Tracer-specific reference tissues selection improves detection of 18 F-FDG, 18 F-florbetapir, and 18 F-flortaucipir PET SUVR changes in Alzheimer's disease.

This study sought to identify a reference tissue-based quantification approach for improving the statistical power in detecting changes in brain glucose metabolism, amyloid, and tau deposition in Alzheimer's disease studies. A total of 794, 906, and 903 scans were included for 18 F-FDG, 18 F-florbetapir, and 18 F-flortaucipir, respectively. Positron emission tomography (PET) and T1-weighted images of participants were collected from the Alzheimer's disease Neuroimaging Initiative database, followed by partial volume correction. The standardized uptake value ratios (SUVRs) calculated from the cerebellum gray matter, centrum semiovale, and pons were evaluated at both region of interest (ROI) and voxelwise levels. The statistical power of reference tissues in detecting longitudinal SUVR changes was assessed via paired t-test. In cross-sectional analysis, the impact of reference tissue-based SUVR differences between cognitively normal and cognitively impaired groups was evaluated by effect sizes Cohen's d and two sample t-test adjusted by age, sex, and education levels. The average ROI t values of pons were 86.62 and 38.40% higher than that of centrum semiovale and cerebellum gray matter in detecting glucose metabolism decreases, while the centrum semiovale reference tissue-based SUVR provided higher t values for the detection of amyloid and tau deposition increases. The three reference tissues generated comparable d images for 18 F-FDG, 18 F-florbetapir, and 18 F-flortaucipir and comparable t maps for 18 F-florbetapir and 18 F-flortaucipir, but pons-based t map showed superior performance in 18 F-FDG. In conclusion, the tracer-specific reference tissue improved the detection of 18 F-FDG, 18 F-florbetapir, and 18 F-flortaucipir PET SUVR changes, which helps the early diagnosis, monitoring of disease progression, and therapeutic response in Alzheimer's disease.

annual change rate

Alzheimer's disease

Alzheimer's Disease Neuroimaging Initiative

amyloid, tau, and neurodegeneration

amyloid‐β

clinical dementia rating

cognitively impaired

cognitively normal

gray matter

mild cognitively impaired

Mini‐Mental Status Examination

Montreal Neurologic Institute

partial volume correction

region of interest

standardized uptake value ratio

INTRODUCTION

Alzheimer's disease (AD) is a progressive neurodegenerative disease associated with memory deficits and cognitive impairments, brain deposition of amyloid‐β (Aβ) peptide plaques, neurofibrillary tangles composed of hyperphosphorylated tau (ptau) protein, and glucose hypometabolism (DeTure & Dickson, 2019; Serrano‐Pozo, Frosch, Masliah, & Hyman, 2011; Uddin et al., 2018). The standardized assessment of pathological processes underlying AD can be accomplished by biomarker‐evidenced amyloid, tau, and neurodegeneration (ATN) framework (Jack Jr et al., 2018). Positron emission tomography (PET) using radiolabeled ligands including 18F‐FDG, 18F‐florbetapir, and 18F‐flortaucipir has been widely used to assess neurodegeneration, deposition of Aβ fibrils, and tau for diagnosis and monitoring progression of AD. Standardized uptake value ratio (SUVR) relative to a reference tissue is commonly used for ATN PET quantification. Specifically, 18F‐FDG SUVR can be used to estimate the metabolic glucose uptake rate ratio (Y. Wu et al., 2012; Y. G. Wu, 2008), while the 18F‐florbetapir and 18F‐flortaucipir SUVRs can be used to approximate the tracer distribution volume ratio (DVR) of binding to Aβ and tau, respectively (Wong et al., 2010; Zhou et al., 2021; Zhou, Endres, Brašić, Huang, & Wong, 2003; Zhou, Sojkova, Resnick, & Wong, 2012).

