Particle Learning Approach to Bayesian Model Selection: An Application from Neurology

Abstract

An improved method is sought to accurately quantify the number of motor units that operate a working muscle. Measurements of a muscle’s contractive potential are obtained by administering a sequence of electrical stimuli. However, the firing patterns of the motor units are non-deterministic and therefore estimating their number is non-trivial. We consider a state-space model that assumes a fixed number of motor units to describe the hidden processes within the body. Particle learning methodology is applied to estimate the marginal likelihood for a range of models that assumes a different number of motor units. Simulation studies of these systems, containing up to 5 motor units, are very promising.