Introduction

Dynamic positron emission tomography (PET) scan protocols allow for accurate quantitative measures [1, 2] of specific binding of PET tracers. Moreover, dynamic scan protocols yield additional information about functional measures such as perfusion [24], which is incorporated in PVElab [25], was used to delineate regions of interest (ROIs) on the co-registered MR scan and superimposed onto the dynamic PET scan to obtain regional time activity curves (TACs). All 68 cortical and subcortical regions from the Hammer template were included. Regional TACs extracted from the PET scans were analysed using a reversible 2-tissue compartment model with blood volume correction (2T4k_VB) and simplified reference tissue model (SRTM) [26]. Receptor parametric map** (RPM) [27] and standardized uptake value ratios (SUVr) were used to obtain parametric images. Cerebellum grey matter (obtained from PVElab) was used as the reference region.

Shortening the Second Part of the Scan (80–130 Min P.I.)

In these analyses, the first part of the scan remained 0–60 min p.i. The second part of the scan was shortened; three shorter time intervals were explored: 80–120 min, 80–110 min and 80–100 min. For each subject, shortened PET scans were acquired by removing 2 to 6 frames to reach the specified scan intervals. Reference region TACs were extracted from these shortened PET scans to estimate kinetic parameters. BPND and R1 values were estimated using RPM from the three different scan durations (0–60/80–100, 0–60/80–110 and 0–60/80–120 min). RPM-derived regional BPND and R1 values were compared to the corresponding non-linear regression (NLR)-based SRTM-derived BPND and R1, and plasma input–derived distribution volume ratio (DVR) values from the original scan duration (0–60/80–130 min). The optimal shortened time interval for the second part was used and fixed during subsequent evaluation of scan shortening of the first part of the PET scan.

Shortening the First Part of the Scan (0–60 Min P.I.)

For shortening the first part of the scan, three time intervals were explored: 0–50, 0–40 and 0–30 min p.i, all in combination with 80–100 min scan interval for the second part of the imaging protocol. For each subject, the corresponding frames were removed to obtain the PET scans with these specified time intervals.

The original scan duration had a gap of only 20 min; the gap in the reference region was interpolated by using cubic interpolation. The larger gap (> 20 min) in the new dual-time-window protocol results in more missing data points in the reference TAC for which proper interpolation is required. Therefore, four different interpolation methods were assessed: population-based plasma input function in combination with a reversible two-tissue compartmental model with blood volume correction (POP-IP_2T4k_VB) to fit the reference tissue TAC, standard cubic interpolation, linear interpolation, and interpolation based on fitting an exponential function to the TAC (excluding points until peak uptake). All scripts were built in house using MATLAB (version R2017B, MathWorks, USA).

The POP-IP_2T4K_VB interpolation method was based on using the population-averaged metabolite-corrected plasma input function and a reversible two-tissue compartmental model with blood volume correction (2T4k_VB). A 2T4k_VB model was used, since it was evaluated in the previous studies [28] that this model best describes the in vivo kinetics of [18F]flortaucipir. So based on the previous research, it was assumed that the cerebellum presents a 2T4k_VB kinetics and the cerebellum TAC with the gap was fitted using this model and the population-averaged metabolite-corrected plasma input function. The fit was visually examined for certainty and the gap in the cerebellum TAC was filled using the values from the fitted curve.

SRTM-derived BPND and R1 estimates using the shortened scan durations and the four different interpolated reference region TACs were obtained. These regional parametric values were compared to the corresponding NLR-based reference region and plasma input–derived values obtained using the original scan duration (0–60/80–130 min).

BPND and R1 parametric images were acquired for the optimally shortened scans with interpolated reference region (using the optimal interpolation technique(s)). Regional parametric values were extracted from these parametric images and were compared to corresponding values derived using plasma input–based and reference tissue–based NLR and RPM from the original scan duration (0–60/80–130 min).

SUVr using the interval 80–100 min (SUVr(80−100 min)) was also evaluated. Regional SUVr values obtained from this time interval were compared with the respective quantitative parameters (DVR, SRTM BPND and RPM BPND) estimated using the original scan duration (0–60/80–130 min).

