Comparing Tau PET Visual Interpretation with Tau PET Quantification, Cerebrospinal Fluid Biomarkers, and Longitudinal Clinical Assessment

Study participants

Participants selected for this study were enrolled in longitudinal observational studies of aging and memory at the Charles F. and Joanne Knight Alzheimer Disease Research Center (Knight ADRC, n = 189, of whom 23 were cognitively impaired, Table 1). All participants met the inclusion criteria of having a tau PET usable for visual reads, and an Aβ PET, MRI, and clinical and cognitive evaluation, all within 18 months; the majority of participants (n = 144) also underwent lumbar puncture within 18 months of their tau PET scan. The study was approved by the Washington University in St. Louis Human Research Protection Office and Institutional Review Board, and all participants or their designees signed an informed consent form.

Clinical and cognitive assessment

Participants were assessed clinically and cognitively using the neuropsychological test battery from the Uniform Data Set (UDS) [13], which includes the Clinical Dementia Rating (CDR^®) [14] and the Mini-Mental State Examination (MMSE) [15]. The CDR assesses three domains of cognition (memory, orientation, judgment, and problem solving) and three domains of function (community affairs, home and hobbies, personal care): scores from the six domains can either be summed to yield the CDR Sum of Boxes score, or passed to a lookup table to yield the CDR Global score.

Tau PET acquisition

Participants were scanned on a Siemens Biograph 40 TruePoint (Siemens Healthineers). Participants received a single intravenous bolus injection (341±29.8 MBq) of ¹⁸F-flortaucipir (Tauvid^TM, Avid Radiopharmaceuticals). Emission data were collected 80–100 min post injection. List-mode data were reconstructed using ordered subset expectation maximization with three iterations and 21 subsets. A low-dose CT scan preceded PET acquisition for attenuation correction.

Tau PET SUVR

Reconstructed PET images were processed using the PET Unified Pipeline (https://github.com/ysu001/PUP) and coregistered to corresponding MR images [16, 17]. After segmenting MR images into ROIs using FreeSurfer version 5.3 [18], regional SUVRs were defined from the reconstructed PET images using a cerebellar gray reference region. The temporal meta-ROI SUVR was defined as the volume-weighted mean SUVR of the amygdala, entorhinal cortex, fusiform, parahippocampal, inferior temporal, and middle temporal ROIs [7, 8].

Tau PET visual interpretation

Two radiologists with training in nuclear medicine (J.A.L. and M.R.P.) followed the manufacturer’s guidelines for ¹⁸F-flortaucipir PET visual interpretation of participant scans using MIM Encore (MIM Software). Reconstructed PET images were coregistered with corresponding MR images. A ROI was drawn around the whole cerebellum in the axial plane that maximizes its cross-sectional area. A color scale with a rapid transition at 1.65 times the mean cerebellar counts was defined. The temporal lobe was divided into the anterolateral, anterior mesial, posterolateral, and posterior mesial temporal quadrants by placing the horizontal crosshair posterior to the brainstem nuclei, and the vertical crosshair at the widest portion of the temporal pole. An image was considered positive if it showed contiguous activity above the rapid transition/cutoff in the cortical gray matter of the posterolateral temporal, occipital, or parietal/precuneus regions. An image was considered negative if it showed no activity above the cutoff in the cortical gray matter of the posterolateral temporal, occipital, or parietal/precuneus regions, or if it showed activity above the cutoff in the cortical gray matter restricted to the medial temporal, anterolateral temporal, and frontal regions. Off-target binding, which may be seen in the choroid plexus, striatum, and brainstem nuclei, and small foci of noncontiguous activity, which may be seen throughout the cortical gray matter, were not used when determining tau positivity. Radiologists were blinded to all other information about each participant. In addition to following the manufacturer’s guidelines for ¹⁸F-flortaucipir PET visual interpretation, in this study, radiologists also reported whether radiotracer activity was symmetric across left and right hemispheres and whether there was off-target binding in the choroid plexus, striatum, brainstem nuclei, or bone/meninges. Notable findings (such as incidental meningiomas) were also reported. Both radiologists determined off-target binding and incidental findings using MR imaging. Additionally, incidental findings were confirmed with a neuroradiologist (T.L.S.B.).

