Emergent behavior search in complex biosignals

Non-invasive sensor technology can detect electromyographic (EMG) signals at the skin surface close to a muscle of interest. These signals represent the combined electrical effect of cellular-level motor units that drive the overall contraction behavior of muscles. There is considerable clinical and scientific interest in being able to decompose EMG signals into constituents that represent incontrovertible evidence for the firing behavior of as many simultaneously active motor units as possible. The last decade has seen Artificial Intelligence (AI) methodologies play a significant role in the development of decomposition algorithms. For example, the IPUS algorithm from AI for Integrated Processing and Understanding of Signals is incorporated into a recent version (PD-IPUS) of the Precision Decomposition system previously developed at Boston University. Generally acknowledged as the most successful algorithm in the field so far, the automatic mode of PD-IPUS is typically able to characterize the firing behaviors of 5 to 15 motor units at a variety of muscle contraction levels with accuracy ranging from 75% to 90%. In this dissertation, we present the detailed structure and an accompanying performance analysis of a new Precision Decomposition algorithm we have developed for accurately characterizing the firing behavior of simultaneously active motor units. The algorithm uses sophisticated Artificial Intelligence techniques to control a complex process of emergent behavior search (EBS) in a space of possible PD-IPUS answers. An initial prototype of this PD-EBS algorithm has been implemented in C++ and successfully tested on a database of real and reconstructed EMG signals for a variety of muscles and muscle contraction levels. With respect to the database of real EMG signals, it is found to decompose an average of 22 motor units per EMG signal with an average accuracy ranging from 90% to 95% per EMG signal. Validation of accuracy has been performed using a two-source test as well as a novel "comparable complexity" test. As the PD-EBS algorithm and its accompanying non-invasive sensor technology move from the laboratory to real-world settings, they have the realistic potential to evolve into practical diagnostic tools for clinicians and a key resource in fields such as motor control, movement science, and sports science.