ATHENA I—An architecture for near real-time physiological signal monitoring and pattern detection

Abstract

Smart healthcare has wide application possibilities, including physiological signals analysis. With the development and improvement of microelectronic devices and sensors, wearable sensors can acquire physiological signals with quality and support remote health monitoring. Due to the growing use of these devices, there is a need for easy and flexible ways to use them in health monitoring. To contribute in this context, we developed the ATHENA I, an architecture focused on using wearables to acquire physiological signals and applying artificial intelligence to classify health patterns in real-time. Because of the constant improvement of wearable devices, we focused on developing a flexible and efficient architecture. Since many health centers outside the central regions of the metropolis have reduced connectivity, we designed an architecture capable of operating entirely outside the data network. The architecture prototype features BITalino wearable sensor and a Raspberry Pi single-board computer. For a complete prototype evaluation, we performed stress-inducing experiments conducted by psychologists to acquire a new dataset of physiological signs. We used this dataset in pattern classification in offline operations and near real-time situations. Our purpose with these experiments is to predict stress, being able to differentiate the normal state from the stress state. The classification results for the stress patterns showed promising results, with accuracy above 98.72% for binary classification (stress, no-stress); and 92.72% for classification with three classes (pre-stress, stress, and post-stress). In the real-time classification, we obtained an accuracy of 69.00%. The architecture presented excellent communication and operation stability. During the experiments, the architecture performed short and long acquisitions efficiently.