Reproducibility of [¹⁸F]MK-6240 kinetics in brain studies with shortened dynamic PET protocol in healthy/cognitively normal subjects

Abstract

Background

[¹⁸F]MK-6240 is a neurofibrillary tangles PET radiotracer that has been broadly used in aging and Alzheimer’s disease (AD) studies. Majority of [¹⁸F]MK-6240 PET studies use dynamic acquisitions longer than 60 min to assess the tracer kinetic parameters. As of today, no consensus has been established on the optimum dynamic PET scan time. In this study, we assess the reproducibility of [¹⁸F]MK-6240 quantitative metrics using shortest dynamic PET protocols in cognitively normal subjects. PET metrics were measured through two-tissue compartment model (2TCM) and Logan model to estimate VT and DVR, as well as SUVR from 90 to 120 min (SUVR_{90 − 120 min}) post-tracer injection for brain regions. 2TCM was carried out using the 120 min dynamic coffee break dataset (first scan from 0 to 60 min p.i., second scan from 90 to 120 min p.i.) and then repeated after stepwise shortening it by 5 min. The dynamic scan length that reproduced the 120 min dynamic scans-based VT to within 10% error was defined as the shortest acquisition time (SAT). The SAT SUVR_{90 − 120 min} was deduced from the SAT dataset by extrapolation of each image pixel time-activity curve to 120 min. The reproducibility of the 120 min dynamic scans-based VT_2TCM, DVR_2TCM, DVR_Logan, and SUVR using the SAT was assessed using Passing-Bablock analysis. The limits of reproducibility of each PET metrics were determined using Bland-Altman analysis.

Results

A dynamic SAT of 40 min yielded < 10% error in [¹⁸F]MK-6240 VT_2TCM’s for all brain regions, compared to those measured using the 120 min datasets. SAT-based analysis did not show statistically significant systemic or proportional biases in VT_2TCM, DVR_2TCM, DVR_Logan, or SUVR compared to those deduced from the full dynamic dataset of 120 min. A mean difference between the 120 min- and SAT-based analysis of less than 4%, 10%, 15%, and 20% existed in the VT_2TCM, DVR_2TCM, DVR_Logan, and SUVR respectively.

Conclusion

Kinetic modeling of [¹⁸F]MK-6240 PET can be accurately performed using dynamic scan times as short as 40 min. This can facilitate studies with [¹⁸F]MK-6240 PET and improve patients accrual. Further work would be necessary to confirm the reproducibility of these results for patients in dementia spectra.

Background

[¹⁸F]MK-6240 is a PET radiotracer used to diagnose, stage, and monitor response to treatment in patients with Alzheimer’s disease by abnormal Tau protein detection in the form of neurofibrillary tangles (NFT) in the brain [1, 2]. In the first clinical study using the [¹⁸F]MK-6240, Shuping et al. showed the potential of tau-PET imaging to image and quantitate Tau protein in dementia [3]. In their study, the investigators assessed for the first time the distribution of [¹⁸F]MK-6240 using 20 or 30 min static images of a 80–90 min dynamic PET scan that started concurrently with the radiotracer injection. Other studies with the same tracer have investigated both static and dynamic PET imaging to assess the prognostic and predictive values of Tau-PET in Alzheimer’s disease. For example, Guehl and co-workers showed the SUVR (ratio of standardized uptake value in the target tissue to that in cerebellar grey matter) to reach equilibrium 135 min p.i. in subjects with high binding (usually observed in AD patients), and 30 min post-injection in healthy controls respectively [2]. Similarly, dynamic PET imaging with up to 180 min have been proposed to assess the [¹⁸F]MK-6240 total volume of distribution (VT), distribution volume ratio (DVR) target tissue to that in cerebellar grey matter, and binding potential (BPND), using compartmental kinetic modeling [1, 2]. Furthermore, Salinas et al. performed a test-retest study to assess the repeatability of Tau quantitative metrics in static (using a 90–120 min post-injection time window) and a 150 min dynamic PET scan [4]. The authors showed up to ~ 20% and ~ 10% variability in VT and SUVR respectively for frontal, temporal, parietal, occipital, and insular lobes, hippocampus, amygdala, and cerebellum. In another study, Kolinger et al. [5] used a dual-time window acquisition of two 30-min dynamic acquisitions separated by a 60 min “coffee break”; The SUVR was deduced from the last 30 min dataset, while the DVR values were measured using compartmental kinetic modeling of the whole dynamic dataset, i.e., by combining the early and late dynamic image sets. Moreover, Vanderlinden et al. [6] and Betthauser et al. [7] showed an increased extracerebral [¹⁸F]MK-6240 uptake in both healthy subjects and those with Mild Cognitive Impairment (MCI) at 90 min post-radiotracer injection, which can increase the variability in longer scans evaluation due to spillover effects [6, 7].

