Statistical parametric mapping of ^99mTc-HMPAO-SPECT images for the diagnosis of Alzheimer's disease: Normalizing to cerebellar tracer uptake

Background: For a quantitative comparison of images obtained during ^99mTc-hexamethylpropylene amine oxime (HMPAO) single-photon emission computed tomography (SPECT), brain activity values are usually normalized to a reference region. In studies of Alzheimer-type dementia (ATD), the cerebellum is often used as a reference region, assuming that it is spared any major pathological involvement. Statistical parametric mapping (SPM) may enhance the evaluation of SPECT scans in ATD patients. However, current SPM software only allows scaling to average whole brain activity (i.e., global normalization). The aim of this study was to develop an easily applied, objective, and reproducible method for determining average cerebellar tracer uptake so that images can be scaled specifically to cerebellar activity prior to the performance of SPM analysis. We also investigated whether cerebellar normalization increases the sensitivity and specificity of SPM analysis of ATD patients compared with global normalization. Methods: Image files were taken from a parallel study investigating the use of SPECT as a diagnostic tool for early onset of ATD. Two methods for determining cerebellar activity were developed: one manually, using templates, the other automated, using specified coordinates entered into a Matlab routine. Group comparison of ATD patients versus controls (= healthy volunteers and depressed patients) was performed on a voxel-by-voxel basis using SPM 96 on Windows 95. Receiver operator characteristics (ROC) were computed for 20 student raters examining patient and control scans with and without single-subject SPMs. Results: The reduction of cerebral blood flow in the group of ATD patients appeared 1.7 times greater in spatial extent when the tracer uptake was normalized to cerebellum rather than to average whole brain activity. Computing the reverse contrast (reductions in the control group compared with ATD patients) produced clusters of significance in globally normalized images which were not manifest after normalizing to cerebellum. This is consistent with the notion that the cerebellum is spared in ATD. Analysis of the area under the ROC curve showed that cerebellar-normalized SPM produced significantly improved accuracy over perfusion scans alone. Conclusion: An easily applied, objective, reproducible method was developed for normalizing images to cerebellum prior to the performance of SPM analysis. Cerebellar normalization produced more extensive abnormalities in SPM analyses of ATD patients than global normalization. Furthermore, cerebellar normalization produced marginally more accurate diagnostic results in single-scan SPM analysis of ATD patients than did global normalization.