Aβ PET

Participants were scanned on either a Siemens Biograph 40 TruePoint, Biograph mMR, or Biograph Vision 600 (Siemens Healthineers). Participants received either a single intravenous bolus injection (539±159 MBq) of ¹¹C-Pittsburgh compound B (PiB) or (369±22.4 MBq) of ¹⁸F-florbetapir (Amyvid^TM, Avid Radiopharmaceuticals). Emission data were either collected 30–60 min post injection (¹¹C-PiB) or 50–70 min post injection (¹⁸F-florbetapir). Reconstructed PET images were formed and pre-processed in the same manner as tau PET. An Aβ PET SUVR was defined for each radiotracer [16, 17] and standardized to the Centiloid scale [19, 20].

MR acquisition

Participants were scanned on either a Siemens Biograph mMR or Magnetom Vida (Siemens Healthineers). Across all scanners, T1-weighted head MR images were acquired using a magnetization prepared rapid gradient echo (MPRAGE) generalized autocalibrating partial parallel acquisition (GRAPPA) sequence using a repetition time = 2300 ms, echo time = 2.95 ms, flip angle = 9°, at 1.1×1.1×1.2 mm³ voxel resolution.

CSF

CSF was collected under standardized operating procedures. Participants underwent lumbar puncture in the morning following overnight fasting and 20–30 ml of CSF was collected in a 50 ml polypropylene tube via gravity drip using an atraumatic Sprotte 22-gauge spinal needle. CSF samples were kept on ice and centrifuged at low speed within 2 h of collection, then transferred to another 50 ml tube to remove cells. CSF was aliquoted at 500μl into polypropylene tubes and stored at –80°C. Concentrations of CSF p-tau181, Aβ₄₂, and Aβ₄₀ were measured by chemiluminescent enzyme immunoassay using a fully automated platform (LUMIPULSE G1200, Fujirebio) according to the manufacturer’s specifications.

Statistical analyses

Cutoffs for binarizing tau PET, Aβ PET, CSF p-tau181, and CSF Aβ₄₂/Aβ₄₀ values were determined by fitting a two-component univariate Manly mixture model [21] in R software [22] to all relevant baseline PET SUVR or CSF measurements available in the Knight ADRC Data Freeze 17 (Supplementary Table 1) and finding the decision boundary. Manly mixture modeling was used to account for possible severe skewness in the data that would be difficult to model using Gaussian mixture modeling, and to account for skewness that can vary from component to component, which would be impossible to model using log or Box-Cox transformations [21]. Cohen’s kappa (κ) was used to measure inter-rater reliability between the two radiologists’ tau PET visual interpretations, as well as between tau PET visual interpretation and tau PET SUVR quantification, and between tau PET visual interpretation and CSF p-tau181 concentration.

Study participants

Overall, participants were on average (±standard deviation) 69.8±8.51 years old, most were cognitively normal with a global Clinical Dementia Rating (CDR^®) [14] of 0 (n = 166/189, 87.8%) and most did not carry the APOE ɛ4 allele (n = 120/188, 63.8%) (Table 1). Cognitively normal participants had a mean tau PET temporal meta-ROI SUVR of 1.15±0.106 and a mean Centiloid of 19.9±34.4. Cognitively impaired participants (n = 23/189, 12.2%) had a clinical diagnosis of either uncertain dementia (n = 9), a CDR = 0.5 in memory only (n = 1), or AD dementia (n = 13). They also had a mean tau PET temporal meta-ROI SUVR of 1.44±0.364 and a mean Centiloid of 74.3±45.6.