A major limitation in such studies is the long scan time. Reduced PET scan time can offer several benefits, including increased patient comfort, fewer movement artifacts, lower imaging costs, and increased patient accrual rate [5, 8, 9].

In this study, we have investigated the feasibility to reduce the scan time in dynamic [¹⁸F]MK-6240 PET studies using cognitively normal subjects (CN). The kinetic parameters (VT and DVR) and SUVR deduced from the shortened scans protocol were compared to those from the standard ones.

Methods

Patients specifics

Twenty cognitively normal subjects, 14 female and 6 male (69 ± 8 years old), underwent 120 min dynamic [¹⁸F]MK-6240 coffee-break PET protocol and T1-MRI for this study on two separate days up to 60 days apart. The study has been approved by the institutional review board of our institution, and all subjects signed an informed consent form.

Image acquisition

PET

Subjects underwent dynamic coffee-break PET protocol study on a Siemens Healthineers Biograph™ mCT64 PET/CT scanner post intravenous bolus injection of 217 ± 66 MBq of [¹⁸F]MK-6240. No fasting or diet restrictions were required before the study. Dynamic PET of the brain started simultaneously with the radiotracer injection and continued for 60 min using the following time-binning sequence (12 × 10s, 2 × 60s, 1 × 120s, 1 × 240s, 10 × 300s) on the first scan. On the second scan 6 × 300s time-binning sequence was used, from 90 min p.i. to 120 min.

PET emission data were first corrected for attenuation using a low-dose CT (120 kVp; 45 mAs), scatter, and random events, and then reconstructed into a 400 × 400 × 109 matrix (voxel dimensions, 1.08 × 1.08 × 2.03 mm³) using 3D OSEM algorithm provided by the manufacturer (4 iterations and 21 subsets, and 4.0 mm cutoff frequency smoothing filter) with time-of-flight.

MRI

MRI were acquired on a Siemens Healthcare TIM TRIO 3.0T scanner. T1 magnetization-prepared rapid gradient-echo (T1-MPRAGE) acquisition protocol was used to provide optimal gray matter (GM)/white matter contrast using a 64-channel head coil, with TR = 2400 ms, TE = 16 ms, TI = 900s, FOV = 256 mm, and 0.5 mm isotropic voxels. T1-weighted images have a 512 × 512 × 416 pixels (256 × 256 × 208 mm) matrix, 16- bits per pixel, 2 pixels/mm resolution, 0.5 × 0.5 × 0.5 mm voxel size.

Image processing and quantification

Inter-frame PET motion correction was performed by coregistering each of the frames to the average of frames 16 and 17 (reference), using Normalized Mutual Information in PMOD^® v3.9 (Fuse-It tool).

The Brainstem, Frontal lobe, Temporal lobe, Parietal lobe, Occipital lobe, Insula, Parahippocampi, Fusiform, Cingulate, Precuneus, White Matter (WM), Cortical Gray Matter (GM), Cerebellar GM, Thalamus, Caudate, Putamen, Pallidum, Hippocampus, and Amygdala VOIs were segmented using Freesurfer (http://surfer.nmr.mgh.harvard.edu/) [10] and then overlaid on the PET dynamic images.

PET-to- T1-weighted MRI image registration was done using the same PMOD^® image registration tool above.

