Tau PET overlap index correlation with neuropathological findings

Abstract

INTRODUCTION

The tau positron emission tomography (PET) overlap index (OI) has shown promise in maximizing signal‐to‐noise for longitudinal tau PET imaging, particularly for early tau pathology, but requires validation against neuropathology.

METHODS

Fifty‐seven participants who underwent serial tau PET imaging (flortaucipir) and subsequent autopsy were included. Tau PET OI and standardized uptake value ratios (SUVRs) were compared across neuropathological diagnoses.

RESULTS

Tau PET OI showed greater concordance with neurofibrillary tangle (NFT) severity in the entorhinal cortex (a key region for Alzheimer's disease [AD] tauopathy) than SUVR, particularly in early Braak tangle stages (positivity: 52.2% for OI vs. 13.0% for SUVR). OI detected overlapping tau voxels that exhibited spatial correspondence with immunohistochemical and autoradiography measures of tau deposition across both AD and non‐AD tauopathies.

DISCUSSION

These findings demonstrate the enhanced capacity of OI in serial tau PET to robustly detect early and spatially localized tau pathology, supporting its application as a sensitive imaging metric in AD and select non‐AD tauopathies.

Highlights

• Tau positron emission tomography (PET) overlap index (OI) quantifies the spatial consistency of elevated standardized uptake value ratio (SUVR) voxels across two serial tau PET scans and is applicable at this point to serial imaging only.

• Tau PET OI improves detection of tau signal compared to SUVR in the Alzheimer's disease spectrum as well as in various pathologic diagnosis groups (Lewy body dementia, argyrophilic grain disease, corticobasal degeneration, progressive supranuclear palsy, etc.).

• Tau PET OI can detect tau signal in gray matter and sometimes in white matter in the 4R tau pathological diagnosis group.

• Tau PET OI correlated with neuropathological immunohistochemistry (IHC) and autoradiography findings with various pathological diagnosis groups.

• Tau PET OI can capture tau overlapped signal at early Braak tangle stage that visually corresponds to neurofibrillary tangle patterns seen on IHC.

1. BACKGROUND

Tau, a microtubule‐associated protein, functions to stabilize microtubule networks. ¹ , ² Detachment of tau from microtubules, driven by hyperphosphorylation, leads to the formation of neurofibrillary tangles (NFTs), which represent a principal pathological feature of Alzheimer's disease (AD). Other neurodegenerative diseases have other forms of tau accumulation, such as frontotemporal lobar degeneration with tau pathology (FTLD‐tau), progressive supranuclear palsy (PSP), and corticobasal degeneration (CBD). ¹ , ² Post mortem research of tau in the brain is essential for understanding these tauopathies, which remain under investigation in various ways. ³ , ⁴ , ⁵ , ⁶

Tau positron emission tomography (tau PET) enables in vivo quantification of tau deposition in the brain, allowing investigations of tau neuropathology in living participants. ⁷ Studies that compare tau PET with neuropathological findings have advanced our understanding of tauopathies and the progression of tau aggregation in the ante mortem brain. ⁷ , ⁸ , ⁹ Lowe et al. demonstrated that [^{1 8}F]‐flortaucipir PET (FTP PET) uptake correlated with neuropathologic findings, with elevated FTP PET standardized uptake value ratios (SUVRs) observed in cases with Braak tangle stage above IV. ⁷ Pontecorvo et al. compared FTP PET SUVR with post mortem measures of phosphorylated tau and amyloid beta (Aβ) across multiple brain regions in end‐stage dementia cases. ⁸ Cortical FTP PET SUVR showed a correlation with phosphorylated tau (r = 0.81) but no association with Aβ. ⁸ Furthermore, correlations between FTP PET and post mortem tau pathology were observed across neocortical regions, even in cases with low tau densities, while no significant associations were found with Aβ plaques, TAR DNA‐binding protein 43 pathology, or tau pathology in individuals with possible primary age‐related tauopathy (PART). ⁹

The tau PET overlap index (OI) technique was developed to assess the spatial consistency of tau deposition by overlapping baseline and follow‐up scans. ¹⁰ Identifying voxel‐wise spatial consistency may reduce the impact of statistical noise components in PET data. In prior work, OI demonstrated greater sensitivity for evaluating tau signal and detecting longitudinal changes compared to standard region of interest (ROI)–based measures like SUVR. ¹⁰ While OI has shown potential for detecting tau signal, ¹⁰ , ¹¹ further validation of the tau PET OI is required. Therefore, in this study, we aimed to (1) validate the tau PET OI against neuropathologic diagnoses determined at autopsy, (2) visually assess its correspondence with immunohistochemistry (IHC) and autoradiography (ARG) at autopsy, and (3) evaluate its ability to detect early Braak tangle stages.

2. METHODS

2.1. Participants

All participants (n = 57) were part of the Mayo Clinic Study of Aging (MCSA) or Mayo Clinic Alzheimer's Disease Research Center, all of whom underwent autopsy and at least two serial FTP tau PET scans and 3T T1‐weighted magnetic resonance imaging (MRI) within an average of 16 months. Clinical diagnoses were made by the Mayo Clinic standard consensus diagnosis method after thorough clinical and cognitive assessment by neurologists, geriatricians, neuropsychologists, and study coordinators, which included psychometric assessments. ⁷ Neuropathological assessments were performed 2.54 ± 1.36 years after the most recent tau PET.

All participants or designees provided written consent with approval of Mayo Clinic and Olmsted Medical Center Institutional Review Boards.

2.2. Neuroimaging methods

All participants underwent PET/computed tomography (CT) imaging after intravenous bolus administration of 370 MBq (range: 333–407 MBq) of FTP. PET/CT image acquisition commenced 80 to 100 minutes post‐injection, during which a 20‐minute static PET scan was performed. ⁷ , ¹⁰

RESEARCH IN CONTEXT

1. Systematic review: We performed a literature review using PubMed to identify studies comparing tau positron emission tomography (PET) to neuropathological findings. Prior studies have shown correlations between standardized uptake value ratio (SUVR) and neurofibrillary tangle (NFT) burden, particularly in later Braak stages, but few studies have evaluated longitudinal tau PET overlap index (OI) in relation to immunohistochemistry (IHC)‐ or autoradiography (ARG)‐confirmed tau pathology, especially in early disease stages or across non‐Alzheimer's disease tauopathies.

2. Interpretation: Our findings demonstrate that tau PET OI shows higher concordance with NFT severity and post mortem tau pathology than conventional SUVR, particularly in early Braak stages and in 4R tauopathies. OI‐detected tau voxels corresponded spatially with IHC and ARG findings, supporting its enhanced sensitivity for detecting early and localized tau accumulation.

3. Future directions: Further studies are needed to validate OI against broader neuropathological controls and assess its predictive value for cognitive decline. Integration of OI with standard SUVR may improve tau PET–based diagnostics and staging across diverse tauopathies.

The image processing pipeline adhered to previously validated methods described in prior literature. ¹⁰ For each participant, tau PET images were rigidly co‐registered to their corresponding high‐resolution T1‐weighted MRIs using six degrees of freedom via the coregistration module in SPM12. T1‐weighted MRIs were segmented into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) using the Unified Segmentation algorithm with the MCSA202 template and tissue class priors. An image consisting of labeled ROIs was propagated from template space to individual space using the inverse of the spatial normalization parameters for each participant. These ROI labels were thereby transformed from template space to each individual's native MRI space. Rigid co‐registration was iteratively applied to align the T1 images from all time points to their mean image, which was updated until convergence. ¹² The resulting registration was applied to all PET images of each participant to put all images for a given participant into alignment with the individual's mean T1. The tau PET images for each time point in the serial scans were resampled into the space of the mean T1 magnetization‐prepared rapid gradient echo. SUVR maps were generated using the cerebellar crus as the reference region. For each ROI, median tracer uptake was calculated after excluding voxels classified as CSF. A tau PET meta‐ROI was defined to include the amygdala, entorhinal cortex (ERC), fusiform gyrus, parahippocampal gyrus, and inferior and middle temporal gyrus—brain regions known for brain area with early tau deposition. ⁷ , ¹⁰ Meta‐ROI SUVR was computed as the arithmetic mean of median values across these ROIs, intentionally avoiding voxel‐size weighting to maintain sensitivity to small but pathologically significant regions such as the ERC and amygdala. ¹⁰ All analyses were performed on non‐partial volume corrected (PVC) PET images.

