Sign In / Sign Out
Navigation for Entire University
- ASU Home
- My ASU
- Colleges and Schools
- Map and Locations
The existing healthcare model of the medical system is based on episodic examination or short-term monitoring for disease diagnosis and treatment. The major issues in such a system are the overlook of individual variability and the lack of personal baseline data, due to limited frequency of measurements. Continuous or non-intermittent monitoring is the key to create big data of individual health record for studying the variability and obtaining the personal baseline. Recent advancements in wireless body area networks (WBAN) and bio-sensing techniques has enabled the emergence of miniaturized, non-invasive, cost-effective wireless sensor nodes (WSNs) that can be deployed on the human body for personal health and clinical monitoring . Through WBAN, the monitored data can be transmitted to a near-field mobile aggregator for on-site processing. Through Internet infrastructures, the data can be uploaded to remote servers for storage and data analysis. These technology advancements will eventually transform the existing model of health related services to continuous monitoring for disease prediction and prevention . Such a wireless health revolution will make healthcare systems more effective and economic, benefiting billions of individuals and the society they live in.
One of the key challenges faced by the long-term wireless health monitoring is the energy efficiency of sensing and information transfer. Due to the limited battery capacity of WSNs, continuous sensing inevitably increases the frequency of battery recharging or replacement, making it less convenient for practical usage. In the WSNs for bio-sensing applications, the energy cost of wireless transmission is about 2 orders of magnitude greater than other operations (e.g., analogto-digital conversion (ADC)). State-of-the-art radio transmitters exhibit energy efficiency in the nJ/bit range while every other component consumes at most tens of pJ/bit . Therefore, reducing the data size for information transfer is the key to improve energy efficiency.
The existing CS framework [1, 4] offers a universal and simple data encoding scheme that can compress a variety of physiological signals, providing a viable solution to realizing energy-efficient WSNs for long-term wireless health monitoring. However, the compression ratio (CR) demonstrated by existing frameworks is limited given a signal recovery quality required for diagnosis purposes. In recent studies [5, 6] percent root-mean-square difference (PRD) of 8.5% and 9% is reported at a CR of 5x and 2.5x for ECG signals, respectively. These frameworks all deal with the sparsity of physiological signals on pre-determined bases and fail to take into account the individual variability in signals that is critical to exact signal recovery.
In this project, we propose an energy-efficient data acquisition framework, customized for the long-term electrocardiogram (ECG) monitoring, which exploits online dictionary learning (ODL) on server nodes to train personalized bases that capture the individual variability for further improving the sparsity of ECG signals. By incorporating such prior knowledge into signal recovery, the CS performance in terms of accuracy-CR trade-off is significantly enhanced, leading to further data size reduction and energy saving on sensor nodes. Additionally, the proposed framework does not require any pre-processing stages on sensor nodes. Alternatively, high reconstruction quality is enforced by pre-processing training data prior to the dictionary learning stage, to eliminate the impact of noise and interference on trained bases, enabling simpler and more cost-effective sensor structures. Experimental results based on MIT-BIH database show that our framework is able to achieve an average PRD of 9% at a CR of 10x. This indicates that our framework can achieve 2-4x additional energy saving on sensor nodes (for the same reconstruction quality) compared to the reference designs [1, 5, 6, 7]. Due to the training and personalization of the dictionary, the proposed framework has the potential to be generally applied to a wide range of physiological signals.
Copyright © 2015-2023 Parallel Systems and Computing Laboratory. All right reserved. | Site Admin: Fengbo Ren | Powered by Drupal.