Compartmental kinetic modeling

The [¹⁸F]MK-6240 VT and DVR of each of the aforementioned segmented brain structures were measured using a two-tissue compartment model (VT_2TCM and DVR_2TCM) as well as Logan model (DVR_Logan) [11] with the cerebellar GM as reference tissue [1, 2]. 2TCM quantification was carried out using an image-derived input function (IDIF) with the method described by Kang et al., 2018 [12]. Specifically, the lumen was first segmented using a 4 mm diameter ROI in 4 consecutive slices over the C2 portion of the carotid arteries (bilateral) and in the third PET time frame (i.e. 20–30 s p.i.). The lumen VOI was then dilated within each of the 4 slices by 2 PET voxels (i.e. ~2.2 mm), and the resulting hollow disk region was used to correct for spill-over to adjacent epithelial tissue. The epithelial VOI was then again dilated by 4 voxels (i.e. ~4.8 mm), resulting in a background VOI. Partial volume correction (PVC) to the IDIF was finally carried out using a regional voxel-based PVC tool from PMOD and a 1.1 × 1.1 × 0.9 mm³ point spread function (estimated in an ongoing study). Whole blood IDIF was fitted with 3-exponential function and then metabolite correction was performed using population-based data previously reported by Guehl et al. 2019 [2].

Effect of Dynamic scan Time

Dynamic scans times ranging from 60 min to 30 min in 5 min intervals were investigated to identify the shortest acquisition time-window (SAT) that would reproduce the [¹⁸F]MK-6240 VT_2TCM using a 120 min dynamic scan (gold standard). The SAT was defined as the shortest dynamic time that would result in an error in VT_2TCM, compared to that measured from the 120 min dynamic dataset, that is within its limits of reproducibility (LOR) for all the segmented brain regions. 2TCM- and Logan reference tissue - based models, as well as the SUVR were also investigated.

The reproducibility of [¹⁸F]MK-6240 kinetic parameters, specifically VT_2TCM and DVR_2TCM and DVR_Logan respectively, deduced from the 120 min dynamic image set (gold standard), were then assessed using the aforementioned SAT that reproduced the VT_2TCM results, as described above.

SUVR was also measured using 90–120 min post-injection after 3-exponential extrapolating the SAT dynamic dataset to 120 min (SUVR_{90 − 120−Extrap}). Results were finally compared to those from the original 120 min dynamic dataset (SUVR_{90 − 120 min}). The cerebellar grey matter was used as a reference region in all cases [1, 2].

Statistical analysis

Statistical analysis was performed using a hierarchical approach. First, the correlations between deduced from the SAT VT_2TCM, DVR_2TCM and DVR_Logan, and those measured from the 120 min were respectively calculated using the Spearman correlation coefficient (R). Non-parametric Passing–Bablok regression analysis [13] was performed to test for systematic bias (95% confidence interval [CI] for intercept (α) does not include 0) and proportional (95% CI for slope (β) does not include 1) between the sets of parameters that exhibited a correlation R > 0.5; Passing-Bablock analysis should not be used in the case of weak correlations [13]. Random differences between the SAT and 120 min dynamic scans were measured using residual standard deviation. If the slope and intercept were not significantly different from 1 to 0, respectively, Bland–Altman analysis [14] was performed to calculate the 95% LOR, after testing for the normality assumption on the differences between the two sets of kinetic rate constants using the Kolmogorov–Smirnov test. All statistical analyses were performed using, using MedCalc Statistical Software version 20.217 (MedCalc Software bv, Ostend, Belgium; https://www.medcalc.org; 2020).

Results

VT_2TCM for all brain regions deduced from 120 min dynamic scans (gold standard) were reproducible down to 35 min dynamic scan time. Figure 1 shows the percent differences for pallidum, that presented the biggest variabilities in VT_2TCM, DVR_2TCM, DVR_Logan, and SUVR between 120 min and shortened dynamic scan times among all other brain regions studied here. In the subsequent analysis, we elected to use a conservative SAT threshold of 40 min.