2.3. OI calculation

OI, as previously published, is a reliable method for detecting tau accumulation using tau PET imaging. ¹⁰ OI quantifies the spatial consistency of elevated SUVR voxels across two serial tau PET scans. ¹⁰ For all participants, voxels within ROIs in each individual's MRI atlas template and voxels within the meta‐ROI exceeding an intensity threshold, defined as an SUVR‐based threshold applied for voxel binarization in OI calculation and set to 1.3 based on prior work, ¹⁰ were binarized to form baseline (M_b ) and follow‐up (M_f ) masks. Small clusters (< 20 voxels, 18‐connectivity) were excluded independently from M_b and M_f . OI was calculated as the ratio of overlapping voxels (N_overlap ; number of overlapping voxels between M_b and M_f ) between scans to the number of suprathreshold voxels in the baseline mask M_b (N_b ). ¹⁰

$$OI=\frac{N_{overlap}}{N_b}$$

Unlike symmetric metrics (e.g., Dice coefficient or Jaccard index), OI is asymmetrically normalized to the baseline to better capture early, spatially limited tau distribution. ¹⁰ An additional metric, overlap size (OS), was computed as the ratio of N_overlap to the total number of voxels in the ROI (N_ROI). These indices were calculated for each pair of serial scans to assess spatial persistence.

$$OS=\frac{N_{overlap}}{N_{ROI}}$$

Positivity cutoffs were defined as 0.5 for OI and 1.29 for SUVR. ⁷ , ¹⁰ For visual comparison, an overlapped clustered mask image of individual participants between M_b and M_f was extracted.

2.4. Neuropathology methods

Standardized post mortem neuropathologic evaluations and brain tissue sampling were performed following the guidelines of the Consortium to Establish a Registry for Alzheimer's Disease (CERAD) protocol. ¹³ Formalin‐fixed, paraffin‐embedded brain sections were routinely stained with hematoxylin and eosin (H&E) and a modified Bielschowsky silver stain to assess histopathological features. Neuropathologic staging included assignment of Braak NFT stage and quantification of neuritic plaque density, in accordance with the criteria established by the National Institute on Aging–Alzheimer's Association (NIA‐AA). ¹⁴ , ¹⁵ Detection of argyrophilic grains was carried out using Bielschowsky staining and AT8 immunohistochemistry, with supplemental 4R tau immunolabeling where necessary. ⁷ Aβ immunohistochemistry was performed in the neocortex, hippocampus, basal ganglia, and cerebellum to determine Thal amyloid phase: phase 1 (neocortex), phase 2 (CA1/subiculum), phase 3 (basal ganglia or dentate fascia), phase 4 (midbrain or CA4), and phase 5 (cerebellum). ⁷ Cerebrovascular pathology was evaluated based on the presence of WM rarefaction and/or cribriform changes in the basal ganglia, in conjunction with moderate to severe arteriolosclerosis, with or without coexisting microinfarcts, lacunar infarcts (<1 cm), or larger territorial infarcts. ⁷ Apolipoprotein E genotyping was conducted ante mortem using DNA extracted from peripheral blood samples, following standard laboratory protocols. ⁷

Neuropathologic assessments were performed by board‐certified neurologists and neuropathologists at the Mayo Clinic. Findings from post mortem brain examinations were systematically documented in standardized autopsy reports, which included detailed evaluations of NFT stage, amyloid plaque burden, and coexisting neurodegenerative or vascular pathologies. Neuropathologic diagnoses for this study were assigned based on the documented findings in these reports. To determine primary and secondary pathologic classifications, we used a research‐based rubric that considered the relative severity, frequency, and rarity of each pathology, independent of the individual's clinical presentation. ⁷ For instance, individuals exhibiting high‐level AD pathology (refer to neuropathological diagnostic criteria in the Supplemental Methods in supporting information), defined by Braak stage V or greater and at least moderate neuritic plaque burden, were assigned a primary diagnosis of AD, regardless of any coexisting pathologies. Conversely, when AD pathology was limited in extent but α‐synucleinopathy was more prominent, the primary diagnosis was assigned as dementia with Lewy bodies (DLB), such as diffuse (DLBD) or brainstem‐predominant (BLBD) subtypes, as exemplified by participants #38 to #46, as shown in Table 1. Similarly, vascular disease was classified as secondary when occurring alongside advanced AD pathology. Primary tauopathies, which rarely overlap with high‐level AD pathology, were considered primary diagnoses. In contrast, common age‐related tauopathies were regarded as secondary, reflecting their frequent co‐occurrence with other neurodegenerative conditions. In cases in which multiple coexisting neuropathologies were reported in the autopsy reports, both pathologies were jointly considered when assigning the primary diagnosis. For example, participants #53, #54, and #55, as shown in Table 1, exhibited multiple coexisting pathologic features, and their diagnostic classification reflected the combined contribution of these coexisting diseases.

TABLE 1.

Individual participant data.

