Predicting Cardiac Output in Young Adults Using Wearable Sensors and Machine Learning

Article Impact Level: HIGH
Data Quality: STRONG
Summary of  Bioengineering  https://doi.org/10.3390/bioengineering13040477 
Dr. Chih-Hao Chang  et al.

Points

Cardiovascular disease is increasingly affecting young adults aged 20 to 29 due to rising global rates of obesity, hypertension, hyperlipidemia, and diabetes.
Advanced hemodynamic evaluations typically require specialized hospital equipment, creating significant diagnostic barriers for at-risk populations in resource-limited healthcare settings.
An international research team developed a data-augmented artificial neural network model to non-invasively predict cardiac index using basic physiological inputs.
The artificial intelligence system achieved a classification accuracy of 97.78% when utilizing three physiological parameters derived from non-invasive skin sensors.
Robust predictive performance was maintained even under two-parameter input conditions, offering a practical alternative to complex and invasive cardiac monitoring procedures.

Summary

This study evaluated the performance of a data-augmented artificial neural network (ANN) integrated with non-invasive Internet of Things (IoT) sensing devices for predicting cardiac index (CI). With cardiovascular disease rising among young adults aged 20–29 due to increasing rates of obesity and metabolic conditions, non-invasive hemodynamic monitoring offers an alternative to resource-intensive hospital equipment. Led by Patricia Angela R. Abu, researchers combined data from non-invasive sensors—including body composition analyzers, blood pressure monitors, and blood flow analyzers—to develop a robust AI prediction framework.

To assess predictive accuracy, the model was trained using key physiological parameters derived from wearable skin sensors and non-invasive hemodynamic instruments, including the TERUMO ES-P2000 blood pressure monitor and PhysioFlow PF07 Enduro cardiac analyzer. Experimental results demonstrated that when configured with three input physiological parameters, the ANN achieved a classification accuracy of 97.78%. Furthermore, the neural network maintained strong predictive reliability even when evaluated under reduced, two-parameter input conditions, significantly outperforming traditional clinical estimation approaches.

The findings establish that integrating feature preprocessing with data-augmented neural networks enables accurate, real-time prediction of cardiac index without requiring invasive arterial catheterization or complex clinical setups. By reducing the number of required physiological inputs while maintaining high classification accuracy, this non-invasive AI framework provides a scalable diagnostic strategy for evaluating cardiac performance and guiding cardiovascular care in resource-limited primary care settings.

Link to the article: https://www.mdpi.com/2306-5354/13/4/477

References

Chang, C.-H., Chan, M.-L., Fang, Y.-H., Huang, P.-L., Chen, T.-Y., Chi, T.-K., Cha, I. E., Ger, T.-R., Li, K.-C., Chen, S.-L., Wang, L.-H., Wang, J.-C., & Abu, P. A. R. (2026). Robust non-invasive cardiac index prediction via feature integration and data-augmented neural networks. Bioengineering, 13(4), 477. https://doi.org/10.3390/bioengineering13040477