Modified Information Theoretic Criteria for Low Complexity Estimation of the Amount of Components in MEG Measurements
Elnaz Javidi, João Paulo C. L. da Costa, Ricardo Kehrle Miranda, João Paulo A. Maranhão, José Alfredo Ruiz Vargas · 2019
Visual Evoked Potential (VEP) allows the diagnostics of illnesses in the optic nerve and the examination of seizure disorders, such as epilepsy. In order to analyze the VEP, variations of the neural electric tensions on the area of visual cortex in the occiput are measured by Electroencephalography (EEG) and Magnetoencephalography (MEG). Although traditionally no algorithm is applied for the estimation of the amount of components and this estimation is subjectively and visually performed by an expert, there are schemes for this task in the literature, such as ICASSO. Drawbacks of ICASSO are its huge computational complexity and its dependence on human intervention to define thresholds in order to correctly find the amount of resolvable components. In addition, traditional eigenvalue based information theoretic criteria (ITC), such as Akaike Information Criterion (AIC) and Minimum Description Length (MDL), completely fail to estimate the amount of components when directly applied to MEG measurements. In this work, we propose a modification of the traditional eigenvalue based ITC, such that their estimates of the amount of components are similar to the ICASSO estimates. Moreover, the modified ITC presents a dramatical reduction in the computational cost in comparison with ICASSO. We validate our results using measurements from the Jena University Hospital.