The following quantitative cutoffs were identified through statistical modeling and are used to determine biomarker positivity in the remainder of the analyses: tau PET temporal meta-ROI SUVR cutoff = 1.32, Aβ PET (Centiloid) cutoff = 21.6, CSF p-tau181 cutoff = 58.1 pg/ml, and CSF Aβ₄₂/Aβ₄₀ cutoff = 0.0737.

Tau PET visual interpretation and tau PET SUVR

Of the 189 ¹⁸F-flortaucipir PET images, 20 (10.6%) were read as positive by both radiologists. Both radiologists also read 169 images as negative and thus agreed on the overall visual interpretation of each image in the current study (n = 189/189, 100%, κ= 1). Agreement between visual interpretation and SUVR quantification was high (n = 186/189, 98.4%, κ= 0.910) (Fig. 1).

Fig. 1

Comparison of tau PET visual interpretation with tau PET SUVR. Each PET study was assessed by visual interpretation using the manufacturer’s guidelines to determine positivity (x-axis) and by temporal meta-ROI SUVR analysis using a cutoff of SUVR = 1.32 to determine positivity (y-axis). The color indicates the Aβ PET status for each case (positive Aβ PET, red; negative Aβ PET, blue; cutoff = 21.6 Centiloids).

The three participants who had discordant results between visual interpretation and SUVR quantification all had tau-positive visual interpretations and tau-negative SUVRs. One participant (Fig. 2a) demonstrated elevated ¹⁸F-flortaucipir uptake in the right precuneus and was Aβ PET, CSF Aβ₄₂/Aβ₄₀, and CSF p-tau181 negative (Table 2). Additional MR imaging revealed a hypointensity on T2*-weighted MRI that colocalized with the elevated right precuneus radiotracer uptake on ¹⁸F-flortaucipir PET, suggesting a hemorrhagic infarct to be the cause of elevated radiotracer uptake instead of AD NFT deposition (Fig. 2b). Upon review of the additional T2*-weighted MR imaging, the readers also revised their interpretation of the image to be tau negative.

Fig. 2

Three cases with tau-positive visual interpretations, but tau-negative SUVRs. a) Tau PET coregistered with MRI of a male participant in his 80s with elevated right precuneus uptake. b) Corresponding T2*-weighted MRI showing a hypointensity (indicated by the crosshair) that colocalizes with the elevated right precuneus uptake from (a). c) Tau PET coregistered with MRI of a female participant in her 70s with elevated occipital lobe uptake, left greater than right. d) Tau PET coregistered with MRI of a female participant in her 70s with elevated posterolateral temporal, occipital, and parietal/precuneus lobe uptake, right greater than left.

The other two participants (Fig. 2c, d) demonstrated lateralized occipital uptake, with greater uptake in either the left (Fig. 2c) or right (Fig. 2d), and were Aβ PET, CSF Aβ₄₂/Aβ₄₀, and CSF p-tau181 positive (Table 2). The participant with greater left occipital uptake than right, likely has an occipital-predominant form of AD tau pathology [23].

The participant with greater right occipital uptake than left also had posterolateral temporal and parietal/precuneus uptake. The temporal meta-ROI SUVR was borderline negative, suggesting that, perhaps due to the lateralized uptake, the SUVR was artificially low for this case.

Incidental findings

In terms of incidental findings, frontal meningiomas were identified in two participants. One participant had a meningioma in their left posterior frontal lobe (Fig. 3a, b); the other participant had it in their left frontal lobe (Fig. 3c, d). Both meningiomas had elevated levels of radiotracer uptake. The first participant also had elevated right posterolateral temporal uptake and tau-positive visual interpretation and SUVR and was Aβ PET positive (Table 3). The other participant had tau-negative visual interpretation and SUVR and was Aβ PET negative.