Various reference tissues‐based SUVRs have been used in previous AD studies, leading to different statistical power in PET assessments (Chen et al., 2015; Zhou et al., 2021). The pons reference region previously demonstrated the best preservation of glucose metabolism in AD and therefore was deemed as a reliable reference tissue for brain 18F‐FDG PET normalization (Minoshima, Frey, Foster, & Kuhl, 1995). Reference tissues including whole brain (Nugent et al., 2020), cerebellum gray matter (GM; Förster et al., 2012; Ossenkoppele et al., 2012), and pons (Alexander, Chen, Pietrini, Rapoport, & Reiman, 2002; Ortner et al., 2019; Schmidt et al., 2008) have been used to calculate 18F‐FDG PET SUVR in longitudinal AD studies. The effect of reference tissues including cerebellum GM, pons, and whole brain on 18F‐FDG PET SUVR have been also evaluated in cross‐sectional (Minoshima et al., 1995; Yakushev et al., 2008) and longitudinal AD studies (Nugent et al., 2020; Verger, Doyen, Campion, & Guedj, 2021). Similarly, different reference tissues including cerebellar GM, centrum semiovale, pons, and corpus callosum have been used to quantify 18F‐florbetapir and 11C‐PIB amyloid PET (Blautzik et al., 2017; Chen et al., 2015; Chiao et al., 2019; Heeman et al., 2020; Shokouhi et al., 2016; Su et al., 2015; Wang et al., 2021; Xie et al., 2020) and 18F‐flortaucipir tau PET (Baker et al., 2017; Cho et al., 2020; Devous Sr. et al., 2018; Southekal et al., 2018; Zhao, Liu, Ha, Zhou, & Alzheimer's Disease Neuroimaging Initiative, 2019). The amyloid PET SUVRs calculated from different reference tissues were compared and evaluated by correlation analysis of SUVR versus cognitive assessment (Chen et al., 2015), test–retest analysis (Blautzik et al., 2017), and effect size for evaluation of the treatment response (Chiao et al., 2019). Our previous research has demonstrated that spatially constrained kinetic model with dual reference tissues comprising of cerebellum GM and centrum semiovale significantly improves quantification of relative perfusion and tau binding (Zhou et al., 2021). In previous longitudinal 18F‐FDG and 18F‐florbetapir PET studies, different reference tissues based SUVRs were compared, but the comparisons were limited to the region of interest (ROI) levels. Also, these previous studies focused 18F‐FDG PET in normal aging (Nugent et al., 2020; Verger et al., 2021) or amyloid treatment effects in mild cognitive impairment (MCI) and AD participants (Chen et al., 2015; Chiao et al., 2019). Moreover, for tau PET studies, to the best of our knowledge, there has been no evaluation for multiple reference tissues in longitudinal studies.

The selection of an appropriate reference tissue is reliant on multivariable factors and imperative aspects such as the studied population, study sample size, PET acquisition protocol, and the type of radiopharmaceutical used. The objective of this study is to improve statistical power for detecting 18F‐FDG, 18F‐florbetapir, and 18F‐flortaucipir PET SUVR changes in AD by selecting appropriate reference tissues. Using the AD Neuroimaging Initiative (ADNI), we performed longitudinal analysis on individuals with disease progression regardless of their disease stage, whether from cognitively normal (CN) to AD, MCI to AD, or CN to MCI. The impact of the reference tissue selection on discriminating between CN and cognitively impaired (CI) were also evaluated. This study is the most comprehensive comparative analysis of different reference tissues measured with multiple radiotracers to monitor the progression of AD. Our study may improve clinical staging diagnosis through quantitative PET which has important implications for biomarker‐guided precision medicine.

MATERIALS AND METHODS

Participants

All 18F‐FDG, 18F‐florbetapir, and 18F‐flortaucipir PET and structural MRI data in this study were obtained from the AD Neuroimaging Initiative (ADNI) dataset (https://adni.loni.usc.edu). Informed written consent was obtained from all participants at each site. In total, we downloaded 794 18F‐FDG‐PET scans, 906 18F‐florbetapir‐PET scans, and 903 18F‐flortaucipir‐PET scans, which encompass 420, 434, and 666 participants, respectively. Demographics and clinical assessments including the Mini‐Mental Status Examination (MMSE), Clinical Dementia Rating (CDR), and clinical diagnostic status of CN, MCI, and AD participants were also obtained.

PET data acquisition and image preprocessing

Raw T1‐weighted structural MRI and preprocessed 18F‐FDG, 18F‐florbetapir, and 18F‐flortaucipir PET images of each subject were downloaded from the ADNI database. The downloaded PET images were aligned, averaged, reoriented, and interpolated into a standard 160 × 160 × 96 voxel image grid and smoothed with an 8 mm in full width at half maximum (FWHM) 3D Gaussian filter by the ADNI consortium with 1.5 mm cubic voxels. Further details of PET and T1‐weighted MR acquisition protocols can be found at http://adni.loni.usc.edu/methods/pet-analysis-method/pet-analysis/ and http://adni.loni.usc.edu/methods/documents/mri-protocols/, respectively.

