Spatial data analysis in the quantitative assessment of cerebral white matter pathology on MRI in HIV infection

Abstract

This study was carried out using MRI (proton density--and T2-weighted) on 16 HIV-negative controls, 9 symptom-free HIV-positive patients and 25 with CDC IV HIV disease. The studies from this last group had previously been allocated by a radiologist to the following categories: 8 with focal mass lesions and normal-appearing white matter; 9 with diffuse encephalopathy (high signal on T2-weighted images, affecting most or all of the white matter) and 8 with patchy encephalopathy (high signal affecting only one or two areas within the white matter). Moran's I, a statistic of spatial autocorrelation, was calculated for the grey-scale values of a sampled pixel array from a central white matter region of each of the images. All values of Moran's I calculated in this study showed a large positive excess over the expected value under randomisation, indicating highly significant positive auto-correlation in the spatial arrangement of the grey-scale values. On T2-weighted images a statistically significant increase in the mean value of Moran's I, compared with controls, was found in the diffuse encephalopathy group, indicating that quantifiable changes in the spatial autocorrelation of pixel data can be related to recognised qualitative changes in the appearance of white matter in subjects with HIV disease. A lesser, but significant, rise in the mean value of Moran's I was also found in the focal mass lesion group, suggesting that changes in spatial autocorrelation may indicate pathological change in advance of qualitative MRI changes.