Advertisement

EFFECTS OF USING A NOVEL LONGITUDINAL PROCESSING PIPELINE FOR MEASURING CHANGE OVER TIME IN PIB-PET

      Background

      Accurate and reliable automated measures of change-over-time in amyloid PET are crucial to observational and clinical trials for Alzheimer’s disease. Previous works have examined how varying methodological choices, such as reference region, affect change measurements computed from cross-sectional (computed intra-timepoint) measurements. This work examines the effects of using longitudinal (simultaneous cross-timepoint) measurement methods.

      Methods

      We implemented a novel, automated longitudinal pipeline. Corresponding serial T1-weighted images were coregistered using a group-wise ANTs nonlinear registration to produce a mean-space T1-weighted single-subject template image (T1-SST). This T1-SST was segmented with SPM12. ANTs nonlinear registration to an in-house standard atlas was used to localize regions on T1-SST (a standard cortical target and seven candidate reference regions) for SUVR calculations. Serial PiB scans were coregistered using group-wise rigid registration with SPM12 to produce a mean-space PiB single-subject template image (PET-SST). A rigid registration between the PET-SST and the T1-SST was used to B-spline resample each PET scan to the T1-SST space for each timepoint’s SUVR calculation. We hypothesized that our approach (using a single mean-space T1-weighted image for segmentation and region localization, and a groupwise alignment of PET images with a single rigid registration to the mean-space T1) would yield improved longitudinal measurement performance compared to traditional intra-timepoint measurement methods. Our metrics of evaluation were: (1) reliability (larger R2of linear intra-subject fits) and (2) plausibility (smaller percent of subjects with significantly negative slopes, indicating biologically-implausible decreasing amyloid) of intra-subject serial trajectories. We tested this hypothesis using scans of 128 Mayo Clinic study participants with 3 serial timepoints with MRI and PiB scans with baseline PiB SUVR ≤ 2.5 (and consequently expected non-decreasing amyloid trajectories).

      Results

      Compared to the traditional cross-sectional method, the proposed longitudinal pipeline showed increased reliability of PiB-PET SUVR measures when using 6/7 tested reference regions. The percentage of subjects with biologically-implausible decreasing slopes using the longitudinal method was also ≤ that of the traditional method when using 6/7 tested reference regions.

      Conclusions

      Our proposed automated longitudinal pipeline measuring PiB-PET SUVR by simultaneously processing serial scans produces measurements with improved reliability and plausibility compared to traditional intra-timepoint methods.
      Figure thumbnail fx1
      Figure 1The reliability of intra-subject trajectories using the proposed longitudinal method was > that of the traditional cross-sectional method in 6/7 comparisons with varying reference regions.
      Figure thumbnail fx2
      Figure 2The percentage of subjects with biologically-implausible, significantly decreasing slopes using the proposed longitudinal method was ≤ that of the traditional method when using 6/7 tested reference regions.
      Table 1Reliability and plausibility performance for each combination of method and reference region
      MethodReference RegionMean R2 (reliability)% Significantly Negative (implausible) slopes
      Cross-SectionalSUVR_cereGM0.6840.781
      LongitudinalSUVR_cereGM0.7040.000
      Cross-SectionalSUVR_cereWhole0.6901.563
      LongitudinalSUVR_cereWhole0.7033.125
      Cross-SectionalSUVR_pons0.7444.688
      LongitudinalSUVR_pons0.7373.125
      Cross-SectionalSUVR_SupraWMero30.6671.563
      LongitudinalSUVR_SupraWMero30.7111.563
      Cross-SectionalSUVR_SupraWMero50.6691.563
      LongitudinalSUVR_SupraWMero50.6921.563
      Cross-SectionalSUVR_atlasSupraWM0.6612.344
      LongitudinalSUVR_atlasSupraWM0.7000.781
      Cross-SectionalSUVR_compositeRef0.7234.688
      LongitudinalSUVR_compositeRef0.7532.344