The PET and MRI data were further processed by partial volume correction (PVC) and spatial normalization, both using Statistical Parametric Mapping (SPM12, Wellcome Department of Imaging Neuroscience, Institute of Neurology, London, UK) in the MATLAB R2020b (The MathWorks Inc., Natick, MA) environment, as reported in our earlier studies (Paranjpe et al., 2019; Yan et al., 2020, 2021). PVC was performed to minimize the possibility of underestimation in PET images, especially for small brain regions such as amygdala and striatum. The reblurred Van Cittert iteration method was applied for PVC in individual PET images, where a 3D Gaussian Kernel of 8 mm FWHM was used as the spatial smoothing function with step length α of 1.5 (Tohka & Reilhac, 2008). All PET images were then coregistered to the individual's own structural MRI images, which were normalized to the standard Montreal Neurologic Institute (MNI) space using an MRI template (image volume: 121 × 145 × 121, voxel size: 1.5 × 1.5 × 1.5 mm in x, y, z). The median (interquartile range) of time intervals between PET and MRI are 28(49) days, 30(48) days, and 51(129.5) days for 18F‐FDG, 18F‐florbetapir, and 18F‐flortaucipir, respectively. The transformation parameters determined by MRI spatial normalization were then applied to the coregistered PET images for PET spatial normalization. SUVR images were calculated relative to the cerebellum GM (SUVRCereb_GM), centrum semiovale (SUVRCS), and pons (SUVRPons) reference tissues. The ROI SUVR values were calculated by applying ROIs on the SUVR images in the standard MNI space for minimizing variance related to the variability of ROI volume and shape in native space (Gottesman et al., 2017; Liu et al., 2019; Paranjpe et al., 2019; Tudorascu et al., 2018; Yan et al., 2021). A total of 18 ROIs including three reference tissues (cerebellum GM, centrum semiovale, and pons) and an additional 15 ROIs including the orbital frontal, prefrontal, superior frontal, medial temporal, inferior temporal, lateral temporal, parietal, posterior precuneus, anterior cingulate, posterior cingulate, occipital, entorhinal cortex, amygdala, hippocampus, and parahippocampal gyrus regions were manually delineated on the MRI template using the PMOD software program (PMOD 4.002, PMOD Technologies Ltd., Zürich, Switzerland) in standard MNI space. These ROI templates were previously developed in the Johns Hopkins Department of Radiology and have been validated in our former studies (Liu et al., 2019; Paranjpe et al., 2019; Yan et al., 2020, 2021; Zhou et al., 2021).

Longitudinal SUVR PET analysis for cognitively declined participants

The effects of different reference tissues were evaluated on the sensitivity of SUVR measurements for the detection of cognitively declined populations. The baseline and last scans were defined as the subject's first and last scan in the downloaded data. Participants at the last scan who had an increased CDR score (Morris, 1993) or evidence of clinical disease progression (CN to MCI, MCI to AD, or CN to AD) were included in the longitudinal studies. This population inclusion criterion was consistent across 18F‐FDG, 18F‐florbetapir, and 18F‐flortaucipir studies. Based on the baseline and last scan SUVR values of each subject, paired statistical t values were calculated at both ROI‐ and voxel‐levels for each reference tissue. The annual change rates for 18F‐FDG, 18F‐florbetapir, and 18F‐flortaucipir uptake were further calculated as follows: where x represents the reference tissue (cerebellum GM, centrum semiovale, or pons), and are the SUVR values of the ROI at the last scan and baseline, T is the time interval from baseline to the last scan in years.

Cross‐sectional SUVR PET analysis for CN and CI participants

To comprehensively assess the performance of the cerebellum GM, centrum semiovale, and pons reference tissues in discriminating between CN or CI individuals, ROI‐ and voxelwise‐based cross‐sectional statistical analyses were performed. All baseline scans were included for cross‐sectional analysis and participants were classified as either CN (CDR = 0) or CI (CDR ≥ 0.5; Zhou et al., 2021). To investigate the sensitivity of the PET SUVR measurement in discriminating CN from CI, effects sizes were approximated using Cohen's d (Chand et al., 2020; Cohen, 1988; Lopresti et al., 2005; Sullivan & Feinn, 2012; Zhou et al., 2021) of CN and CI groups as follows: where represents the standard deviation of SUVR in pooled population, represents either cerebellum GM, centrum semiovale, or pons reference tissues. Since the 18F‐FDG SUVR decreased while the 18F‐florbetapir and 18F‐flortaucipir SUVR increased with disease progression, we set group 1 to be CN and group 2 to be CI for 18F‐FDG, and group1 to be CI and group2 to be CN for 18F‐florbetapir and 18F‐flortaucipir.

For correcting the influence of covariates, the two‐sample independent t test adjusted by age, sex, and education levels were performed at voxelwise level using SPM12. For ROI‐based analysis, the generalized linear model was used to assess the group difference in SUVR for each ROI by adjusting for covariates, the Bonferroni‐corrected p‐value < .05 was defined as significant.

RESULTS

Study cohort characteristics

Study cohort characteristics for participants in the longitudinal analyses are summarized in Table 1. A total of 53, 55, and 20 participants were included with a mean time interval between the baseline and last scan of 63.42 ± 27.15 months, 57.05 ± 18.75 months, and 19.88 ± 8.03 months for 18F‐FDG, 18F‐florbetapir, and 18F‐flortaucipir, respectively. For each tracer, there were substantial differences between the baseline and last scan in age, education level, and CDR. There was no significant difference in the baseline and last scan for the MMSE of the participants in 18F‐flortaucipir. Comparison across tracers demonstrated that MMSE values at the baseline showed differences between 18F‐FDG and 18F‐flortaucipir groups and between 18F‐florbetapir and 18F‐flortaucipir groups. Other groups showed no significant difference in age and CDR in all three tracers.

