Bilateral spatial filtering: refining methods for localizing brain activation in the presence of parenchymal abnormalities.

Functional MRI (fMRI) is an important tool for pre-surgical localization of eloquent cortex prior to resection of brain lesions. To increase the inherently low activation signal to noise ratio, fMRI pre-processing steps often include spatial smoothing. However, the effects of smoothing in the presence of brain lesions have not been studied. We have adapted the widely used method of Gaussian spatial filtering to include an "edge stopping" function. This method, termed bilateral filtering, minimizes blurring of apparent brain activity across anatomic boundaries and into regions of non-activation. fMRI data were acquired in a patient with a known low grade glioma during a blocked finger-tapping paradigm. Simulated activity was superimposed on baseline images of non-activated brain using the same paradigm, with additive signal equal to 1, 3, and 5% of the mean physiologic background. Comparison of Gaussian and bilateral filtering suggests that the modified technique more accurately locates brain activation and increases the significance of activation bordering sharp transitions. Thus, spatial pre-processing with a bilateral filter may be particularly useful in the pre-operative assessment of brain lesions.