Movement correction method for human brain PET images: application to quantitative analysis of dynamic 18F-FDDNP scans

Abstract

Head movement during a PET scan (especially a dynamic scan) can affect both the qualitative and the quantitative aspects of an image, making it difficult to accurately interpret the results. The primary objective of this study was to develop a retrospective image-based movement correction (MC) method and evaluate its implementation on dynamic 2-(1-{6-[(2-(18)F-fluoroethyl)(methyl)amino]-2-naphthyl}ethylidene)malononitrile ((18)F-FDDNP) PET images of cognitively intact controls and patients with Alzheimer's disease (AD).

Methods: Dynamic (18)F-FDDNP PET images, used for in vivo imaging of beta-amyloid plaques and neurofibrillary tangles, were obtained from 12 AD patients and 9 age-matched controls. For each study, a transmission scan was first acquired for attenuation correction. An accurate retrospective MC method that corrected for transmission-emission and emission-emission misalignments was applied to all studies. No restriction was assumed for zero movement between the transmission scan and the first emission scan. Logan analysis, with the cerebellum as the reference region, was used to estimate various regional distribution volume ratio (DVR) values in the brain before and after MC. Discriminant analysis was used to build a predictive model for group membership, using data with and without MC.

Results: MC improved the image quality and quantitative values in (18)F-FDDNP PET images. In this subject population, no significant difference in DVR value was observed in the medial temporal (MTL) region of controls and patients with AD before MC. However, after MC, significant differences in DVR values in the frontal, parietal, posterior cingulate, MTL, lateral temporal (LTL), and global regions were seen between the 2 groups (P < 0.05). In controls and patients with AD, the variability of regional DVR values (as measured by the coefficient of variation) decreased on average by more than 18% after MC. Mean DVR separation between controls and patients with AD was higher in frontal, MTL, LTL, and global regions after MC. Group classification by discriminant analysis based on (18)F-FDDNP DVR values was markedly improved after MC.

Conclusion: The streamlined and easy-to-use MC method presented in this work significantly improves the image quality and the measured tracer kinetics of (18)F-FDDNP PET images. The proposed MC method has the potential to be applied to PET studies on patients having other disorders (e.g., Down syndrome and Parkinson's disease) and to brain PET scans with other molecular imaging probes.

FIGURE 1

General illustrated overview of the retrospective MC method proposed in this study.

FIGURE 2

Head movement can cause transmission-emission (TX-EM) misalignment during a dynamic brain PET. (A) An early non-AC EM frame is shown to be fused well with the TX image taken at the start of the study. Both images are shown at the same transaxial plane. (B) A late non-AC EM frame is now fused with the original TX image at the same transaxial plane from before. Since the subject’s head moves away from the PET gantry after the acquisition of the early non-AC EM frame, the mismatched TX-EM would cause an incorrect ACF matrix to be applied to the late non-AC EM frame. (C) After applying the first part of the MC method, the original TX image is matched to the late EM image. This produces an accurate ACF matrix that can be used to properly reconstruct the late EM frame.

FIGURE 3

For the control subject with negligible head movement, the MC method did not introduce any apparent image degradation to the [¹⁸F]-FDDNP DVR image. Before MC, the DVR image of the AD subject with considerable head movement was subject to image artifacts (i.e., abnormally high scalp uptake and asymmetrical [¹⁸F]-FDDNP binding near the top of the head as seen in the coronal view) due to transmission-emission misalignment. However, after MC, the image quality of the DVR image was improved. All images correspond approximately to the same cross-sectional locations in their respective coronal, transaxial and sagittal views.

FIGURE 4

Displacement maps for a control subject with minor head movement (A) and an AD subject with significant head movement (B). Each map shows the displacement undergone by each voxel in the reference non-AC emission frame as it is co-registered to the last non-AC emission frame in the dynamic image. Note the difference in scale for both color bars.

FIGURE 5

Mean displacement of the reference non-AC emission frame as it is co-registered to each of the remaining frames in the dynamic image for a control subject with negligible head movement and an AD subject with large head movement. The mean was calculated from the displacements of all the voxels in the head.