TABLE 1

Characteristics of longitudinal study cohort

		Mean ± SD			p value
		18F‐FDG	18F‐florbetapir	18F‐flortaucipir	18F‐FDG versus 18F‐florbetapir	18F‐FDG versus 18F‐flortaucipir	18F‐florbetapir versus 18F‐flortaucipir
Participant (CN/MCI/AD)		29/22/2	25/29/1	9/8/3
Scans		106	110	40
Sex (M/F)		24/29	25/30	9/11
Education (years)		15.40 ± 2.66	15.40 ± 2.65	15.50 ± 2.56
Age (years)	Baseline	74.53 ± 5.65	73.98 ± 6.90	77.17 ± 7.89	.65	.12	.09
	Last scan	79.82 ± 5.94†	78.73 ± 7.26†	78.83 ± 7.98†	.40	.56	.96
MMSE	Baseline	28.13 ± 1.83	27.89 ± 1.74	26.45 ± 3.39	.48	.01**	.02*
	Last scan	25.08 ± 4.51†	24.65 ± 5.25†	25.25 ± 4.63	.66	.15	.66
CDR	Baseline	0.21 ± 0.25	0.26 ± 0.25	0.35 ± 0.37	.25	.06	.25
	Last scan	0.67 ± 0.48†	0.78 ± 0.51†	0.70 ± 0.52†	.24	0.82	0.54

Note: Significant differences between the baseline and last scan values using two‐sample t‐test are denoted as †. Significant differences between the 18F‐FDG and 18F‐florbetapir group, 18F‐FDG and 18F‐flortaucipir group, 18F‐florbetapir, and 18F‐flortaucipir group are denoted by *, **, and *** when p < .05, .01, and 0.001, respectively.

Study cohort characteristics for participants in the cross‐sectional study are listed in Table 2. There were no significant differences between CN and CI groups in terms of their age and education level, but considerable differences were observed for the MMSE and CDR scores for 18F‐FDG and 18F‐florbetapir PET studies. Significant differences were detected in age, education level, MMSE, and CDR scores in 18F‐flortaucipir PET study.

TABLE 2

Characteristics of cross‐sectional study cohort

Characteristics	18F‐FDG			18F‐florbetapir			18F‐flortaucipir
	CN	CI (MCI/AD)	p value	CN	CI (MCI/AD)	p value	CN	CI (MCI/AD)	p value
Participants (n)	168	252(220/32)		257	177(162/15)		395	271(205/66)
Sex (M/F)	74/94	151/101		104/153	106/71		104/153	162/109
Age (years)	72.79 ± 6.14	71.76 ± 7.89	.16	72.28 ± 6.41	71.77 ± 7.71	.45	73.31 ± 7.08	75.78 ± 8.37	<.0001***
Education (years)	16.57 ± 2.61	16.24 ± 2.66	.21	16.63 ± 2.52	16.33 ± 2.66	.23	16.84 ± 2.32	16.08 ± 2.68	.0001**
MMSE score	29.13 ± 1.14	27.43 ± 2.58	<.0001***	29.13 ± 1.06	27.53 ± 2.76	<.0001***	29.08 ± 1.18	26.48 ± 3.70	<.0001***
CDR	0.00 ± 0.04	0.52 ± 0.14	<.0001***	0.00 ± 0.04	0.52 ± 0.13	<.0001***	0.02 ± 0.12	0.56 ± 0.35	<.0001***

Note: Significant differences between CN and CI groups are denoted by *, **, and *** indicating p < .05, .01, and .001, respectively.

Effect of reference tissue selection on sensitivity to detect longitudinal PET SUVR changes in AD

18F‐FDG PET

The statistic t maps based on the 18F‐FDG SUVR images for three reference tissues are illustrated in Figure 1. It was evident that the t‐values calculated from SUVR images were in the order of t(SUVRPons) > t(SUVRCereb_GM) > t(SUVRCS) in the frontal, temporal and parietal regions. ROI‐based analysis showed consistent results as demonstrated in Figure 2. The t(SUVRPons) showed the greatest sensitivity in the orbital frontal, prefrontal, superior frontal, lateral temporal, inferior temporal, posterior precuneus, anterior cingulate, posterior cingulate, caudate, entorhinal cortex, amygdala, hippocampus, and parahippocampal gyrus, 86.62 ± 47.63% higher than the t(SUVRCS) and 38.40 ± 29.13% higher than the t(SUVRCereb_GM). In contrast, SUVRCS had the lowest sensitivity and failed to detect SUVR decrease in the orbital frontal, prefrontal, and amygdala (t < 1.67). Compared with the SUVRCS, the SUVRCereb_GM detected changes in the orbital frontal and prefrontal. The ACRs of ROIs for each reference tissue are listed in Table 3. The pons demonstrated the largest ACR across all listed brain regions, with an average of 73.8 ± 42.4% and 92.2 ± 61.7% higher than the ACRCereb_GM and ACRCS, respectively. Regardless of which reference tissue was used, the caudate, anterior cingulate, lateral temporal, parahippocampal gyrus, posterior cingulate, parietal, and entorhinal cortex showed the greatest longitudinal annual change rates of 18F‐FDG uptake with each having at least a 1% reduction (Table 3).