Fig. 3

Two cases of incidental meningioma with tau PET uptake. a) Tau PET coregistered with MRI of a female participant in her 70s with a left frontal posterior meningioma (indicated by the crosshair) with tau radiotracer uptake. b) Corresponding MRI image. c) Tau PET coregistered with MRI of a male participant in his 70s with a left frontal meningioma (indicated by the crosshair) with tau radiotracer uptake. d) Corresponding MRI image.

CSF p-tau181

Agreement between visual interpretation and CSF p-tau181 was moderate (n = 124/144, 86.1%, κ= 0.526, Table 4). Two participants had tau-positive visual interpretations but were CSF p-tau181 negative (Fig. 4a, b). One participant was previously identified as having a tau-positive visual interpretation but tau-negative SUVR (the same case as in Fig. 2a, b). The other participant demonstrated posterolateral temporal uptake in both hemispheres and was Aβ PET and CSF Aβ₄₂/Aβ₄₀ positive. In addition, 18 participants had tau-negative visual interpretations but were CSF p-tau181 positive (Fig. 4a, b). These cases were mostly Aβ PET positive (n = 14/18, 77.8%) and/or CSF Aβ₄₂/Aβ₄₀ positive (n = 17/18, 94.4%).

Fig. 4

Comparison of tau PET visual interpretation with CSF p-tau181 concentration. Each participant is plotted by visual interpretation (x-axis) and CSF p-tau181 concentration (y-axis); participants with p-tau181 ≥58.1 pg/ml were considered positive. In (a), the color indicates the Aβ PET status for each participant (positive Aβ PET, red; negative Aβ PET, blue; cutoff = 21.6 Centiloid). In (b), the color indicates the CSF Aβ₄₂/Aβ₄₀ status for each participant (positive CSF Aβ₄₂/Aβ₄₀, red; negative CSF Aβ₄₂/Aβ₄₀, blue; cutoff = 0.0737). In (c), the color indicates the tau PET temporal meta-ROI SUVR status for each participant (positive tau PET, red; negative tau PET, blue; cutoff = 1.32 SUVR).

Clinical and cognitive assessment

Six participants were assessed at baseline to be cognitively normal but tau-positive on visual interpretation (Table 5). One participant (Case 1) was previously mentioned to have PET radiotracer uptake colocalized to a parietal/precuneus hypointensity on T2*-weighted MRI and no other positive AD biomarkers (Fig. 2 and Table 2). The remaining five participants were all Aβ PET positive. No participant reliably converted from cognitively normal to AD dementia. One participant (Case 2) did convert to AD dementia at their three-year follow up but was reassessed to have a clinical diagnosis of uncertain dementia, more specifically, possible non-AD dementia of vascular origin at their five-year follow up. Another participant (Case 4) converted to AD dementia at their two-year follow up but was reassessed to have frontotemporal dementia (FTD) at their four-year follow up.

Four participants were assessed at baseline to be cognitively normal and tau-negative on visual interpretation but would convert to AD dementia at follow up (Table 5). Two participants (Case 7 and Case 10) converted to AD dementia at their one-year follow ups but were reassessed as cognitively normal at their two-year follow ups. The remaining two participants (Case 8 and Case 9) converted to AD dementia at their second- and fourth-year follow ups, respectively, but only Case 2 demonstrated Aβ PET positivity at baseline.

Twenty-three participants were assessed at baseline to have cognitive impairment (Table 6). Nine of these participants received a clinical assessment of uncertain dementia and two of the nine had a baseline tau-positive visual assessment. Both cases (Case 2 and Case 4) converted to AD dementia by their first- and second-year follow ups, respectively and were both Aβ PET positive. Nonetheless, three cases with a tau-negative visual interpretation at baseline (Case 5, Case 7, and Case 8) converted to AD dementia at their two-, two-, and three-year follow ups, respectively, although Case 5 was reassessed to be cognitively normal at their five-year follow up.

