Considerations and code for partial volume correcting [18F]-AV-1451 tau PET data

Abstract

[¹⁸F]-AV-1451 is a leading tracer used with positron emission tomography (PET) to quantify tau pathology. However, [¹⁸F]-AV-1451 shows "off target" or non-specific binding, which we define as binding of the tracer in unexpected areas unlikely to harbor aggregated tau based on autopsy literature [1]. Along with caudate, putamen, pallidum and thalamus non-specific binding [2], [3], we have found binding in the superior portion of the cerebellar gray matter, leading us to use inferior cerebellar gray as the reference region. We also addressed binding in the posterior portion of the choroid plexus. PET signal unlikely to be associated with tau also occurs in skull, meninges and soft tissue (see e.g. [4]). We refer to [¹⁸F]-AV-1451 binding in the skull and meninges as extra-cortical hotspots (ECH) and find them near lateral and medial orbitofrontal, lateral occipital, inferior and middle temporal, superior and inferior parietal, and inferior cerebellar gray matter. Lastly, the choroid plexus also shows non-specific binding that bleeds into hippocampus. We are providing the code (http://www.runmycode.org/companion/view/2798) used to create different regions of interest (ROIs) that we then used to perform Partial Volume Correction (PVC) using the Rousset geometric transfer matrix method (GTM, [5]). This method was used in the companion article, "Comparison of multiple tau-PET measures as biomarkers in aging and Alzheimer's Disease" ([6], DOI 10.1016/j.neuroimage.2017.05.058).

Fig. 1

OC subject. A: MPRage. B: Corresponding [18 F]-AV-1451 scan with no PVC. C: Rousset PVC image using ROI group 7. D: Calculated pre-PVC, C smoothed to the resolution of the scanner. E: Calculated pre-PVC – original SUVR (D-B).

Fig. 2

The standard deviation of the residuals normalized by the original SUVR reflects how well the PVC ROI configuration (x-axis) explained the original SUVR data. The lower the value (y-axis), the better the ROI configuration explained the original SUVR data. ROI configuration 1 was all ROIs in Table 1 + choroid plexus. ROI configuration 2, choroid plexus was divided into low and high ROIs. ROI configuration 3 was the same as 2 plus ECH (SUVR threshold=1.6). ROI configuration 4 was the same as 2, no ECHs, and the c3 mask and c4+c5 mask were added. ROI configuration 5 included ECHs (threshold=1.6) and c3 ROI and a c4+c5 ROI. ROI configuration 6 was the same as ROI configuration 4 except the c3 mask and c4+c5 mask were both divided into high and low masks (threshold=1). ROI configuration 7 included ECHs (threshold=1.6), c3 low and high (threshold=1) and c4+c5 low and high (threshold=1) ROIs. ROI configuration 8 was the same as 7 except the inferior cerebellar gray was scanned for high voxels. ROI configuration 9 was the same as 8 except a threshold of 1.3 was used for ECH. ROI configuration 10 was the same as 8 and 9 except a threshold of 1.9 was used for ECH. The ROIs used for calculating the normalized standard deviation corresponded to those in Table 1.

Fig. 3

Examples of binding in dorsal cerebellum, bleed in from neighboring ROIs into dorsal cerebellum, as well as bleed in from ECH into inferior cerebellum.

Fig. 4

Areas of high (A) and low (B) choroid plexus binding are shown in the masks (top row) and SUVR image (bottom row) for one sample subject. Histograms from 4 subjects show the bimodal distribution of the choroid plexus [¹⁸F]-AV-1451 binding.

Fig. 5

young healthy control subject with ECH near the occipital lobe. A is the original SUVR image. B has the calculated pre-PVC for the ROI group with an ECH threshold of 1.3. C is the same as B but ECH threshold was 1.6, and D had an ECH threshold of 1.9.

Fig. 6

Age versus volume of ECH; Spearman correlation shows age is correlated with the volume of ECH in healthy controls but is not significant in the AD/MCI subgroup within that subsample.

Supplementary Fig. 1

For each subject, the whole choroid plexus distribution of SUVR values was analyzed. A. shows the cutoff for each subject of the bimodal distribution, B. the kurtosis and C. the skewness.