FIGURE 1

Paired statistical t map of longitudinal (mean follow‐up period: 63.42 ± 27.15 months, n = 53) 18F‐FDG SUVR changes in participants with cognitive decline. The SUVR was calculated for reference tissue cerebellum GM, centrum semiovale, and pons, respectively

FIGURE 2

Statistical t values of ROI‐based longitudinal 18F‐FDG SUVR changes in subjects with cognitive decline. Statistical p values indicate *p < .05, **p < .001, ***p < .0001. ACC, anterior cingulate; Amy, amygdala; Caud, Caudate; EC, entorhinal cortex; Hip, hippocampus; InfT, inferior temporal; LatT, lateral temporal; MT, medial temporal; OrbF, orbital frontal; Par, parietal; PCC, posterior cingulate; PHip, parahippocampal gyrus; PPrC, posterior precuneus; PreF, prefrontal; SupF, superior frontal

TABLE 3

Comparison of the annual change rates (%) of 18F‐FDG, 18F‐florbetapir, and 18F‐flortaucipir SUVR

		OrbF	PreF	SupF	LatT	InfT	MT	Par	PPrC	ACC	PCC	Caud	EC	Amy	Hip	PHiP	Occ
18F‐FDG	Cereb GM	−0.4 ± 2.8	−0.3 ± 2.6	−0.5 ± 2.6	−1.2 ± 2.6	−0.8 ± 2.4	−0.6 ± 2.4	−1.1 ± 2.6	−0.8 ± 2.7	−1.2 ± 2.9	−1.1 ± 2.6	−1.5 ± 3.0	−1.0 ± 3.2	−0.3 ± 2.9	−0.7 ± 2.4	−1.1 ± 3.0
	CS	−0.3 ± 3.3	−0.3 ± 3.5	−0.5 ± 3.3	−1.1 ± 3.3	−0.8 ± 2.9	−0.6 ± 2.8	−1.0 ± 3.0	−0.8 ± 3.3	−1.2 ± 2.6	−1.0 ± 3.2	−1.4 ± 3.4	−1.0 ± 2.9	−0.3 ± 2.9	−0.7 ± 2.6	−1.1 ± 2.4
	Pons	−0.9 ± 2.2	−0.8 ± 2.2	−1.0 ± 2.4	−1.7 ± 2.5	−1.3 ± 2.2	−1.1 ± 2.1	−1.5 ± 2.5	−1.3 ± 2.7	−1.7 ± 1.7	−1.6 ± 2.3	−2.0 ± 2.2	−1.5 ± 3.0	−0.8 ± 2.5	−1.2 ± 2.3	−1.6 ± 2.4
18F‐florbetapir	Cereb GM	1.3 ± 2.7	1.4 ± 2.4	1.3 ± 2.5	1.0 ± 2.0	0.8 ± 2.2		1.2 ± 2.2	1.3 ± 2.1	1.0 ± 3.0	1.8 ± 2.5						1.1 ± 2.3
	CS	1.8 ± 2.9	1.9 ± 2.4	1.8 ± 2.3	1.5 ± 2.2	1.3 ± 2.1		1.7 ± 2.1	1.8 ± 2.1	1.6 ± 3.6	2.4 ± 2.9						1.5 ± 1.9
	Pons	1.6 ± 2.7	1.7 ± 2.4	1.6 ± 2.4	1.4 ± 2.2	1.2 ± 2.2		1.5 ± 2.3	1.6 ± 2.2	1.4 ± 3.6	2.2 ± 2.9						1.4 ± 2.2
18F‐flortaucipir	Cereb GM			0.5 ± 3.1	0.5 ± 2.7	0.5 ± 3.1
	CS			3.3 ± 6.4	3.2 ± 5.9	3.2 ± 5.7
	Pons			0.8 ± 5.7	0.8 ± 4.8	0.7 ± 4.9

Note: The values demonstrate ACRs of SUVRs in 18F‐FDG, 18F‐florbetapir and 18F‐flortaucipir when using the cerebellum GM, centrum semiovale, and pons as reference tissues.

Abbreviations: ACC, anterior cingulate; Amy, amygdala; Caud, Caudate; EC, entorhinal cortex; Hip, hippocampus; InfT, inferior temporal; LatT, lateral temporal; MT, medial temporal; Occ, Occipital; OrbF, orbital frontal; Par, parietal; PCC, posterior cingulate; PHip, parahippocampal gyrus; PPrC, posterior precuneus; PreF, prefrontal; SupF, superior frontal.