Test-Retest Repeatability Analysis

For the test-retest repeatability (TRT) analysis, the test-retest cohort was used. The TRT of RPM BPND and R1 values derived from the optimal shortened scan duration were compared to the test-retest repeatability of RPM BPND obtained using the original scan duration (0–60/80–130 min). In addition, TRT for SUVr(80–100) was also assessed. The TRT was calculated using Eq. 1.

$$ \mathrm{TRT}\ \left(\%\right)=\frac{\left(\mathrm{Retest}\ \mathrm{value}-\mathrm{Test}\ \mathrm{value}\right)\ }{\left(\mathrm{Retest}\ \mathrm{value}+\mathrm{Test}\ \mathrm{value}\right)} \times 200 $$
(1)

Statistical Analysis

Linear regression fitting and correlation coefficients (r2) were used to compare BPND and R1 for the shortened scan durations and SUVr(80−100 min) against corresponding parametric values for the original scan duration (0–60/80–130 min) derived from plasma input–based and reference tissue–based NLR and RPM. Furthermore, Bland-Altman plots were used to assess and illustrate TRT performance.

Results

Shortening the Second Part of the Scan (80–130 Min P.I.)

The RPM BPND values obtained from the three shortened scan durations (0–60/80–120, 0–60/80–110 and 0–60/80–100 min) provided excellent correlations with plasma input DVR-1, SRTM BPND and RPM BPND obtained using the original acquisition time window (0–60/80–130) (Table 1; all r2 > 0.93). Reduction of the time interval of the second part to 100 min had negligible effects on the RPM BPND estimation: correspondence to DVR-1 (HC: r2 = 0.94, slope = 0.95; AD: r2 = 0.94, slope = 0.92), SRTM BPND (HC: r2 = 0.98, slope = 1.05; AD: r2 = 0.96, slope = 0.85) and RPM BPND (HC: r2 = 0.98, slope = 1.04; AD: r2 = 0.99, slope = 0.94). Comparison of regional SRTM BPND values obtained using the shorter time interval (0–60/80–100) to plasma input DVR-1 and SRTM BPND obtained with the original scan duration (0–60/80–130) are presented in Supplementary Table 1.

Table 1. RPM BPND obtained using shorter time intervals compared to plasma input DVR-1, SRTM BPND and RPM BPND obtained with the original scan duration

The RPM R1 values obtained from the three shortened scan durations (0–60/80–120, 0–60/80–110 and 0–60/80–100) also provided excellent correlations with SRTM R1 and RPM R1 estimated using the original acquisition time window (0–60/80–130) (Supplementary Table 2).

Shortening the First Part of the Scan (0–60 Min P.I.)

In Fig. 1, the different interpolations of a typical reference TAC for the shortest dual-time-window (0–30/80–100 min) assessed in this study are presented. For all shortened scan durations, SRTM BPND using the reference TACs interpolated with either POP-IP_2T4k_VB or cubic interpolation methods had the best correspondence with plasma input DVR-1 and SRTM BPND (r2 > 0.90, Table 2) obtained with the original scan duration. Reduction of the time interval of the first part of the scan to 30 min and using POP-IP_2T4k_VB for reference region interpolation had negligible effects on the quantitative accuracy of the estimated kinetic parameters with respect to that estimated using the original scan duration: DVR-1 (HC: r2 = 0.93, slope = 0.94; AD: r2 = 0.92, slope = 0.97) and SRTM BPND (HC: r2 = 0.96, slope = 1.02; AD: r2 = 0.98, slope = 0.89). SRTM BPND values obtained with 0–30/80–100 min data using cubic interpolation for reference region had similar agreement with DVR-1 (HC: r2 = 0.94, slope = 0.91; AD: r2 = 0.91, slope = 0.92) and SRTM BPND (HC: r2 = 0.96, slope = 0.98; AD: r2 = 0.98, slope = 0.85) from the original scan duration. Good correlations were observed for linear and exponential interpolation methods (r2 > 0.90, Table 2). However, these interpolation methods resulted in higher underestimation (15–25 %) of the parametric values.

