Electroneurography (ENG) is a minimally invasive method of identifying activation of separate neuron fascicles within a peripheral nerve from the measurements voltage distributions on the surface of the nerve. ENG can be applied to control a robotic prosthetic arm by the intact peripheral nerves of an amputation patient, or to gain muscle control of a spinal chord injury patient. Mathematically, the ENG problem is an ill-posed inverse problem with high degree of non-uniqueness and instability, and the signal suffers from a low signal-to-noise ratio. In this talk, we discuss the computational model, the beamformer algorithm that has been suggested for the ENG problem, and a Bayesian hierarchical version of it, allowing a noise reduction method to improve the identification of the active fascicles.
Bayesian source separation in MEG
Magnetoencephalography (MEG) is a completely non-invasive brain-mapping modality which uses measurements of the magnetic field outside the head induced by electrical brain activity to localize and characterize the activity inside the brain. Potentially, it is particularly useful in the study of epilepsy as a tool for localizing the foci of the onset of seizures. A key issue in MEG is the separation of sources of a different nature. Non-focal sources from both inside and outside of the brain produce interference, making the inverse problem of identifying the focal source signal extremely difficult. In this talk we show how Bayesian methods can be used to address this issue. In particular, we illustrate how a mixed prior distribution is able to separate sources which are statistically different from each other. Furthermore, we propose using a depth scan to identify activity from deep focal sources. Numerical simulations are used to generate controlled data in order to validate the model.