Thirteen of the 23 participants with baseline cognitive impairment received a clinical assessment of AD dementia. All 13 participants were Aβ PET positive (Table 6). Twelve of these participants had tau-positive visual interpretation; the remaining participant (Case 18) was tau PET negative, but at their one-year follow up had their clinical assessment changed to FTD. Additionally, Case 21 was tau PET positive, but was reassessed to be cognitively normal at their one- and two-year follow ups.

Conclusions

¹⁸F-flortaucipir PET visual interpretation was found to be consistent between readers in this study (n = 189/189, 100%, κ= 1) and highly consistent with SUVR quantification (n = 186/189, 98.4%, κ= 0.910), suggesting these two approaches to determining tau PET positivity give similar results despite their different methodologies. However, three participants had discordant visual interpretations and SUVRs, likely due to a hemorrhagic infarct with elevated radiotracer uptake, an atypical, occipital-predominant presentation of AD NFT deposition, and a highly lateralized presentation of AD NFT deposition, respectively. These cases suggest the need for MR imaging to accompany ¹⁸F-flortaucipir PET visual interpretation, and the need to consider regions outside the temporal meta-ROI for SUVR quantification.

Among non-AD sources of ¹⁸F-flortaucipir uptake, the most studied is off-target binding in the choroid plexus, striatum, brainstem, and bone/meninges [24, 25]. In this study, off-target binding did not mimic the appearance of the AD tau pattern when assessed by visual readers, nor did it cause any tau PET temporal meta-ROI SUVR to be falsely positive when compared to visual interpretation. However, we observed two other sources of off-target binding that were not mentioned in the manufacturer’s guidelines for ¹⁸F-flortaucipir PET visual interpretation and which can potentially confound tau PET interpretations: hemorrhagic infarcts and meningiomas. The hemorrhagic infarct case was the case previously described as having a tau-positive visual interpretation and a tau-negative SUVR quantification. The two meningioma cases demonstrated elevated levels of radiotracer uptake in the frontal lobe, which is immaterial when assessing tau PET positivity by visual interpretation, but meningiomas in the posterolateral temporal, occipital, or parietal/precuneus regions might plausibly interfere with visual interpretation and SUVR quantification.

¹⁸F-flortaucipir PET visual interpretation was found to be moderately consistent with CSF p-tau181 (n = 124/144, 86.1%, κ= 0.526). Most discordant cases (n = 18/20) are amyloid-positive and CSF p-tau181 positive, but tau-negative on visual interpretation. This suggests that the discordance between ¹⁸F-flortaucipir PET and CSF p-tau181 may be attributed to participants with early changes in AD pathophysiology. Moloney and colleagues have found in an autopsy study that p-tau181, 205, 217, and 231 fluid biomarker sites are present in the early stages of NFT maturity [26]. Wennström and colleagues have found that p-tau217 can be found within NFTs and neuropil threads along with p-tau181, 231, 202, 202/205, and 369/404, and that p-tau217 area fraction correlated with antemortem plasma p-tau217 in individuals with confirmed Aβ plaque pathology [27]. Furthermore, plasma p-tau and CSF p-tau have been shown to be strongly correlated [28–30]. Taken together, these findings suggest a possibility that CSF p-tau181 is tracking changes in AD pathophysiology that occur earlier than the more advanced stages of NFT aggregation that ¹⁸F-flortaucipir PET may be more sensitive to.

When interpreting tau PET visual interpretation alongside clinical diagnosis after the study (both visual interpretation and clinical diagnosis were performed independently) a few relationships between the two kinds of AD diagnoses were remarkable. First, a baseline tau-positive visual interpretation in participants who were cognitively normal at baseline did not reliably predict conversion to AD dementia at follow up. If anything, tau PET positivity in cognitively normal participants was more likely to be either a sign of atypical AD, of related dementias (vascular dementia or FTD), or of resilience to AD dementia. Second, a baseline tau-negative visual interpretation in participants who were cognitively normal at baseline did not rule out conversion to AD dementia at follow up. Four cases were found to demonstrate conversion to AD dementia at follow up under these circumstances, although two of these were later reassessed to be cognitively normal. Third, baseline tau PET positivity in cognitively impaired participants did not guarantee a diagnosis of AD dementia at follow up: one participant was assessed to be cognitively normal at follow up even under these circumstances and another was reassessed to have FTD. Finally, baseline tau PET negativity in cognitively impaired participants cannot be used to rule out conversion to AD dementia at follow up: three such participants converted to AD dementia at their follow up visits, respectively, although one was reassessed to be cognitively normal at a later date.