‐florbetapir PET

In contrast to the 18F‐FDG PET analysis where the SUVRPons demonstrated the greatest sensitivity, the SUVRCS in 18F‐florbetapir was superior in detecting longitudinal amyloid depositions (Figure 3). As demonstrated in Figure 4, significant longitudinal differences in the orbital frontal, prefrontal, superior frontal, lateral temporal, inferior temporal, parietal, posterior precuneus, occipital, anterior cingulate, and posterior cingulate were detected using the centrum semiovale reference tissue. The t(SUVRCS) was 27.12 ± 12.31% and 52.12 ± 23.78% higher than the t(SUVRCereb_GM) and t(SUVRPons), respectively, which was consistent with the t map at the voxel‐level. The cerebellum GM and pons demonstrated similar sensitivities in monitoring all regions, however, the use of these reference tissues was unable to detect changes in the anterior cingulate. The greatest amyloid ACR for disease‐progressed participants was observed in the posterior cingulate (2.2%) when centrum semiovale was used as reference tissue (Table 3). The average of ACRs in all the studied brain regions was 1.2% using a cerebellum GM reference tissue (ACRCereb_GM), 1.7% using a centrum semiovale reference tissue (ACRCS), and 1.6% using the pons reference (ACRPons) tissue (Table 3). ACRCS was 10.2 ± 1.8% and 44.3 ± 10.1% greater than the ACRCereb_GM and ACRPons, respectively, highlighting the superiority of the centrum semiovale in detecting longitudinal amyloid changes.

FIGURE 3

Paired statistical t map of longitudinal (mean follow‐up period: 57.05 ± 18.75 months, n = 55) 18F‐florbetapir SUVR changes in participants with cognitive decline. The SUVR was calculated for reference tissue cerebellum GM, centrum semiovale, and pons, respectively

FIGURE 4

Statistical t values of ROI‐based longitudinal 18F‐florbetapir SUVR changes in subjects with cognitive decline. Statistical p values indicate *p < .05, **p < .001, ***p < .0001. ACC, anterior cingulate; InfT, inferior temporal; LatT, lateral temporal; Occ, occipital; OrbF, orbital frontal; Par, parietal; PCC, posterior cingulate; PPrC, posterior precuneus; PreF, prefrontal; SupF, superior frontal

‐flortaucipir PET

The ROI and voxelwise results for 18F‐flortaucipir are displayed in Figures 5 and 6, respectively. Centrum semiovale reference tissue could only identify longitudinal changes in known AD‐sensitive regions including the superior frontal, lateral temporal, and inferior temporal. The t(SUVRCS) ranged from 2.02 to 2.34, which was 8.46 and 15.42 times higher than the t(SUVRCereb_GM) and t(SUVRPons), respectively (Figure 6). The ACRs of these significant brain regions using three reference tissues are also listed in Table 3. The ACRs of the superior frontal, lateral temporal, and inferior temporal demonstrated an average of 3.2% when using the centrum semiovale as the reference tissue, exceeding ACRCereb_GM (0.5%) and ACRPons (0.8%) by 5.5 and 3.2 times, respectively.

FIGURE 5

Paired statistical t map of longitudinal (mean follow‐up period: 19.88 ± 8.03 months, n = 20) 18F‐flortaucipir SUVR changes in participants with cognitive decline. The SUVR was calculated for reference tissue cerebellum GM, centrum semiovale, and pons, respectively

FIGURE 6

Statistical t values of ROI‐based longitudinal 18F‐flortaucipir SUVR changes in subjects with cognitive decline. Statistical p values indicate *p < .05, **p < .001, ***p < .0001

Effect of reference tissue selection on sensitivity to detect PET SUVR differences between CN and CI groups in cross‐sectional study cohort

The Cohen's d maps (Figure 7) generated from the cerebellum GM, centrum semiovale, and pons reference tissue‐based SUVR images were visually comparable. The average d‐values over 15 ROIs for the cerebellum GM, centrum semiovale and pons in 18F‐FDG were d(SUVRCereb_GM) = 0.33 ± 0.89, d(SUVRCS) = 0.30 ± 0.15 and d(SUVRPons) = 0.38 ± 0.35; in 18F‐florbetapir were d(SUVRCereb_GM) = 0.44 ± 0.16, d(SUVRCS) = 0.46 ± 0.14, and d(SUVRPons) = 0.49 ± 0.13; in 18F‐flortaucipir were d(SUVRCereb_GM) = 0.60 ± 0.20, d(SUVRCS) = 0.56 ± 0.23, and d(SUVRPons) = 0.59 ± 0.21. The covariates adjusted t maps (Figure 8) generated from reference tissue‐based SUVR images were comparable in 18F‐florbetapir and 18F‐flortaucipir, but pons had increased performance in 18F‐FDG. The average t values from ROIs for the cerebellum GM, centrum semiovale, and pons in 18F‐FDG were t(SUVRCereb_GM) = 3.52 ± 1.09, t(SUVRCS) = 3.37 ± 0.96, and t(SUVRPons) = 4.76 ± 0.91; in 18F‐florbetapir were t(SUVRCereb_GM) = 4.54 ± 1.67, t(SUVRCS) = 4.75 ± 1.43, and t(SUVRPons) = 5.51 ± 1.20; in 18F‐flortaucipir were t(SUVRCereb_GM) = 8.39 ± 2.58, t(SUVRCS) = 7.51 ± 2.90, and t(SUVRPons) = 7.96 ± 2.67.