Participant	Sex	Primary clinical Dx	Primary pathologic Dx	Secondary pathologic Dx	Tau, SUVR a	Tau, OI b	Thal	Braak	Neuritic plaques d	Neurofibrillary tangles d	Scan to death (mos)
1	F	Probable AD	AD (High)	ALB	2.95	0.9957	4	VI	Frequent	Frequent	15
2	F	Dementia with PVD	AD (High)	ARTAG	2.39	0.9552	5	VI	Frequent	Frequent	7
3	F	Probable AD	AD (Intermed)	AA	2.38	0.9726	3	VI	Frequent	Frequent	8
4	F	Dementia with PVD	AD (High)	AA	2.31	0.9973	5	VI	Frequent	Frequent	30
5	F	Probable AD	AD (High)	DLBD	2.18	0.9900	5	V	Frequent	Frequent	45
6	M	Probable AD	AD (High)	ALB	2.12	0.9964	4	VI	Missing	Missing	20
7	F	Probable AD	AD (High)	ALB	2.12	0.9966	4	VI	Frequent	Frequent	31
8	M	Probable AD	AD (High)	CAA	2.11	0.9736	4	VI	Frequent	Frequent	16
9	M	Probable AD	AD (High)	ALB	2.08	0.9703	4	VI	Frequent	Frequent	29
10	M	Probable AD	AD (High)	DLBD	1.90	0.9866	5	V	Frequent	Frequent	21
11	M	Probable AD	AD (Intermed)	DLBD	1.86	0.9870	3	VI	Frequent	Frequent	53
12	M	Probable AD	AD (High)	MSA	1.85	0.9792	5	VI	Frequent	Moderate	28
13	F	Dementia with PVD	AD (High)	ALB	1.82	0.9933	4	VI	Frequent	Frequent	17
14	M	MCI	AD (High)	ALB	1.79	0.9718	4	VI	Frequent	Frequent	81
15	F	Probable AD	AD (High)	AA	1.78	0.9776	5	V	Missing	Missing	21
16	F	Dementia c	AD (High)	AA	1.72	0.9776	5	VI	Frequent	Frequent	11
17	F	Probable AD	AD (High)	LATE	1.70	0.9792	4	VI	Frequent	Frequent	34
18	M	Probable AD	AD (High)	DLBD	1.69	0.9687	4	V	Moderate—Frequent	Frequent	21
19	F	Dementia c	AD (High)	ALB	1.68	0.9148	4	V	Frequent	Frequent	24
20	M	Probable AD	AD (High)	ALB	1.61	0.9454	5	VI	Frequent	Frequent	28
21	F	Probable AD	AD (High)	A‐TDP	1.61	0.9588	3	V	Frequent	Frequent	15
22	M	lvPPA	AD (High)	ALB	1.50	0.9085	5	VI	Frequent	Frequent	11
23	M	LBD	AD (High)	DLBD	1.49	0.9386	5	V	Frequent	Frequent	17
24	M	Probable AD	AD (High)	ALB	1.48	0.9139	4	V	Frequent	Frequent	14
25	M	Normal	AD (High)	LATE	1.45	0.9135	5	VI	Frequent	Frequent	35
26	F	Dementia c	AD (High)	VaD	1.34	0.9148	5	V	Frequent	Sparse—Frequent	10
27	M	FTD	AD (High)	AA	1.32	0.7941	4	V	Frequent	Frequent	37
28	F	MCI	AD (Intermed)	AA	1.31	0.8017	5	IV	Moderate	Frequent	7
29	M	Probable AD	AD (High)	BLBD	1.29	0.8953	5	VI	Rare	Moderate	30
30	F	Normal	AD (High)	AA	1.28	0.7398	4	V	Frequent	Frequent	39
31	F	MCI	AD (Low)	A‐TDP	1.27	0.8426	2	IV	Sparse	Frequent	38
32	M	Probable AD	AD (High)	AA	1.21	0.5138	4	V	Moderate	Moderate	56
33	M	Probable AD	AD (Low)	LATE	1.18	0.8383	4	II	Sparse	Moderate	9
34	M	Normal	AD (Intermed)	ALB	1.11	0.1209	4	IV	Sparse	Sparse	29
35	M	MCI	AD (Low)	AA	1.07	0.1195	4	II	Sparse	Sparse	31
36	M	Normal	AD (Intermed)	TLBD	1.05	0.0415	2	III	Missing	Missing	9
37	M	MCI	AD (Intermed)	AA	0.92	–	3	IV	Sparse	Moderate	3
38	M	PD	DLBD	AD (Low)	1.33	0.8331	3	III	Moderate	Frequent	11
39	F	Dementia c	DLBD	AD (Low)	1.30	0.7062	4	II	Moderate	Frequent	22
40	M	LBD	DLBD	AD (High)	1.28	0.9264	3	V	Missing	Missing	23
41	M	LBD	DLBD	AD (Intermed)	1.26	0.6732	4	IV	Frequent	Frequent	14
42	M	LBD	DLBD	AD (Low)	1.21	0.7786	5	I	None	Sparse—Moderate	27
43	M	LBD	DLBD	PART‐AD (Low)	1.19	0.6725	1	II	None	Moderate	32
44	M	LBD	DLBD	AD (Low)	1.16	0.5532	4	II	Frequent	Frequent	20
45	M	LBD	DLBD	AD (Intermed)	1.14	0.3652	4	IV	Moderate	Frequent	47
46	F	Normal with RBD	DLBD	AD (Intermed)	1.09	0.1097	3	III	None	Moderate	24
47	F	Depression	BLBD	None	1.22	0.6945	0	0	None	None	6
48	M	Normal	BLBD	PART	1.08	0.1584	3	III		Moderate	15
49	F	Normal	AGD	AD (Low)	1.25	0.8267	1	III	Absent	Moderate	44
50	M	FTD	CBD	TLBD	1.16	0.5714	2	III		Frequent	17
51	M	FTD	FTDP‐17 (MAPT p.V337M mutation)	None	1.35	0.8518	0	V	Missing	Moderate—Frequent	15
52	F	FTD	FTDP‐17 (MAPT p.P301L mutation)	VaD	1.04	0.7820	0	0	Absent	Absent	63
53	M	FTD	FTLD‐tau, CBD	AD (Low)	1.30	0.8691	2	II	Sparse	Sparse	31
54	F	Dementia c	FTLD‐tau, CBD	LATE	1.14	0.5985	0	‐ e	Missing	Missing	38
55	M	svPPA	FTLD‐tau, GGT	MTS	1.23	0.7976	0	0	Absent	Absent	52
56	M	PSP	PSP	AA	1.22	0.6572	3	‐ e	Sparse	Mild	15
57	M	Normal	PSP	LATE	1.05	0.1538	0	IV	None	Moderate	57

Abbreviations: AA, amyloid angiopathy; AD (high), Alzheimer's disease high likelihood; AD (intermed), Alzheimer's disease intermediate likelihood; AD (low), Alzheimer's disease low likelihood; AD, Alzheimer's disease; AGD, argyrophilic grain disease; ALB, amygdala‐predominant Lewy bodies; A‐TDP, TAR DNA‐binding protein 43–positive inclusions within amygdala; BLBD, brainstem Lewy body disease; CAA, cerebral amyloid angiopathy; CBD, corticobasal degeneration; DLBD, diffuse Lewy body disease; Dx, diagnosis; F, female; FTD, frontotemporal dementia; FTDP‐17, frontotemporal degeneration with Parkinsonism linked to chromosome 17; FTLD‐tau, frontotemporal lobar degeneration with 4R‐tauopathy; GGT, globular glial tauopathy; LATE, limbic‐predominant age‐related TAR DNA‐binding protein 43 encephalopathy; LBD, Lewy body disease; lvPPA, logopenic variant of primary progressive aphasia; M, male; MAPT, microtubule‐associated protein tau; MCI, mild cognitive impairment; MMSE, Mini‐Mental State Examination; mos, months; MSA, multiple system atrophy; MTS, medial temporal sclerosis; OI, overlap index; PART, primary age‐related tauopathy; PD, Parkinson's disease; PiB, Pittsburgh compound B; PSP, progressive supranuclear palsy; RBD, rapid eye movement behavior disorder; ROI, region of interest; SUVR, standardized uptake value ratio; svPPA, semantic variant primary progressive aphasia; TDP, TAR DNA binding protein; TLBD, transitional Lewy body disease; VaD, vascular disease.

^a Tau SUVR derived from meta‐ROI (refer to sections 2.2 and 2.3).

^b Tau OI derived from meta‐ROI (refer to sections 2.2 and 2.3).

^c Dementia hard to classify.

^d Pathology region: entorhinal cortex.

^e The presence of a tauopathy (other than aging/AD) precludes Braak staging.

To enable visual comparison between tau PET OI patterns and underlying histopathology, immunohistochemistry (IHC) was performed on selected paraffin‐embedded brain tissue sections. Tissue staining was conducted using validated antibodies targeting hallmark tau pathologies. Specifically, tau pathology was assessed using AT8 (phospho‐tau, Ser202/Thr205; dilution 1:1000; Endogen, RRID:AB_223647) and PHF‐1 (phospho‐tau, Ser396/Ser404, RRID:AB_2315150) antibodies. ⁷ The resulting IHC‐stained sections were used for regional evaluation of tau accumulation and subsequently compared to corresponding regions on tau PET for spatial concordance.

2.5. Autoradiographic methods

Autoradiography (ARG) of FTP was performed on adjacent 5 µm formalin‐fixed paraffin‐embedded (FFPE) brain sections from the cases selected from the research brain bank. ⁵ , ¹⁶ Sections were deparaffinized and incubated in 20 µCi of FTP in 500 µL of phosphate‐buffered saline (PBS) for 60 minutes. ⁵ , ¹⁶ To demonstrate displacement ARG for non‐specific binding, 30 µL of the respective reference compound (3.8 mM) was dissolved in 70 µL of ethanol (EtOH), combined with 20 µCi of [¹⁸F]‐labeled compound, and brought up to 500 µL using PBS. ⁵ , ¹⁶ This mixture was applied to adjacent sections in the same manner. After incubation, the slides were washed in PBS for 1 minute, 70% EtOH/distilled water for 2 minutes, 30% EtOH/distilled water for 1 minute, and distilled water for 1 minute to remove the unbound tracer. ⁵ , ¹⁶ Slides were air dried and then exposed to phosphor screens for 60 minutes. The screens were imaged using a Typhoon FLA 7000 (GE) laser scanner. ⁵ , ¹⁶

2.6. Cognitive testing and association analysis

Cognitive testing was completed within 4 months of the follow‐up PET scans. All tests were administered by experienced psychometrists and supervised by board‐certified clinical neuropsychologists. ¹⁷ The memory domain was assessed using the Auditory Verbal Learning Test (AVLT) delayed recall, Wechsler Memory Scale–Revised Logical Memory II, and Visual Reproduction II subtests. Individual test scores were first converted to z scores using the mean and standard deviation from the MCSA 2004 to 2012 cognitively unimpaired (CU) cohorts and weighted to match the 2013 Olmsted County, Minnesota, population. ¹⁷ The memory domain z score was then calculated as the average of the z scores from the three memory subtests. ¹⁷ In addition to the memory domain z score, the total AVLT score (avlt.sum) and Mini‐Mental State Examination (MMSE) score were also obtained, with all cognitive assessments conducted within a standardized neuropsychological framework to ensure consistency across participants.

