Influence of smoothing process on statistical brain function image analysis

Abstract

Spatial smoothing performed after spatial normalization on the easy Z-score Imaging System (eZIS) is considered to affect signal size. The purpose of this study was to analyze quantitatively the influence of the smoothing process on eZIS using the voxel of interest (VOI) method. A normal database (NDB) was established using (99m)Tc-Hexamethylpropyleneamine oxime (HMPAO) brain perfusion SPECT images of thirty healthy volunteers. Then the NDB was smoothed with various Gaussian kernels (2, 4, 8, 12, 16 mm). Artificial lesions with known volumes and reduction of tracer uptake were made on one of the healthy volunteer images, and eZIS analysis was performed on the NDB of the same Gaussian kernel, respectively. The signal size of small-sized lesions was expanded 5.1 times to the true signal size of a 12 mm Gaussian kernel. On the other hand, the medium lesion size, which was approximately the same size as the posterior cingulate gyrus, was expanded 2.9 times to true signal size. Estimation of the false positive area using the extraction estimation method at lesion size medium indicated the lowest value at 8, 12 mm Gaussian kernel smoothing. The smoothing procedure at 8-12 mm Gaussian kernel is effective to detect a focal abnormality in the brain SPECT of Alzheimer's disease.