Conclusions:
- Anomaly detection methods combined with signal correlation monitoring are effective for syncope prediction.
- The developed TCAAD offers one of the longest early prediction times, improving HUTT safety and patient experience.
- This approach facilitates early intervention, potentially avoiding syncope and its associated symptoms.