RESUMO
The analysis of diffuse optical imaging (DOI) data has seen significant developments over the last few years. When compared to fMRI, signals originating from optical imaging are tainted by more physiology and the separation of activation from this background can be difficult in some cases. In this work, we show that the use of time-frequency techniques based on wavelets distinguish different physiological sources from the evoked response to a given stimulus. In particular, we show that analytical complex wavelets identify synchronies in the signal at different scales. These synchronies are then used to extract activation information from the DOI data in order to estimate the evoked hemodynamic response or to define a new type of contrast between two conditions. This work presents both simulations and applications with real data (visual stimulation and motor tasks experiments).
Assuntos
Algoritmos , Mapeamento Encefálico/métodos , Encéfalo/fisiologia , Imagem de Difusão por Ressonância Magnética/métodos , Potenciais Evocados/fisiologia , Espectroscopia de Ressonância Magnética/métodos , Processamento de Sinais Assistido por Computador , HumanosRESUMO
Diffuse optical imaging uses light to provide a surrogate measure of neuronal activation through the hemodynamic responses. The relative low absorption of near-infrared light enables measurements of hemoglobin changes at depths reaching the first centimeter of the cortex. The rapid rate of acquisition and the access to both oxy and deoxy-hemoglobin leads to new challenges when trying to uncouple physiology from the signal of interest. In particular, recent work provided evidence of the presence of a 1/f noise structure in optical signals and showed that a general linear model based on wavelets can be used to decorrelate the structured noise and provide a superior estimator of response amplitude when compared with conventional techniques. In this work the wavelet techniques are extended to recover the full temporal shape of the hemodynamic responses. A comparison with other models is provided as well as a case study on finger-tapping data.