2.7. Statistical analysis

All statistical analyses were conducted using R Statistical Software (v4.4.2). Scatterplot analyses were conducted across the following pairs: (1) meta‐ROI OI versus meta‐ROI follow‐up scan SUVR, (2) meta‐ROI OS versus meta‐ROI follow‐up scan SUVR, (3) meta‐ROI OI versus meta‐ROI OS, and (4) meta‐ROI OI versus meta‐ROI follow‐up scan SUVR within overlapped voxels. For each comparison, local regression was applied as a non‐parametric smoothing technique to visualize data trends. The fitted curves were generated using locally weighted polynomial regression with 95% confidence bands. In each case, the x axis represented OI or OS, and the y axis corresponded to OS, SUVR, or follow‐up SUVR values. Spearman rank correlation coefficient (rho) was used to assess associations between Braak tangle stage and regional tau PET metrics, including the ERC OI and ERC SUVR. In addition, Spearman rank correlation coefficient was evaluated between Braak tangle stage and posterior hippocampus area OI, and posterior hippocampus area SUVR, which included the hippocampus, parahippocampal gyrus, fusiform gyrus, and inferior temporal cortex.

For each participant, associations between these cognitive measures (memory domain z score, avlt.sum, and MMSE) and tau PET metrics—OI, OS, and SUVR—were evaluated. Correlation analyses were performed to assess the relationships between cognitive performance and PET‐derived measures.

3. RESULTS

3.1. Tau PET OI and autopsy participants

Neuropathologic diagnoses are summarized in Table 1, and participant demographics are presented in Table 2. Autopsy diagnoses included AD (n = 37, participants 1–37), DLBD (n = 9, participants 38–46), BLBD (n = 2, participants 47–48), argyrophilic grain disease (AGD; n = 1, participant 49), CBD (n = 1, participant 50), frontotemporal degeneration with Parkinsonism linked to chromosome 17 (FTDP‐17; n = 2, participants 51–52, with distinct mutation types: V337M and P301L), FTLD‐tau with CBD (n = 2, participants 53–54), FTLD‐tau with globular glial tauopathy (GGT; n = 1, participant 55), and PSP (n = 2, participants 56–57). The cohort comprised 35 men and 22 women. While seven participants were clinically “normal,” all exhibited some degree of neuropathology. Primary neuropathologic diagnoses within the AD spectrum were further categorized as AD (low), AD (intermediate), and AD (high) according to autopsy findings. Participants with AD exhibited a moderate degree of cognitive impairment, with a mean MMSE score of 19.8. Additional demographic and clinical characteristics of the cohort are provided in Tables 1 and 2.

TABLE 2.

Demographics by neuropathology diagnosis.

Demographics	AD (n = 37)	DLBD (n = 9)	BLBD (n = 2)	CBD (n = 1)	FTLD‐tau, GGT (n = 1)	FTLD‐tau, CBD (n = 2)	FTDP‐17 (n = 2)	PSP (n = 2)	AGD (n = 1)	Total (n = 57)
Sex
Female	16	2	1	0	0	1	1	0	1	22
Male	21	7	1	1	1	1	1	2	0	35
Education
Mean (SD)	14.4 (2.6)	15.3 (2.6)	14.5 (3.5)	16	12	14 (2.8)	13.5 (0.7)	19 (1.4)	18	14.7 (2.6)
Age at scan (years)
Mean (SD)	78.3 (9.9)	76 (8.5)	83.4 (3.6)	74	57	67.8 (5.1)	60 (12.4)	69.7 (19.4)	75	76.3 (10.4)
Scan to death (months)
Mean (SD)	25.1 (16.1)	24.4 (10.5)	10.5 (6.4)	17	52	34.5 (4.9)	39 (33.9)	36 (29.7)	44	26.4 (16.1)
Range	3–81	11–47	6–15			31–38	15–63	15–57		3–81
APOE 𝜀4, no.
No	15	5	1	1	1	2	2	1	1	29
Yes	21	4	1	0	0	0	0	1	0	27
MMSE a								(n = 1)
Mean (SD)	19.8 (6.8)	21.8 (8)	26 (4.2)	13	17	–	–	28	30	20.7 (7)
GMWM PVCN PiB SUVR b								(n = 1)
Mean (SD)	2.45 (0.54)	2.01 (0.41)	1.39 (0.23)	1.4	–	1.36 (0.08)	–	1.26	1.4	2.22 (0.61)
Range	1.32–3.84	1.34–2.51	1.23–1.55		–	1.3–1.42	–			1.23–3.84
GMWM PVCN FTP‐SUVR c
Mean (SD)	1.84 (0.62)	1.25 (0.08)	1.18 (0.12)	1.24	1.24	1.31 (0.09)	1.25 (0.26)	1.2 (0.13)	1.31	1.63 (0.57)
Range	0.97–3.64	1.12–1.37	1.1–1.26			1.24–1.37	1.07–1.44	1.1–1.29		0.97–3.64

Abbreviations: AD, Alzheimer's disease; AGD, argyrophilic grain disease; APOE, apolipoprotein E; BLBD, brainstem Lewy body disease; CBD, corticobasal degeneration; DLBD, diffuse Lewy body disease; FTDP‐17, frontotemporal degeneration with Parkinsonism linked to chromosome 17; FTLD‐tau, frontotemporal lobar degeneration with 4R‐tauopathy; GGT, globular glial tauopathy; GM, gray matter; MMSE, Mini‐Mental State Examination; PiB, Pittsburgh compound B; PSP, progressive supranuclear palsy; PVCN, partial volume correction no; SD, standard deviation; SUVR, standardized uptake value ratio; WM, white matter.

^a MMSE was missing in 12 of the group at the time of the imaging visit as they were unable to be tested due to disease severity.

^b Voxel size weighted average of uptake in GM and WM regions prefrontal, orbitofrontal, parietal, temporal, anterior and posterior cingulate, and precuneus scaled by weighted average of cerebellar crus 1 and 2 without PVCN.

^c Voxel size weighted average of uptake in GMWM regions entorhinal, amygdala, parahippocampal, fusiform, inferior and middle temporal scaled by weighted average of cerebellar crus 1 and 2 without PVCN.

3.2. Tau PET OI relationship to autopsy diagnosis

This analysis was first conducted using the meta‐ROI. Meta‐ROI OI and meta‐ROI follow‐up scan SUVR distributions were evaluated across 11 primary neuropathological diagnostic groups, based on autopsy reports (Figure 1). In Figure 1, marker symbols denote the NFT severity in the ERC as reported in autopsy findings. The positivity cutoffs were set at 0.5 for OI and 1.29 for SUVR, as described in section 2.3. ⁷ , ¹⁰ One participant was excluded from OI analysis because no cluster map meeting the intensity threshold of 1.3 was detected, and therefore OI could not be calculated. Based on autopsy findings, AD cases were categorized into three subgroups: AD (low), AD (intermediate), and AD (high). Across all AD groups, the proportion of individuals with NFT severity in the ERC rated as sparse or higher who demonstrated tau PET meta‐ROI OI and meta‐ROI follow‐up scan SUVR positivity was 83.8% (31/37) and 73% (27/37), respectively. More specifically, in the AD (high) subgroup, these proportions were 92.9% (26/28) for OI and 85.7% (24/28) for follow‐up scan SUVR. In the AD (intermediate) subgroup, both OI and follow‐up scan SUVR positivity rates were 50% (3/6). Notably, in the AD (low) subgroup, meta‐ROI OI positivity (66.7%, 2/3) showed greater detection of cases with NFT severity rated as sparse or higher in the ERC compared to meta‐ROI follow‐up scan SUVR, which showed 0% positivity (0/3).

FIGURE 1.