Fig. 1.
figure 1

Interpolation of the gap in reference region TAC (30 to 80 min) with different interpolation methods.

Table 2. Shortened time intervals interpolated using four different methods are compared with plasma input DVR-1 and SRTM BPND obtained with the original scan duration

Figure 2 presents the correspondence of SRTM BPND values obtained with the shortened scan durations using POP-IP_2T4k_VB for reference region interpolation against SRTM BPND values estimated from the original scan duration. The bias increased as the first part was shortened. An underestimation of 9 % was observed for SRTM BPND values with the shortened scan duration for both groups (0–30/80–100 min) with respect to that obtained with original scan duration. SRTM R1 values derived from the shortened scan duration (0–30/80–100 min) showed excellent correlations with SRTM and RPM R1 values obtained with the original scan duration for HC and AD patients for each interpolation method (r2 > 0.95, Supplementary Table 3).

Fig. 2.
figure 2

Comparison of SRTM BPND estimated using shortened time intervals a 0–50/80–100 min, b 0–40/80–100 min, c 0–30/80–100 min and POP-IP_2T4k_VB interpolation for reference region against SRTM BPND obtained from the original scan duration (0–60/80–130 min). LOI, line of identity.

An example of the RPM BPND images for the original scan duration and shortened scan duration (0–30/80–100) is illustrated in Fig. 3a. Comparison of RPM BPND obtained from the shortest scan duration (0–30/80–100 min) using POP-IP_2T4k_VB for reference region interpolation against RPM BPND obtained with the original scan duration is shown in Fig. 3b and Supplementary Fig. 1. RPM BPND obtained with the shortest scan duration (0–30/80–100 min) using either POP-IP_2T4k_VB or cubic methods for reference region interpolation and SUVr80-100 min compared to original DVR-1, SRTM BPND and RPM BPND values are shown in Table 3 for HC and AD patients separately. The same comparisons were made for RPM R1 and are illustrated in Supplementary Table 4.

Fig. 3.
figure 3

a An example of the BPND images for a representative AD patient are displayed for the original scan (0–60/80–130 min) duration and shortened scan duration (0–30/80–100 min using POP-IP_2T4k_VB interpolation) along with the corresponding MR. b Comparison of BPND obtained from the shortened scan duration (0–30/80–100 min using POP-IP_2T4k_VB interpolation) against BPND obtained with the original scan duration (0–60/80–130 min). c Bland-Altman plot of the original test-retest differences for RPM DVR (BPND+1) values. d Bland-Altman plot of the test-retest differences for RPM DVR (BPND+1) values using shortened scan duration (0–30/80–100 min) and POP-IP_2T4k_VB method for reference region interpolation.

Table 3. Comparison of RPM BPND obtained with the shortest scanning interval (0–30/80–100) and SUVr(80-100 min) to plasma input DVR-1, SRTM BPND and RPM BPND derived from the original scan duration (0–60/80–130)

Test-Retest Repeatability

The RPM BPND values estimated for both test and retest scans using the shortest assessed scan duration (0–30/80–100) and POP-IP_2T4k_VB or cubic interpolation for the reference region correlated well (Table 4). The test-retest parametric correlations using the short scan duration and interpolated reference TAC (POP-IP_2T4k_VB and cubic interpolation methods) were similar to the correlations using the original scan duration. The TRT across all Hammers’ regions of interest for RPM BPND was − 1 % ± 4 for HC and 0 % ± 4 for AD patients using the original dual-time-window acquisition (Table 4). Furthermore, TRT for R1 was 0 % ± 6 for both groups using the original dual-time-window acquisition. The TRT for BPND remained the same for HC when using the shortest dual-time-window 0–30/80–100 min with either of the interpolations methods (POP-IP_2T4k_VB or cubic interpolation). For AD patients, the TRT changed to 0 % ± 5 when using the shortest dual-time-window 0–30/80–100 min with POP-IP_2T4k_VB or cubic interpolation. The TRT for R1 remained similar as with the original data (0 % ± 6) for the shortest dual-time-window (0–30/80–100 min) both with POP-IP_2T4k_VB and cubic interpolation for both groups. TRT for SUVr80-100 min was − 1 % ± 5 for HC and − 1 % ± 6 for AD patients (Table 4).

