Advertisement

A ROBUST, SIMPLIFIED BRAAK-TYPE CLASSIFICATION SCHEME FOR FLORTAUCIPIR F-18 TAU PET IMAGES

      Background

      Using ROIs based on the Braak staging scheme, patterns of [18F]-flortaucipir binding consistent with expectations from tau neuropathology can be observed in vivo [Schwarz et al., Brain 2016]. However, small regions of interest (ROIs) are potentially sensitive to variations in image preprocessing, atrophy and experimental noise. Medial temporal lobe (MTL) regions distinguishing stages 1-3 are prone to contamination from adjacent extraparenchymal signals and binding to the choroid plexus. Our aim was to identify a simplified and practical flortaucipir PET staging scheme for AD, robust to the above nuisance factors.

      Methods

      Four ROIs were defined from standard Harvard-Oxford or Juelich (FSL) atlas structures: (1) an MTL ROI comprising the anterior hippocampus, the parahippocampal/entorhinal and fusiform gyri; (2) a lateral temporal lobe (LTL) ROI comprising the anterior inferior and middle temporal gyri; (3) anterior superior temporal gyrus; and (4) primary visual cortex. We compared larger and smaller variants on these ROIs, and evaluated the effect of grey-matter masking. Individual subject ROI profiles were binarized at 3SD above the mean signal in each ROI from 14 young controls and matched to predefined Braak-like patterns. Staging (0, 1-3, 4, 5 or 6) was performed on each hemisphere independently and the most advanced stage taken. Stages 1-3 were not differentiated. Performance was evaluated in terms of maximizing test-retest consistency (N=21) and minimizing the number of unassigned profiles (N=21+236). The algorithm was then prospectively applied to flortaucipir scans from ADNI-2 (N=101).

      Results

      The larger ROIs and simplified staging scheme yielded higher test-retest consistency and reduced number of unassigned profiles than the published approach. The variant ROIs and grey matter masking did not improve performance. Applied to the ADNI data set, 91% of subjects were classified into a typical predefined estimated Braak stage pattern (Figure). Higher stage tau patterns were associated with more advanced disease severity and amyloid positivity.

      Conclusions

      This simplified flortaucipir staging scheme provides a practical and robust means of classifying tau patterns in vivo in terms of involvement of the sentinel brain regions proposed by Braak. It is easily implemented using widely available atlases, and may complement other tau PET summary measures.
      Figure thumbnail fx1
      Figure thumbnail fx2