Comparison of meta‐ROI OI and meta‐ROI follow‐up scan SUVR across primary neuropathologic diagnoses. Based on autopsy reports, 57 participants were classified into 11 primary neuropathologic diagnostic groups, with AD further subdivided into three categories: low, intermediate, and high. Marker symbols indicate NFT severity in the ETC as reported in the autopsy findings, categorized into eight levels. Additionally, based on autopsy findings, cases corresponding to 4R tauopathies (AGD; CBD; PSP; FTDP‐17; FTLD‐tau, GGT; FTLD‐tau, CBD) with NFT severity in the ETC classified as “missing,” “none or absent” were denoted using a distinct purple marker. The meta‐ROI included the amygdala, fusiform gyrus, ETC, parahippocampal gyrus, middle temporal gyri, and inferior temporal gyri. An OI cutoff value of 0.5 is indicated by a vertical dashed line, and an SUVR cutoff value of 1.29 is similarly represented by a vertical dashed line. AD, Alzheimer's disease; AGD, argyrophilic grain disease; BLBD, brainstem Lewy body disease; CBD, corticobasal degeneration; DLBD, diffuse Lewy body disease; ETC, entorhinal cortex; FTDP‐17, frontotemporal degeneration with Parkinsonism linked to chromosome 17; FTLD‐tau, frontotemporal lobar degeneration with tau pathology; GGT, globular glial tauopathy; NFT, neurofibrillary tangle; OI, overlap index; PSP, progressive supranuclear palsy; ROI, region of interest; SUVR, standardized uptake value ratio. SUVR is displayed as SUVr within the figure panels.

In the DLBD group, meta‐ROI OI revealed positivity above the predefined cutoff (= 0.5) in 67% (6/9) of participants, whereas meta‐ROI SUVR at follow‐up indicated positivity in only 22% (2/9). Importantly, most DLBD cases exhibited moderate to frequent NFT severity in the ERC based on autopsy findings, suggesting that the OI signal in this group likely represents secondary NFT pathology consistent with true positive detection.

Meta‐ROI OI also indicated that a greater number of individuals in non‐AD pathological diagnosis groups showed OI signal above the predefined cutoff (= 0.5): CBD (1/1), AGD (1/1), and PSP (1/2), whereas meta‐ROI follow‐up scan SUVR values remained below threshold in most cases: CBD (0/1), AGD (0/1), and PSP (0/2). These findings likely reflect the influence of non‐specific FTP binding rather than true 4R tau binding, and AD‐type tau co‐pathology could also contribute in some cases.

Meta‐ROI analyses identified a subset of participants with positive meta‐ROI OI values despite negative meta‐ROI SUVR values, while their NFT severity was classified as “missing,” “none or absent.” For participant P.52, who was diagnosed with FTDP‐17 with a p.P301L mutation (Figure S1 in supporting information), the additional autopsy report described diffuse, severe 4R‐tau‐immunoreactive grains with ballooned neurons, most prominent in the frontotemporal neocortices. Baseline and follow‐up tau PET SUVR maps demonstrated a widespread distribution of meningeal off‐target signal. Inspection of the overlap mask indicated that the meningeal off‐target signal could contribute to the OI calculation. Nevertheless, localized tau signal was also observed in the ERC and inferior temporal cortex (white arrows, Figure S1), and these focal signals were captured within the overlapping voxel distribution. Accordingly, the positive meta‐ROI OI in this participant likely reflects the combined detection of meningeal off‐target signal and localized temporal tau signal, while the post mortem diagnosis of 4R tauopathy indicates an influence of non‐specific FTP binding on the tau PET signal. For participant P.54, who was neuropathologically diagnosed with FTLD‐tau with CBD (Figure S1), the additional autopsy report indicated tau‐4‐immunoreactive lesions and cortical NFTs, with the greatest involvement in the temporal lobe. Similar to these findings, baseline and follow‐up tau PET SUVR maps showed elevated tau signal throughout the temporal cortex, and overlapping tau voxels were similarly detected across this region. For participant P.55, who was diagnosed with FTLD‐tau with GGT (Figure S1), the additional pathological report described severe, diffuse 4R‐tau‐positive glial inclusions in GM, characterized by globular glial cytoplasmic inclusions, with widespread involvement and greatest severity in the temporal lobe. Baseline and follow‐up tau PET SUVR maps demonstrated high tau signal throughout the temporal cortex, and the overlap mask confirmed a corresponding distribution of overlapping tau voxels. Although both P.54 and P.55 were reported post mortem with 4R tauopathy, the tau PET signal observed across the temporal cortex appears to have been influenced by 4R‐tau FTP binding; nonetheless, detection of overlapped tau voxels throughout the temporal cortex by the meta‐ROI OI met criteria for positivity. In contrast, participant P.47, who was diagnosed with BLBD (Figure S2 in supporting information), showed localized regions of elevated tau signal on baseline and follow‐up tau PET SUVR images, with corresponding overlapping tau voxels observed focally across the brain. In the right hemisphere amygdala, overlapping tau voxels demonstrated spatial correspondence with IHC findings indicating NFT pathology in the amygdala region, as defined by the brain atlas. No overlapping tau voxels or NFTs were visually observed in the ERC, whereas overlapping tau voxels were detected in the right hemisphere inferior temporal and middle temporal cortices. Although ERC NFT was negative in this participant, the meta‐ROI OI was positive, reflecting detection of spatially persistent tau signal within other regions included in the meta‐ROI. The correspondence between amygdala IHC findings and overlapping tau voxels, together with detection of tau signal in inferior and middle temporal cortices, illustrates the ability of OI to identify tau accumulation within the meta‐ROI.

Examining the distribution of overlapping voxels within the meta‐ROI using OS (Figure S3A in supporting information), the mean value for the AD (high) group was 0.79 ± 0.23 (range, 0.22–0.99), for the AD (intermediate) group was 0.42 ± 0.47 (range, 0–0.97), and for the AD (low) group was 0.22 ± 0.20 (range, 0.01–0.39). Among the other pathological diagnosis groups, the mean values were as follows: DLBD, 0.23 ± 0.18 (range, 0.003–0.56); BLBD, 0.02; CBD, 0.15; AGD, 0.30; PSP, 0.15; FTDP‐17, 0.10 (P301L mutation) and 0.60 (V337M mutation); FTLD‐tau with GGT, 0.24; and FTLD‐tau with CBD, 0.35. When examining the mean SUVR of overlapping voxels between baseline and follow‐up scans (Figure S3B), the AD (high) and AD (intermediate) groups demonstrated a broader distribution compared to the other neuropathological diagnosis groups.

Meta‐ROI OI, meta‐ROI OS, and meta‐ROI follow‐up scan SUVR demonstrated strong but non‐linear associations with one another (meta‐ROI OI vs. meta‐ROI OS, ρ = 0.94, p < 0.001; meta‐ROI OI vs. meta‐ROI follow‐up scan SUVR, ρ = 0.93, p < 0.001), with individuals in the AD (high) group showing the highest OI, SUVR values and a tendency toward broader distribution of overlapped voxels within meta‐ROI (Figure S3C). Similar patterns were observed in the relationships between meta‐ROI OI and meta‐ROI follow‐up scan SUVR in overlapping voxels (ρ = 0.93, p < 0.001) and between meta‐ROI OS and meta‐ROI follow‐up scan SUVR (ρ = 0.97, p < 0.001; Figure S3D).