Table 4. RPM BPND, RPM R1 and SUVr values obtained from the test scan are compared to corresponding values obtained from the retest scan for the original scan duration (0–60/80–130), and for the shortened scan duration (0–30/80–100) interpolated with cubic or POP-IP_2T4k_VB interpolation method

Bland-Altman plots for the RPM BPND+1 illustrating TRT for all Hammers’ regions obtained with the original scan duration and the shortest scan duration with POP-IP_2T4k_VB reference region interpolation are presented in Fig. 3c and d.

Discussion

The current study demonstrated that for [18F]flortaucipir expanding the break in the dual-time-window protocol with just a 50-min overall scanning time (early interval of 0–30 min, than a coffee break, followed by a late interval of 80–100 min) had minimal effect on the quantitative accuracy. The optimal shortened dual-time-window protocol (0–30/80–100 min) allows sufficiently accurate estimation of BPND while reducing patient burden and enables interleaved scanning, where other patients could use the camera during breaks within the scan period.

An excellent correlation was observed between the shortened acquisition protocol (0–30/80–100 min) and the original dual-time-window (0–60/80–130 min) protocol. Four different interpolation methods were used to interpolate the missing data between the two time windows for the reference region TAC (cerebellum grey matter). According to our results, POP-IP_2T4k_VB interpolation, which uses a population-averaged plasma input function, showed a good correspondence of the estimated kinetic parameters to that obtained from the original scan protocol, and the lowest under/over-estimation(s) compared to other interpolation methods. Heeman et al. [29] showed that POP-IP_2T4k_VB interpolation method also works well to interpolate the missing data points in a dual-time-window protocol for [18F]flutemetamol and [18F]florbetaben. They concluded that the introduction of a gap with a maximum of 60 min in a dual-time-window protocol (early interval of 0–30 min followed by a late interval of 90–110 min) does not affect quantitative accuracy for [18F]flutemetamol and [18F]florbetaben. As such, POP-IP_2T4k_VB interpolation does not only work well for [18F]flortaucipir but also for [18F]flutemetamol and [18F]florbetaben, possibly because the model describes the in vivo kinetics of the tracers best and is therefore ideal for estimating the missing reference region data points accurately.

The correlations for all interpolation methods were comparable (Table 2). This was not expected, since linear and exponential interpolation did not follow the course of tracer as can be observed in Fig. 1. A possible explanation could be that the gap between the dual-time-windows was not substantially large enough to see significant differences in correlations between the interpolation methods. However, a substantial underestimation (at times up to 20 % or even more) was observed in AD patients for the shortened SRTM BPND values obtained with linear and exponential interpolation methods when compared to plasma input DVR-1 and SRTM BPND obtained with the original scan duration. This indicates that these interpolation methods are not suitable for quantitatively accurate kinetic parameter estimation for [18F]flortaucipir. For SRTM BPND values obtained with POP-IP_2T4k_VB interpolation, the biases remained within ~ 10 % for the same comparisons (Table 2). Comparisons of RPM BPND obtained with the shortest scanning interval (0–30/80–100) to plasma input DVR-1, SRTM BPND and RPM BPND derived from the original scan duration (0–60/80–130) showed that POP-IP_2T4k_VB interpolation had a minimal and acceptable bias of ~ 5 % (Table 3). Since RPM BPND is the main parameter outcome for [18F]flortaucipir, shortening the scan duration to 0–30/80–100 min with POP-IP_2T4k_VB interpolation for the reference region will provide quantitative acceptable accurate results. However, when individual regions are assessed, relatively higher bias (~ 8 %) was observed in regions with higher tau uptake (RPM DVR > 2) (Supplementary Fig. 1).