Fig. 1

Variability assessed for VT, DVR and SUVR in Pallidum for different duration scan times, compared with 120 min scan

A strong and statistically significant Spearman correlation (R > 0.75; p-value < 0.05) was observed between the 40 min and 120 min -based for all brain regions in VT_2TCM Similar results were observed for DVR_2TCM but for the putamen (R = 0.73). For DVR_Logan, all brain regions presented strong correlation, except occipital, parahippocampi, precuneus and thalamus, that presented medium spearman correlation ( 0.5 < R < 0.75) with statistical significance. Frontal, GM, occipital, parietal, precuneus, putamen, temporal and WM presented strong correlation between SUVR_{90 − 120 min} and SUVR_{90 − 120Extrap}. All other regions presented medium correlation.

Figure 2 shows the average Spearman correlation, R, values between the 40 min and 120 min -based PET metrics, averaged over all subjects, for each of the brain segments.

Fig. 2

Bar plot of the correlation values between different parameters calculated from [¹⁸F]MK-6240 PET shortened scan (40-min) and full scan (120-min). Dashed line represents correlation cutoff (≥ 0.500)

In Fig. 3 (A-D) the Passing-Bablok plots for each of the four PET metrics deduced from 40 min vs. 120 min dynamic scans are shown; both intercept and slope of the best fits were within the corresponding 95% confidence intervals, thus resulting in no statistically significant systemic or proportional biases respectively in all cases. Furthermore, Bland-Altman’s plots which presents the LOR of each of the four PET metrics between the 120 min and 40 min dynamic scans, averaged over all brain regions, are presented in Fig. 3 (E-H). The detailed LOR results for each of the brain regions studies here are tabulated in Table 1. For VT_2TCM, DVR_2TCM, and DVR_Logan the corresponding mean percent differences between the 40 min- and 120 min- based measurements, as well as the respective LOR’s were within ~ 20%. However, an increase error of up to ~ 30% in SUVR_{90 − 120−Extrap} was observed particularly for small brain regions.

Fig. 3

Passing-Bablock plots for (A) VT_2TCM, (B) DVR_2TCM, (C) DVR_Logan, and (D) SUVR and Bland-Altman plots (E) VT_2TCM, (F) DVR_2TCM, (G) DVR_Logan, and (H) SUVR for comparison between the shortened-time scan and full-time scan

Table 1 Bland-Altman results for comparison in VT_2TCM, DVR and SUVR calculated from 40 and 120 min [¹⁸F]MK6240 PET scans. Regions with correlation R < 0.5 were excluded from further statistical analysis, i.e. passing-Bablok and then Bland-Altman, as was recommended by passing and Bablok et al. [13]

Discussion

[¹⁸F]MK-6240 PET imaging is routinely performed using either dynamic acquisitions of up to 90 min [1, 2, 5] to deduce the corresponding kinetic parameters, specifically the DVR, or static acquisition, mostly between 70 min and 120 min post-radiotracer injections to measure the SUVR [1,2,3,4,5,6,7,8]. The long scan time in the formal and the long wait time in the latter can jeopardize patients accrual rates and increase the associated imaging costs. In this study, we have investigated the feasibility to shorten the PET dynamic scan time and assessed the respective errors in the [¹⁸F]MK-6240 quantitative metrics.

Twenty cognitively normal subjects underwent a 120 min dynamic [¹⁸F]MK-6240 PET imaging, as per our protocol. The DVR for a set of selected brain regions were measured using [1] full 2TCM with image-derived input function measured from the carotid arteries, and [2] graphical Logan model with the cerebellar GM as reference tissue. The latter has the advantage that it is more robust and does not require an input function which suffers from partial-volume effect, and hence can result in large errors when assessing the tracer kinetic metrics.