We further compared OI and SUVR distributions in the hippocampus, ERC, and superior/middle temporal gyri (Figure 2). In these plots, the different shapes represent NFT severity in the respective regions as reported in the autopsy findings. The overall distribution patterns were consistent with those observed in the previous meta‐ROI analysis. In the AD (high) group, all participants demonstrated positive meta‐ROI OI results. In contrast, individual ROI‐level OI analyses showed positive values for all participants except one (P.32; Table 1), who exhibited negative OI values in the hippocampus and ERC (hippocampus OI = 0.1395; ERC OI = 0.2511). Despite this negativity in the ERC, the meta‐ROI OI for this participant remained positive (OI = 0.5138), indicating that positive OI values in the temporal cortex and other regions comprising the meta‐ROI contributed to the overall meta‐ROI OI positivity. Interestingly, within the AD (low) group, tau PET OI identified a greater proportion of individuals with NFT severity in the ERC rated as moderate or higher compared to tau PET SUVR. Among participants who exhibited positive meta‐ROI OI but negative meta‐ROI SUVR values, and had NFT status classified as missing, none, or absent, in the participant diagnosed neuropathologically with FTLD‐tau with CBD (P.54), hippocampus and ERC OI values were negative (hippocampus OI = 0.0122; ERC OI = 0.2550), whereas OI values in the superior/middle temporal cortex were positive (OI = 0.5220). Although the hippocampus was not included in the meta‐ROI, the positive meta‐ROI OI indicates that elevated OI values within temporal cortical regions drove the overall meta‐ROI result, in contrast to the hippocampus and ERC OI. In contrast, another participant with the same neuropathological diagnosis (P.53) showed positive OI values across all three regions displayed in Figure 2. This pattern suggests that elevated tau PET signal was broadly distributed throughout the temporal lobe, resulting in a high meta‐ROI OI value and a positive detection. For the participant diagnosed with FTLD‐tau with GGT (P.55), hippocampus OI was negative (OI = 0.4354), whereas OI values in the remaining regions were positive (ERC OI = 0.7016; superior/middle temporal cortex OI = 0.7682). This participant similarly demonstrated a high tau PET signal distributed throughout the temporal lobe, accompanied by a positive meta‐ROI OI with a high OI value. The presence of negative hippocampal OI and positive OI in other regions suggests that OI effectively captures the regional distribution of tau signal, when assessed at the individual ROI level. For the BLBD participant (P.47), hippocampus OI was negative (OI = 0.2759), and ERC OI could not be calculated because a low tau PET signal in the baseline scan prevented cluster formation, resulting in exclusion of this region from the samples of the figure. In contrast, OI in the superior/middle temporal cortex was positive (OI = 0.7352). The positive meta‐ROI OI in this case indicates that OI values of focal overlapped voxels within temporal cortical regions contributed to the overall meta‐ROI outcome.

FIGURE 2.

Comparison of OI and SUVR in three regions—the hippocampus, ETC, and superior/middle temporal gyrus—based on primary neuropathologic diagnoses from autopsy reports. Fifty‐seven participants were classified into 11 primary diagnostic groups, with AD further subdivided into low, intermediate, and high categories. Marker symbols indicate eight levels of NFT severity in the three regions as reported in the post mortem findings. In addition, based on autopsy findings, cases corresponding to 4R tauopathies (AGD; CBD; PSP; FTDP‐17; FTLD‐tau, GGT; FTLD‐tau, CBD) with NFT severity in the ETC classified as “missing,” “none or absent” were denoted using a distinct purple marker. OI and SUVR values were calculated separately for the hippocampus, ETC, and superior/middle temporal gyrus (temporal superior/middle cortex). An OI cutoff of 0.5 is denoted by a vertical dotted line, and an SUVR cutoff of 1.29 is similarly indicated by a vertical dotted line. AD, Alzheimer's disease; AGD, argyrophilic grain disease; BLBD, brainstem Lewy body disease; CBD, corticobasal degeneration; DLBD, diffuse Lewy body disease; ETC, entorhinal cortex; FTDP‐17, frontotemporal degeneration with Parkinsonism linked to chromosome 17; FTLD‐tau, frontotemporal lobar degeneration with tau pathology; GGT, globular glial tauopathy; NFT, neurofibrillary tangle; OI, overlap index; PSP, progressive supranuclear palsy; SUVR, standardized uptake value ratio. SUVR is displayed as SUVr within the figure panels.

Meta‐ROI analysis captures tau signal by averaging across multiple temporal lobe regions; ¹⁰ however, when early tau accumulation is confined to small regions such as the ERC or hippocampus, as in Braak stage I, this averaging may attenuate localized signal. In contrast to SUVR, OI enabled detection of tau accumulation not only at the meta‐ROI level but also within individual ROIs, including the ERC. These findings indicate that OI more effectively preserves region‐specific sensitivity to early tau deposition and may better capture early disease‐related tau changes. Furthermore, because neuropathological assessments provide NFT severity at the subregional level, including regions such as the ERC, combined meta‐ROI and regional ROI analyses enable evaluation of whether OI and SUVR signals reflect underlying pathology. Consistent with this framework, individual regional analyses demonstrated that OI was more effective than SUVR in capturing early tau accumulation within specific regions.

3.3. Tau PET OI examples with IHC and ARG

Figure 3 presents representative tau PET OI images for AD (high), AD (intermediate), FTDP‐17 (V337M mutation), FTLD‐tau with CBD, CBD, PSP, DLBD, and BLBD. Overlapped tau voxels used in OI calculation were visually compared to IHC and ARG findings. In AD (high), extensive overlapped tau voxels were detected in the ERC, with corresponding IHC and ARG results of tau deposition. In AD (intermediate), despite negative meta‐ROI OI and SUVR, a small, overlapped voxel cluster was observed in the ERC, which correlated visually to IHC and ARG results. In non‐AD pathological diagnoses, overlapped tau voxels in FTDP‐17 (V337M mutation), CBD, PSP, and DLBD aligned with tau deposition in IHC and ARG. Notably, in FTLD‐tau with CBD, overlapped voxels in the superior/middle temporal cortex WM corresponded to WM tau deposition in IHC and ARG. These findings indicate that OI enhances the sensitivity of detecting FTP signal corresponding to regions with tau deposition. In 4R tauopathies, OI highlighted subtle FTP elevations in GM and nearby WM regions, which likely reflect regionally specific, non‐specific degenerative signal rather than direct 4R tau binding (Figures S4 and S5 in supporting information; Example: FTLD‐tau with CBD, CBD, PSP). In BLBD, no overlapped voxels were identified, consistent with absent tau burden on IHC and ARG results.

FIGURE 3.

Visual comparison of tau PET OI and IHC/ARG according to neuropathological diagnosis. Nine participants, classified by post mortem reports into the following primary neuropathological diagnoses, were included: AD (high), AD (intermediate), FTDP‐17 (MAPT p.V337M mutation), FTLD‐tau with CBD, CBD, PSP, DLBD, and BLBD. Participant numbers corresponding to Table 1 are shown before each diagnosis label in the format “P.XX.” IHC and ARG data were obtained from tissue slices of the entorhinal cortex, superior/middle temporal cortex, and posterior hippocampus (including the hippocampus, parahippocampal gyrus, fusiform gyrus, and inferior temporal cortex). Tau PET OI images represent coronal views corresponding to the anatomical locations of the IHC and ARG slices for both baseline and follow‐up scans, with the overlapped mask image for the region marked by a purple box shown on the right. Red voxels in the overlapped image indicate overlapping clusters with an OI intensity threshold of 1.3 in both baseline and follow‐up scans, representing the voxels used in OI calculation. Purple arrows in the IHC and overlapped images manually indicate the corresponding anatomical locations. For each participant, values for meta‐ROI OI, OS, SUVR, secondary neuropathological diagnosis, Braak tangle stage, and the scan‐to‐death interval (months) are presented. The meta‐ROI was defined to include the bilateral amygdala, fusiform gyrus, entorhinal cortex, parahippocampal gyrus, middle temporal gyrus, and inferior temporal gyrus. AD, Alzheimer's disease; ARG, autoradiography; BLBD, brainstem Lewy body disease; CBD, corticobasal degeneration; DLBD, diffuse Lewy body disease; FTDP‐17, frontotemporal degeneration with Parkinsonism linked to chromosome 17; FTLD‐tau, frontotemporal lobar degeneration with tau pathology; IHC, immunohistochemistry; OI, overlap index; OS, overlap size; PET, positron emission tomography; PSP, progressive supranuclear palsy; SUVR, standardized uptake value ratio. SUVR is displayed as SUVr within the figure panels.