A bias of the current study lies in the inclusion of cognitively normal participants. In a clinical setting, ¹⁸F-flortaucipir PET is indicated for use in patients with cognitive impairment. Two of the three cases discordant between visual interpretation and SUVR quantification in this study were from cognitively normal participants and would not warrant the use of ¹⁸F-flortaucipir visual interpretation in a clinical setting to begin with. Six of the 20 cases discordant between visual interpretation and CSF p-tau181 quantification were from cognitively normal participants and also would not warrant the use of ¹⁸F-flortaucipir visual interpretation in a clinical setting. Furthermore, the inclusion of cognitively normal participants, who represent the majority of the cases studied, also introduces a negative case bias, as they also represent a majority of the tau PET negative cases. Since most of these cases have tau PET SUVRs much lower than the SUVR positivity threshold, the agreement between visual interpretation and SUVRs is higher in the current study compared to a more challenging study comparing exclusively borderline cases. Nonetheless, exploring tau positivity in cognitively normal participants in this study identified individuals who have atypical AD tau and clinical progression.

The current study is focused on ¹⁸F-flortaucipir PET, and its conclusions do not necessarily apply to other tau PET radiotracers, which may have their own distinctive characteristics regarding off-target binding and sensitivity, which need to be accounted for on a tracer-by-tracer basis in studies of visual interpretation guidelines [31].

Future studies may also explore the discordances between tau PET visual interpretation and tau PET SUVR more thoroughly by investigating those participants with questionable or very mild dementia. Future studies may also explore the consequences of using regions of interest beyond the temporal meta-ROI used in the current study, such as the MUBADA [32]. These alternative ROIs may be appropriate for increasing concordance between visual interpretation and SUVR quantification, especially if crucial differences between the two lie in elevated radiotracer uptake outside the temporal meta-ROI. That said, the most direct way of harmonizing visual interpretation and SUVR quantification may be to construct an entirely new ROI for SUVR quantification that covers the critical regions in visual interpretation (posterolateral temporal, occipital, and parietal/precuneus regions) and develop a quantification method sensitive to contiguous uptake within this ROI.

In conclusion, ¹⁸F-flortaucipir PET visual interpretation can identify atypical AD NFT deposition that may be missed by SUVR quantification depending upon the regions of interest used. However, while the manufacturer’s guidelines for ¹⁸F-flortaucipir PET visual interpretation address non-AD sources of uptake such as off-target binding, they do not address other non-AD sources of uptake such as hemorrhagic infarcts and meningiomas. Temporal meta-ROI SUVR was highly concordant with visual interpretation for the cohort considered in this study. However, SUVR analyses could not detect lateralized occipital-predominant AD NFT deposition because the occipital lobe falls outside the temporal meta-ROI. CSF p-tau181 concentration was moderately concordant with visual interpretation and enabled detection of early changes in AD pathophysiology associated with tau hyperphosphorylation. However, these changes cannot be seen on PET. Finally, a positive visual interpretation did not make a follow up diagnosis of AD dementia inevitable, and a negative visual interpretation did not exclude the possibility of a follow up diagnosis of AD dementia. Additional work is needed to understand how multiple AD PET and CSF biomarkers might conceivably be used in tandem in a clinical setting alongside AD clinical evaluation in order to correctly diagnose and treat all individuals, not just those who demonstrate AD biomarker and clinical findings concordant with group-level trends.