Universal Multi-complexity Measures for Physiological State Quantification in Intelligent Diagnostics and Monitoring Systems

Previously we demonstrated that performance of heart rate variability indicators computed from necessarily short time series could be significantly improved by combination of complexity measures using boosting algorithms. Here we argue that these meta-indicators could be further incorporated into various intelligent systems. They can be combined with other statistical techniques without additional recalibration. For example, usage of distribution moments of these measures computed on consecutive short segments of the longer time series could increase diagnostics accuracy and detection rate of emerging abnormalities. Multiple physiological regimes are implicitly encoded in such ensemble of base indicators. Using an ensemble as a state vector and defining distance metrics between these vectors, the encoded fine-grain knowledge can be utilized using instance-based learning, clustering algorithms, and graph-based techniques. We conclude that the length change of minimum spanning tree based on these metrics provides an early indication of developing abnormalities.