3.4. Tau PET OI related to Braak tangle stage

Figure 4 illustrates OI and SUVR values computed in the ERC, stratified by Braak tangle stage. Figure 4A displays the distribution of ERC OI and SUVR across 11 primary neuropathological diagnostic categories as defined by autopsy findings. Figure 4B shows the distribution of ERC OI (rho = 0.63, p < 0.0001) and SUVR (rho = 0.75, p < 0.0001) in relation to NFT severity in the ERC, as reported in post mortem pathology. Among individuals at Braak stages (0–VI), 52.2% (12 of 23) exhibited entorhinal OI positivity corresponding to NFT severity rated as rare or higher, whereas 13.0% (3 of 23) were SUVR‐positive under the same criteria, indicating greater sensitivity of OI for early NFT detection in the ERC. In the posterior hippocampus, the distribution of OI (ρ = 0.76, p < 0.0001) and SUVR (ρ = 0.80, p < 0.0001) revealed a similar tendency. Among individuals with NFT severity rated as sparse or higher in this region, OI positivity was observed in 47.8% (11/23) and SUVR positivity in 4.3% (1/23), indicating a better detection rate of OI (Figure S6A,B in supporting information). Visual inspection of tau PET OI images from Braak stage 0 to III cases revealed spatial correspondence between overlapping tau voxels and tau deposition confirmed by IHC and ARG results in both the ERC and posterior hippocampus (Figure 4C, Figure S6C). Collectively, these findings indicate that OI was more sensitive for detecting the tau signal for suggesting early tau pathology during the early Braak tangle stages.

FIGURE 4.

Comparison of OI and SUVR by Braak tangle stage. OI and SUVR values were calculated for the entorhinal cortex. An OI cutoff value of 0.5 is indicated by a vertical dotted line, and an SUVR cutoff value of 1.29 is indicated by a vertical dotted line. Spearman rank correlation coefficients (rho, ρ) were 0.63 for OI and 0.75 for SUVR. A, Plot of OI and SUVR by Braak tangle stage, with markers denoting primary neuropathological diagnoses; the legend shows symbols for 11 major diagnostic groups. B, Plot of OI and SUVR by Braak tangle stage, with markers representing entorhinal cortex NFT levels based on autopsy reports; the legend indicates eight NFT severity levels. C, Visual comparison of tau PET OI and IHC/ARG for early Braak tangle stages (0–III). Four participants with Braak tangle stages 0 to III, as determined from post mortem reports, were included. IHC and ARG data were obtained from slices of the entorhinal cortex and anterior hippocampus (including the amygdala and entorhinal cortex). Tau PET OI images present coronal views corresponding to the anatomical locations of the IHC and ARG slices for both baseline and follow‐up scans, with the overlapped mask image for the region indicated by the purple box shown on the right. Red voxels in the overlapped image represent overlapping clusters with an OI intensity threshold of 1.3 in both baseline and follow‐up scans, corresponding to voxels used for OI calculation. Purple arrows in the IHC and overlapped images manually indicate corresponding locations. For each participant, meta‐ROI OI, OS, SUVR, secondary neuropathological diagnosis, Braak tangle stage, and scan‐to‐death interval (months) are reported. The meta‐ROI comprised the amygdala, fusiform gyrus, entorhinal cortex, parahippocampal gyrus, middle temporal gyrus, and inferior temporal gyrus. AD, Alzheimer's disease; AGD, argyrophilic grain disease; ARG, autoradiography; BLBD, brainstem Lewy body disease; CBD, corticobasal degeneration; DLBD, diffuse Lewy body disease; FTDP‐17, frontotemporal degeneration with Parkinsonism linked to chromosome 17; FTLD‐tau, frontotemporal lobar degeneration with tau pathology; GGT, globular glial tauopathy; IHC, immunohistochemistry; NFT, neurofibrillary tangle; OI, overlap index; OS, overlap size; PET, positron emission tomography; PSP, progressive supranuclear palsy; SUVR, standardized uptake value ratio. SUVR is displayed as SUVr within the figure panels.

3.5. Tau PET OI related to cognitive test performance

To assess the relationship between voxel‐wise tau PET metrics and cognitive performance, we examined the associations of meta‐ROI OI, OS, and SUVR with memory z scores, AVLT total score (avlt.sum), and MMSE. Marker symbols represent the 11 primary neuropathological diagnostic groups identified at autopsy. As shown in Figure 5A, all three tau PET measures were negatively associated with memory z scores, with OI (R ² = 0.26, p = 0.038) and OS (R ² = 0.31, p = 0.015) showing significant associations, while SUVR (R ² = 0.21, p = 0.055) trended toward significance. In Figure 5B, stronger associations were observed with avlt.sum, where OI (R ² = 0.41, p = 3 × 10^{− 4}), OS (R ² = 0.51, p = 1.8 × 10^{− 5}), and SUVR (R ² = 0.43, p = 1.5 × 10^{− 4}) each demonstrated robust correlations, with OS showing the strongest relationship. In addition, Figure S7 in supporting information illustrates the associations with global cognitive function as measured by MMSE. All three tau PET metrics were significantly correlated with MMSE scores: OI (R ² = 0.24, p = 0.0007), OS (R ² = 0.22, p = 0.0013), and SUVR (R ² = 0.27, p = 0.0003). These findings suggest that OI, OS, and MMSE are all associated with both domain‐specific memory performance and global cognitive function across a range of neuropathological diagnoses.

FIGURE 5.

Associations between cognitive scores and OI, OS, and SUVR. Cognitive testing was performed within 4 months of the follow‐up PET scan, and only participants with available cognitive test data were included in the analysis. A, Associations between memory domain z scores and OI, OS, and SUVR, with different markers representing each of the 11 primary neuropathological diagnostic groups. B, Associations between total learning score on the Auditory Verbal Learning Test (AVLT.sum) and OI, OS, and SUVR, with different markers indicating the 11 primary neuropathological diagnostic groups. AD, Alzheimer's disease; AGD, argyrophilic grain disease; ARG, autoradiography; BLBD, brainstem Lewy body disease; CBD, corticobasal degeneration; DLBD, diffuse Lewy body disease; FTDP‐17, frontotemporal degeneration with Parkinsonism linked to chromosome 17; FTLD‐tau, frontotemporal lobar degeneration with tau pathology; GGT, globular glial tauopathy; IHC, immunohistochemistry; NFT, neurofibrillary tangle; OI, overlap index; OS, overlap size; PET, positron emission tomography; PSP, progressive supranuclear palsy; SUVR, standardized uptake value ratio. SUVR is displayed as SUVr within the figure panels.

4. DISCUSSION

This study examined the relationship between tau PET OI and post mortem neuropathology. OI demonstrated greater sensitivity than SUVR in reflecting NFT severity within the ERC across AD subgroups and non‐AD tauopathies, particularly at early Braak stages. Overlapping tau voxels identified by OI spatially corresponded with tau deposition observed in IHC and ARG. OI also provided enhanced detection of tau signals in individuals with 4R tauopathies, revealing overlapped tau voxels not only within GM but also in selective WM regions, and both OI and OS were associated with cognitive performance measures.

An important finding was that OI detected tau signals in individuals across a range of neuropathological diagnoses, whereas SUVR failed to do so in several cases. For example, tau PET OI identified tau accumulation in DLBD as well as within the AD spectrum, including the AD (low) group. In addition, focal tau PET OI signals exhibited spatial concordance with tau deposition confirmed by IHC and ARG. These results indicate that OI leveraging spatial coherence enhances the identification of focal tau PET voxel regions that reflect underlying tau pathology. ¹⁰ These features of OI also underscore its greater capability in detecting early tau signal changes. In the comparison of tau PET OI and SUVR distributions across Braak tangle stages, SUVR alterations became more prominent at Braak stage IV and beyond, in line with previous findings. ⁷ In contrast, OI not only effectively distinguished tau accumulation in later Braak stages but also more reliably captured tau signal in the earlier stages. Specifically, OI detected localized, low‐level tau signal in participants spanning Braak stages I to IV, many of whom were diagnosed with non‐AD pathologies. Tau can be passively released from neurons containing dying or dead NFTs, ¹⁸ but it may also be transferred between cells or released into the extracellular space through neuronal activity, ¹⁹ , ²⁰ facilitating propagation to adjacent regions. ²¹ , ²² This process may be further accelerated in the presence of Aβ. ²³ , ²⁴ Tau pathology in the absence of amyloidosis remains incompletely understood but may hold clinical relevance. Prior work has shown that a subset (13%) of DLB cases exhibit amyloid‐negative, tau‐positive pathology at autopsy, suggesting that tau accumulation can occur independently of amyloid. ²⁵ Some evidence further indicates that tau pathology alone may represent a distinct co‐pathology that accrues in synergy with α‐synuclein–related neurodegeneration. ²⁶ These observations support the value of detecting tau pathology even in the absence of amyloid. Accordingly, a positive tau PET OI finding at early stages does not necessarily indicate AD and may reflect presymptomatic AD, age‐related tau changes, secondary tau pathology, or non‐AD tauopathies. While OI shows promise as a sensitive research tool for early tau detection, its clinical diagnostic utility may be most impactful in the context of future tau‐targeted therapies. Spatial correspondence of tau signal was further validated by the tau deposition observed in post mortem IHC and ARG assessments, but these analyses did show incomplete colocalization with neuropathologic findings which may be partially attributed to the time delay between tau PET and death. These findings highlight the advantage of OI in detecting early tau‐related changes and reinforce its utility as reported in earlier work. ¹⁰ While conventional ROI‐based SUVR methods may have limited ability to reflect subtle tau PET signals during the initial phase of deposition which is probably reflecting diminished signal‐to‐noise ratio when a small volume of true radiotracer binding is present ⁷ , ¹⁰ , ²⁷ —due to reliance on median signal values within the region ²⁸ , ²⁹ —they become very valuable in more advanced tauopathy stages with high tracer uptake. Thus, both OI and standard ROI‐based metrics serve complementary roles in interpreting early and late tau PET signal patterns, and a combined approach may provide a more comprehensive representation of tau pathology, supporting the methodological incorporation of both strategies in future tau PET studies. 