FIGURE 7

Statistical Cohen's d images for SUVR differences between cognitively normal (CN) and impaired (CI) groups in 18F‐FDG (a), 18F‐florbetapir (b), and 18F‐flortaucipir (c) PET studies when using cerebellum GM, centrum semiovale, and pons as reference tissue. The SUVR was calculated for reference tissue cerebellum GM, centrum semiovale, and pons, respectively

FIGURE 8

Statistical t images for SUVR differences between cognitively normal (CN) and impaired (CI) groups in 18F‐FDG (a), 18F‐florbetapir (b), and 18F‐flortaucipir (c) PET studies when using cerebellum GM, centrum semiovale, and pons as reference tissue. The SUVR was calculated for reference tissue cerebellum GM, centrum semiovale, and pons, respectively

DISCUSSION

In this study, we evaluated the effect of reference tissue selection on the statistical power to detect longitudinal and cross‐sectional SUVR changes in 18F‐FDG, 18F‐florbetapir, and 18F‐flortaucipir PET studies of AD. Specifically, the results from 18F‐FDG were consistent with previous reference tissue selection studies in which the pons showed superiority in both longitudinal and cross‐sectional analyses, especially in elucidating longitudinal SUVR changes (Minoshima et al., 1995; Nugent et al., 2020; Verger et al., 2021). After the correction of covariates (age, sex, and education level) effects in cross‐sectional analysis, the pons demonstrated better performance in distinguishing CN and CI groups, particularly in the orbital frontal, prefrontal, and inferior temporal (Table S1). Minoshima et al. (1995) evaluated pons and other reference tissues in AD by observing the preservation of glucose metabolism. They concluded that pons was a reliable and appropriate reference tissue for data normalization to distinguish between CN and CI groups. In the investigation of the metabolic changes of normal aging, pons was also identified as the most appropriate area for brain intensity normalization due to its minimal correlation with age and greater significant longitudinal cluster volumes compared to other normalizations (Verger et al., 2021). Similarly, in the context of healthy aging, pons was verified as the optimal reference tissue by examining changes in brain glucose uptake when comparing with the whole brain based on the posterior cingulate and the precuneus regions (Nugent et al., 2020). Although these previous studies have identified pons as the most appropriate reference tissue, these studies did not conduct voxelwise analysis and considered only a limited number of ROIs. More importantly, the longitudinal population these studies included were either normal aging or patients with only follow‐up scans, regardless of the subjects' disease progression and clinical severity. In our study, we confirmed the use of pons in 18F‐FDG for the population with disease progression according to their clinical diagnosis. The annual reduction rates in 18F‐FDG uptake under normal aging in the anterior cingulate cortex and posterior cingulate cortex/precuneus were previously reported as 0.6 and 0.6% (Ishibashi et al., 2018). By using pons, our result illustrated that the ACRs of the subjects with clinical declines in the anterior cingulate and posterior cingulate were 1.7 and 1.6%, as more than twice as that of the normal participants reported (Ishibashi et al., 2018).

For 18F‐florbetapir amyloid PET, our results suggest that the centrum semiovale is the ideal reference tissue to detect 18F‐florbetapir longitudinal changes in AD for monitoring brain amyloid accumulation. The centrum semiovale showed the greatest statistical power to detect brain regions with highly significant t‐values, as well as the highest ACR (ACRCS = 1.7%) compared to the cerebellum GM and pons. The cerebral white matter has been previously validated as the reference tissue with less variability and great statistical power to detect longitudinal increases in Aβ depositions (Chen et al., 2015). Interestingly, the application of subcortical white matter and cerebellar white matter alone or in combination produced an enhanced result when assessed by effect size (Cohen's d; Chiao et al., 2019). Moreover, in a test–retest 18F‐florbetapir longitudinal SUVR study, the brainstem had the highest stability and correlation between PET and concurrent cerebrospinal fluid Aβ1–42 levels (Shokouhi et al., 2016). It is well‐recognized that partial volume effects exist in brainstem‐ or pons‐based SUVR measurements (Chen et al., 2015; Su et al., 2019). In our cross‐sectional analysis including effect size estimates and two‐independent sample t‐test, cerebellum GM, centrum semiovale, and pons all demonstrated comparable statistical power in distinguishing SUVR differences between CN and CI groups. Cerebellum GM was ultimately recommended as the ideal reference tissue in 18F‐florbetapir PET studies especially in studies involving relative cerebellar perfusion measurement (Bilgel et al., 2020; Hsiao et al., 2013).

