A hybrid intelligent recognition system for the early detection of strokes
The increasing prevalence of wearable sensors and low-cost mobile devices have prompted the development of systems for automated diagnosis. Here we focus on models and algorithms for the early detection of strokes that are implanted in a wearable device that generates warning alarms and automatically connects to e-health services, ensuring timely interventions at the onset of a stroke.
The proposed approach employs two wearable devices to monitor movement data that involve two main stages: Human Activity Recognition (HAR) and alarm generation.
Two different HAR methods capable of classifying current human activity are developed and compared: one uses genetic fuzzy finite-state machines, and the other relies on Time Series (TS) analysis. Furthermore, an algorithm using Symbolic Aggregate approXimation (SAX) TS representation is proposed for alarm generation purposes, which is triggered by the detection of anomalous movements.
The proposed methods are evaluated using realistic data gathered from healthy individuals. A discussion of topics related to the learning issues involved in these techniques is included. It is worth mentioning that the proposed algorithms can be easily transferred to embedded systems and can benefit from reduced learning costs.