A notable observation is that OI identified overlapping tau voxels not only in GM regions but also in discrete WM areas in individuals with 4R tauopathies, including FTLD‐tau with CBD, CBD, and PSP cases in this study. In the FTLD‐tau with CBD cases, these overlapping voxels were localized to the superior/middle temporal WM and exhibited spatial correspondence with tau deposition confirmed by IHC and ARG findings. Multiple studies have documented WM tau deposition in patients with CBD, as evidenced by tau PET imaging and post mortem tissue‐based staining methods. ³⁰ , ³¹ , ³² , ³³ , ³⁴ In a tau PET study, Cho et al. reported increased tau signal in localized regions of the precentral gray and WM through voxel‐wise comparisons of FTP PET in corticobasal syndrome (CBS) patients. ³² Smith et al. demonstrated elevated FTP PET signal in the subcortical WM underlying the motor cortex based on group comparisons among CBS, AD, and PSP cohorts. ³³ Tsai et al. observed increased FTP binding in the frontal WM of individuals with CBS. ³⁴ Using ARG and IHC, Lowe et al. confirmed tau deposition in the WM of the rostral substantia nigra and substantia nigra in CBD cases. ⁵ Vega et al. identified disease‐specific tau aggregate morphologies in the frontal WM across CBD, PSP, and AD, highlighting regional heterogeneity in tau burden. ³¹ In PSP, tau pathology in WM was further characterized by Vega et al., while subtle tau PET SUVR signals were observed across both GM and WM in a PSP cohort by Whitwell et al. ³¹ , ³⁵ The finding in this study of the detection of overlapping tau voxels within GM and distinct WM regions in 4R tauopathy cases aligns with these prior reports of WM tau pathology. However, it is important to acknowledge that an elevated FTP signal in 4R tauopathies may not necessarily reflect specific 4R tau binding. Prior studies ³¹ , ³² , ³³ , ³⁴ , ³⁵ , ³⁶ have shown that subtle, regionally specific FTP elevations can also occur in other degenerative diseases, including those without tau isoforms, suggesting that this signal may partially reflect non‐specific or degeneration‐associated FTP binding rather than true 4R tau pathology. In this context, OI may enhance the sensitivity to such marginal FTP signals, which could represent either low‐level tau accumulation or non‐specific degenerative uptake in 4R tauopathies. Recognizing both possibilities provides a more balanced interpretation of the observed OI findings.

From a clinical perspective, meta‐ROI OI and meta‐ROI OS demonstrated significant correlations with cognitive measures, including AVLT total scores, memory z scores, and MMSE. These results support the finding that tau PET signal in the medial temporal lobe was associated with memory decline ¹⁷ and suggest that spatial tau distribution indices are also associated with memory decline.

4.1. Strengths and limitations

This study represents the first direct comparison of tau PET OI to neuropathological findings, supporting the utility of spatial coherence in enhancing the identification of PET voxels that correspond to underlying tau deposition confirmed at autopsy. Tau PET OI demonstrated greater sensitivity than conventional SUVR in detecting tau signal across both AD and non‐AD tauopathies, including Lewy body disease, AGD, CBD, and PSP. OI identified tau accumulation within both cortical GM and discrete WM regions, particularly in 4R tauopathies. Overlapping voxels derived from OI showed high spatial concordance with tau deposition, confirmed by IHC and ARG, across multiple diagnostic groups. Furthermore, OI was able to detect early tau signal at lower Braak stages, where SUVR exhibited limited sensitivity, underscoring its advantage in capturing spatially coherent and low‐intensity tau signal during early pathological progression.

Despite the strengths of this study, several limitations warrant consideration. First, serial scanning is required to perform OI. This means that participants must undergo two or more PET scans to obtain OI. Further studies are needed to assess the feasibility and performance of OI when applied to single PET frames based on dynamic PET imaging datasets. Second, variability in the scan‐to‐death interval—averaging 26.4 ± 16.1 months (range: 3–81 months)—introduces potential uncertainty in interpreting PET‐to‐autopsy correspondence. Third, the relatively modest sample size (n = 57) reflects the inherent rarity of cohorts with both serial PET imaging and subsequent autopsy, which may limit generalizability. Fourth, the absence of a neuropathologically verified cognitively unimpaired control group constrains the ability to fully contextualize OI performance in cognitively and pathologically normal individuals. Fifth, the determination of primary versus secondary neuropathological diagnoses was based on semi‐structured autopsy reports, which may introduce subjectivity or inter‐rater variability. Last, although a uniform diagnostic framework was applied, inter‐institutional variability in neuropathological assessment protocols may introduce inconsistency across diagnostic classifications.

5. SUMMARY

The tau PET OI quantifies the spatial consistency of elevated SUVR voxels across two serial tau PET scans and is applicable to longitudinal imaging. This study demonstrates the utility of tau PET OI for detecting early and spatially localized tau pathology across AD and non‐AD tauopathies, including early Braak tangle stages. OI effectively identified tau‐related signal in GM and selected WM regions in individuals with 4R tauopathies, showing strong spatial concordance with post mortem IHC and ARG findings. The regional agreement with neuropathological features suggests that OI serves as a sensitive marker of disease‐associated neurodegenerative signal. In addition, OI and OS were associated with cognitive performance, supporting their pathological and clinical relevance.

CONFLICT OF INTEREST STATEMENT

Seokbeen Lim, Jeyeon Lee, Paul H. Min, Christina M. Moloney, Carly T. Mester, Sujala Ghatamaneni, Matthew L. Senjem, Hugo Botha, Stuart J. McCarter, Vijay K. Ramanan, Rodolfo Savica, Julie A. Fields, Mary M. Machulda, Dennis W. Dickson, R. Ross Reichard, Aivi T. Nguyen, Christopher G. Schwarz, Jeffrey L. Gunter, Kejal Kantarci, Prashanthi Vemuri¹, and David T. Jones report no disclosures relevant to the manuscript. Jonathan Graff‐Radford serves on a data safety monitoring board (DSMB) for strokeNET. He is a site investigator for trials sponsored by Eisai and Cognition therapeutics. He has received an honorarium from the American Academy of Neurology for being a course director and from IMPACT‐AD for serving as faculty. David S. Knopman serves on a DSMB for the DIAN study; has served on a DSMB for a tau therapeutic for Biogen but received no personal compensation; is a site investigator in Biogen aducanumab trials; is an investigator in clinical trials sponsored by Lilly Pharmaceuticals and the University of Southern California; serves as a consultant for Samus Therapeutics, Roche, Magellan Health, and Alzeca Biosciences but receives no personal compensation; and receives research support from the NIH. Bradley Boeve receives honoraria for SAB activities for the Tau Consortium; is a site investigator for clinical trials sponsored by Alector, Cognition Therapeutics, EIP Pharma, and Transposon; and receives research support from NIH. Clifford R. