Clinical correlative evaluation of an iterative method for reconstruction of brain SPECT images

Brain SPECT and PET investigations have showed discrepancies in Alzheimer's disease (AD) when considering data deriving from deeply located structures, such as the mesial temporal lobe. These discrepancies could be due to a variety of factors, including substantial differences in gamma-cameras and underlying technology. Mesial temporal structures are deeply located within the brain and the commonly used Filtered Back-Projection (FBP) technique does not fully take into account either the physical parameters of gamma-cameras or geometry of collimators. In order to overcome these limitations, alternative reconstruction methods have been proposed, such as the iterative method of the Conjugate Gradients with modified matrix (CG). However, the clinical applications of these methods have so far been only anecdotal. The present study was planned to compare perfusional SPECT data as derived from the conventional FBP method and from the iterative CG method, which takes into account the geometrical and physical characteristics of the gamma-camera, by a correlative approach with neuropsychology.

Correlations were compared between perfusion of the hippocampal region, as achieved by both the FBP and the CG reconstruction methods, and a short-memory test (Selective Reminding Test, SRT), specifically addressing one of its function. A brain-dedicated camera (CERASPECT) was used for SPECT studies with 99mTc-hexamethylpropylene-amine-oxime in 23 consecutive patients (mean age: 74.2 +/- 6.5) with mild (Mini-Mental Status Examination score > or =15, mean 20.3 +/- 3), probable AD. Counts from a hippocampal region in each hemisphere were referred to the average thalamic counts.

Hippocampal perfusion significantly correlated with the MMSE score with similar statistical significance (p < 0.01) between the two reconstruction methods. Correlation between hippocampal perfusion and the SRT score was better with the CG method (r = 0.50 for both hemispheres, p < 0.01) than with the FBP method (r = 0.37 and 0.43, respectively for the right and left hemisphere, p < 0.05 and p < 0.02). The bootstrap procedure showed that such correlation indexes were statistically different both in the right (p < 0.01) and in the left (p < 0.05) hemisphere.

These results are interpreted as a better performance of the CG reconstruction method in correctly detecting counts from hippocampal ROI. By using the same gamma-camera or collimator, alternative methods for brain SPECT reconstruction may improve quality of data and then help SPECT diagnostic accuracy.