Detailed analysis using Passing-Bablok showed reproducibility of the 120 min-based aforementioned [¹⁸F]MK-6240 metrics using the first 40 min dynamic data post-radiotracer injection with no systemic or proportional biases. Table 1 summarizes the average percent differences together with the corresponding limits of reproducibility between the 40 min and 120 min dynamic scans-based measurements of VT_2TCM, DVR_2TVM, and DVR_Logan for each of the brain regions studied here. Less than 10% error and 20% LOR’s respectively were observed in all cases (Table 1) but for the Putamen and Pallidum (LOR: -28.32% – 0.22%). SUVR_{90 − 120−Extrap} increased percent error, yet less than 20% with LOR of as much as ~ 45%, compared to SUVR_{90 − 120} measured from the last 30 min of full dynamic dataset, especially for small brain region, which may be due to their higher susceptibility to uncertainties in the inter-frame motion correction.

The regions that did not presented strong correlation were not analyzed with Passing-Bablok and Bland-Altman, and most part of this was presented in SUVR measurements. These results could be due to the high variability already presented in previous studies [5,6,7], as well as failure of the kinetic model to predict the behavior of the tracer at later time points.

The attempt to shorten the dynamic scan time to 30 min, based on the first part of the dual time window protocol previously described by Kolinger and collaborators [5], failed in our study for some regions such as Amygdala, Parahippocampi, and Accumbens nuclei, due to its variability between scans achieving higher than 30% (Data not shown). The reason for this could be the small size of the aforementioned regions, which are more susceptible to patient head motion, as well as to pill-in from neighboring brain structures.

The [¹⁸F]MK-6240 extracerebral uptake which becomes more prominent beyond 60 min post-radiotracer injection, as it was previously shown by other studies [4, 6, 7], can also yield biases in the cortical region quantification, increasing variability in kinetic parameters, and require accurate partial volume effect correction. With shortened dynamic scans from 0 to 40 min post-injection we can avoid this influence of extracerebral [¹⁸F]MK-6240 uptake, especially when moving to voxel-based analysis.

Another source of variability on this study is the use of IDIF as input function for 2TCM and Logan graphical model [12]. As previously shown [12, 15], the used method presents good agreement with arterial input function and therefore it was chosen due the lack of AIF.

One limitation in this study is the lack of subjects with MCI/AD. Previous studies have shown that MK6240 exhibits slower kinetics in MCI and AD compared to normal brain [1, 2]. However, 2TCM DVR was shown to be time stable (with minim bias) between 50 min and 120 min, with no dependence on the disease status, i.e. healthy vs. MCI/AD [2] and hence it is expected to be a more robust metric to quantitate MK6240 in the brain than SUVR. Yet, still, it would therefore necessary to repeat this work in a cohort of MCI/AD patients to ensure the reproducibility of our results, or determine the corresponding shortest dynamic scan time that would reproduce the Mk6240 kinetics deduced from 120 min dynamic scan (gold standard).

Shortening clinical imaging acquisition times, as in this study, can have multiple advantages such as improving patients comfort, increasing accrual rate, reducing imaging cost, reducing patient motion, etc. Besides, it can facilitate the clinical implementation of dual-tracer PET imaging protocols such as in Nehmeh et al. [16]. Specifically for AD, Rowe et al. showed a correlation between Tau and Ab depositions in brain that were measured using MK6240 and PIB PET respectively, in both AD and cognitively unimpaired controls subjects [17]. Combined information on Tau and Ab can synergistically allow more detailed understanding of their interaction and provide synergistic prognostic and predictive values that are superior to that of each of the two tracers when imaged separately. This requires minimizing the dynamic scan time of each of the two tracers involved, for which our results from this study constitute a building block.

Conclusion

We have showed the reproducibility of [¹⁸F]MK-6240 kinetics using shortened dynamic PET scan times of only 40 min to within 10% using both 2TCM as well as the Logan model. The latter will spare the need for an input function, and hence will facilitate the implementation of [¹⁸F]MK-6240 kinetic modeling. Extrapolation of the 40 min dynamic to estimate the SUVR at delayed time points resulted in ~ 20% uncertainty, which would not be useful to assess small changes in MK6240 due to disease progression or response to putative drug. DVR is a more robust metric to quantitate MK6240 uptake in the brain than SUVR when using shortened PET dynamic acquisition time of 40 min.