P3-090: JOINT DEBLURRING OF LONGITUDINAL DIFFERENTIAL PET IMAGES OF TAU
Background
Differential measurement of tau aggregates in the brain using serial PET imaging is of great significance in image-based biomarker development for aging and Alzheimer's disease. However, partial volume effects arising from the limited spatial resolution of PET pose a challenge to quantitation. We have developed an image deblurring technique that utilizes high-resolution anatomical information to correct for partial volume effects in differential images of tau based on serial [18F]Flortaucipir PET datasets.
Methods
The deblurring technique is based on 1) deconvolution of co-registered difference images using the point spread function of the scanner measured in the image space and 2) minimization of the joint entropy between the PET difference image and a co-registered T1-weighted MPRAGE MR image. Serial [18F]Flortaucipir ([18F]AV1451) PET data was acquired from an elderly cohort (63-90 years) consisting of 96 Harvard Aging Brain Study participants. The time gap between baseline and follow-up scans was 0.6-3.6 years.
Results
As shown in Fig. A, deblurring led to successfully recovery of structural details in the PET difference images while suppressing spurious negative values due to noise. Annualized rates of change based on standardized uptake value ratio (SUVR) differences across the two timepoints were computed corresponding to the following 4 clinically relevant regions-of-interest (ROIs): inferior temporal cortex (ITC), fusiform gyrus (FG), parahippocampal gyrus (PHG), and entorhinal cortex (EC), as shown in Fig. B. We used the coefficient of variation (CV) or relative standard deviation as a measure of biomarker reliability. As shown in Fig. C, deblurring led to prominent reductions in CV for all ROIs for both cognitively normal and impaired cohorts. Importantly, this translated to much smaller samples needed to detect an effect of given size as illustrated in Fig. D.
Conclusions
Joint deblurring of serial datasets led to reduced CV levels and smaller sample size requirements. This finding is particularly critical in the context of drug development as it indicates lower sample sizes required to detect a drug-induced lowering of tau pathology. Thus, our deblurring approach shows promise in the development of novel tau-based quantitative biomarkers for aging and Alzheimer's disease .
A. Sample Images: Transverse slices from a subject showing the ROI labels, the high-resolution TI MR image, the uncorrected PET SUVR difference image, and the jointly deblurred PET (dbPET) SUVR difference image. The ROI labels include inferior temporal cortex (ITC), fusiform gyrus (FG), parahippocampal gyrus (PHG), and entorhinal cortex (EC). B. ROI Mean SUVR - Annualized Change: dbPET led to non-negative measures with reduced standard deviation relative to the original PET difference images. C. Coefficient ofVariation: dbPET generated reduced coefficients of variation (CV) for both cognitively normal and impaired cohorts. D. Minimum Sample Size: The reduction in CV translates to a dramatic reduction in the minimum required sample size for a given effect size.