Our study also revealed the centrum semiovale as the optimal reference tissue for 18F‐flortaucipir tau PET studies. Superior frontal, lateral temporal, and inferior temporal were the only brain regions that were detected as significant using the centrum semiovale reference in tau PET. Considering the shorter time intervals between tau scans, all three reference tissues were indicated as the sensitive regions in tau PET imaging which strongly correlated with neurodegeneration (Zhou et al., 2021). To the best of our knowledge, this is the first study that evaluated reference tissue effects longitudinally for tau PET. For cross‐sectional analysis, the presented d maps and covariates adjusted t maps showed that the cerebellum GM, centrum semiovale, and pons revealed similar performance in 18F‐flortaucipir, but the cerebellum GM was still the recommended reference tissue due to previous cerebral blood flow studies (Rubinski et al., 2021; Zhou et al., 2021). Our previous research has demonstrated that the centrum semiovale reference tissue‐based DVR and SUVR at the late phase, as well as the cerebellum GM reference tissue‐based R1 and SUVR at the early phase, demonstrated higher Cohen's d effect size to detect tau deposition with improved quantification power for dynamic 18F‐flortaucipir PET quantifications (Zhou et al., 2021). Notice that the cerebellum GM was frequently used as reference tissue in full dynamic 18F‐flortaucipir PET studies (Baker et al., 2017; Barret et al., 2017; Devous Sr. et al., 2018; Golla et al., 2017). The average ACR of tau obtained in this study when centrum semiovale as reference tissue was ACRCS = 3.22%, which was close to the previous annual 18F‐flortaucipir change of 4% collected from the inferior temporal region when cerebral white matter was used as the reference tissue (Hanseeuw et al., 2019).

To date, the cerebellum GM reference tissue, a region without relevant specific binding, has been suggested for the quantitation of amyloid and tau burden (Baker et al., 2017; Barret et al., 2017; Joachim, Morris, & Selkoe, 1989; Price et al., 2005; Yamaguchi, Hirai, Morimatsu, Shoji, & Nakazato, 1989), as well as for the quantification of the tracer perfusion (Bilgel et al., 2020; Weiner et al., 2013; Zhou et al., 2007, 2021). For example, cerebellum GM was the suggested reference tissue to estimate relative transport rate R1 for dynamic 18F‐flortaucipir PET studies (Zhou et al., 2021). Based on the cerebellum GM reference, the R1 images derived from 11C‐PIB dynamic PET have been used to assess the cerebral blood flow decreases in AD studies (Bilgel et al., 2020). Previous studies have also proposed and validated the benefits of using a white matter reference in longitudinal amyloid PET studies (Brendel et al., 2015; Chen et al., 2015). The cerebral white voxels were usually collected in the corpus callosum and centrum semiovale, rather than the cerebellum white matter which is close to gray matter or ventricles due to the likelihood of being confounded by the partial volume effect. Although the whole cerebellum is also used for cross‐sectional analyses of florbetapir PET (Doraiswamy et al., 2012; Fleisher et al., 2013; Landau et al., 2013), a reference region including subcortical white matter is suggested for the measurement of the SUVR changes in longitudinal florbetapir PET study (Landau et al., 2015), which is consistent with our longitudinal study. We have also implemented the longitudinal analysis and covariates corrected cross‐sectional analysis using the whole cerebellum (a combination of gray matter and white matter) reference tissue and compared the results. In the longitudinal analysis, the sensitivity of the whole cerebellum reference tissue to detect 18F‐FDG PET SUVR changes is lower than that of cerebellum GM and is close to that of white matter (centrum semiovale; Figure S1). In 18F‐florbetapir and 18F‐flortaucipir PET studies, the whole cerebellum still has comparable sensitivity to detect SUVR changes with centrum semiovale but has remarkable higher detection power as compared with cerebellum GM reference tissue (Figures S2 and S3). The covariates adjusted t‐test showed that the whole cerebellum reference tissue based SUVR has the least sensitivity to discriminate CN and CI groups in both 18F‐FDG (Figure S4) and 18F‐florbetapir (Figure S4) when compared to the cerebellum GM, centrum semiovale and pons based SUVRs, specifically for the frontal and hippocampus (Table S1) in 18F‐FDG and occipital and parahippocampal gyrus in 18F‐florbetapir (Table S2). Whereas in 18F‐flortaucipir, the cerebellum GM, centrum semiovale, pons, and whole cerebellum demonstrate similar sensitivities in distinguishing CN and CI groups (Table S3 and Figure S4).

In conclusion, the reference tissue pons‐based 18F‐FDG SUVR and centrum semiovale‐based 18F‐florbetapir and 18F‐flortaucipir SUVR significantly improved the detection power of longitudinal PET changes in subjects with cognitive decline. For our cross‐sectional analyses, the pons demonstrated a better performance in distinguishing CN and CI groups in 18F‐FDG. For amyloid and tau PET, the cerebellum GM, centrum semiovale and pons revealed a comparable statistical power to distinguish between CN and CI, but the cerebellum GM was suggested as the ideal reference tissue for quantification of relative cerebral blood flow, as well as amyloid and tau depositions using 18F‐florbetapir and 18F‐flortaucipir. The suggested tracer‐specific reference tissues for SUVR calculation provide a basis for clinical quantitative ATN PET normalization and standardization. Our study supports an improved quantitative PET approach for early diagnosis, monitoring of disease progression, and therapeutic response in AD studies.

CONFLICT OF INTEREST

The authors have declared no conflicts of interest for this article.

Appendix S1: Supporting Information

Click here for additional data file.