From Table 3, it is evident that SUVr(80-100 min) presents over-/underestimations when compared to DVR-1, SRTM BPND and RPM BPND (using 0–60/80–130 min scan duration data). In contrast, parameters estimated using a shortened scan duration data had a much better correspondence with the parameters estimated using the whole scan duration data (0–60/80–130 min). Although with a shortened scan duration, a 50-min scanning time is required, which is 30 min more than that required for a static scan; SUVr is still semi-quantitative. Moreover, it is known that SUVr might be effected by blood flow changes overtime, and hence is an unreliable parameter for longitudinal studies, whereas using the shortened dual-time-window protocol (0–30/80–100), a flow estimate (R1) to evaluate the effect of flow can be estimated and therefore is apt for longitudinal studies. Moreover with the proposed method, only 50 min of actual scanning time is required on contrast to 110 min of actual scan duration (decreasing the patient burden by 60 min of scanning).

Dynamic [18F]flortaucipir PET scans can be used to estimate both the specific binding of the tracer to tau pathology (BPND) and as well as rCBF (R1) using RPM. Perfusion imaging is a reliable biomarker to assess neuronal dysfunction in neurodegenerative diseases [30]. R1 is an estimate of the relative blood flow, and it has been shown that it has a high correlation with [18F]FDG uptake [30] and so an accurate estimation of R1 is also necessary. The current study demonstrated that shortening the scan duration to 0–30/80–100 min had negligible effects on R1 estimations. As such, it can be safely assumed that the scanning protocol can be reduced to 0–30/80–100 min with minimal bias (~ 7 % on average). If the implementation for POP-IP_2T4k_VB interpolation is not possible, use of cubic interpolation is a reliable alternative to interpolate the gap between the dual-time-windows for the reference region.

Unfortunately, further reduction of the acquisition time for the first part was not possible since the bias in the estimated parameters was increasing as the scan duration was being reduced (Fig. 2). A reason could be that as the peak approaches, further loss of information makes it difficult to estimate the efflux kinetics of [18F]flortaucipir. Further reduction of the second part of the scan was also not possible, since we want to be able to compare with the internationally mostly used 80–100-min SUVr time interval.

The shortened dual-time-window (0–30/80–100 min) acquisition protocol provided almost identical TRT values when compared to the observed TRT values for the original scan protocol (0–60/80–130) (Fig. 3). This implies that the shortened dual-time-window protocol not only provides quantitatively acceptable estimates but also result in high repeatability, suggesting that it can be reliably used for longitudinal and treatment-monitoring studies. However, the benefits of using a dynamic scanning protocol over static scanning protocol in a longitudinal setup for [18F]Flortaucipir need further validation, even though it has been presented by van Berckel et al. [16] that SUVr does get affected by changes in flow and under-/overestimates the underlying specific binding in case of [11C]PIB. Making use of a dual-time-window protocol also has some challenges to consider. Even if the total acquisition time is reduced with 60 min, it still takes longer than a single static acquisition protocol used for SUVr. Another disadvantage of using a dual-time-window protocol is the added CT attenuation scan for the second part of the scan, which is still present in this new shortened scanning protocol. Furthermore, the pre-processing of the PET images are more demanding compared to simplified methods, for instance, due to a required additional co-registration of the first part to the second part of the scan. Yet, the main advantage is that, apart from obtaining quantitative information on BPND, this protocol also generates parametric R1 data that may be used as a surrogate for flow and/or FDG uptake [15] and this protocol could therefore obviate the need to make a separate FDG scan, when clinically required.

Conclusion

The current study demonstrated that quantitatively acceptable [18F]flortaucipir kinetic parameters can be obtained using just 50 min of total scanning time by implementing a dual-time-window protocol (0–30/80–100 min). The best method to interpolate the gap in the reference tissue is 2T4k_VB tracer kinetic model with population-averaged metabolite-corrected plasma input function. Reducing the dual-time-window protocol enables interleaved scanning, reduces patient burden and enables efficient use of tracer batches and cost-effectiveness.