Signal analysis of human brain function activation: effect of spatial registration noise in brain images

Human brain function is frequently being studied by subtracting two brain images of the same subject, one obtained while resting and the other one while his brain is activated by either a motor task, a cognitive activity, the effects of a medication, or drug abuse. The activation signal has noise induced both by the camera and the patient, i.e. daily metabolic variability, patient positioning errors, etc. The present methodology allows the determination of faint activation signals by finding sharper error bounds in the error induced in the signal by each separate source of noise in the image, combining them later for a final bound on the noise to assess the signal´s statistical significance. Using simple concepts from numerical analysis and directional statistics, we present a general methodology to analyze the effect that spatial registration errors have on each pixel´s signal to noise ratio. This approach to signal analysis is valid for most imaging modalities and registration methods. Experimental results are given for 48 subjects, using Positron Emission Tomography (PET) brain images spatially registered with the Centroidal Principal Axes registration method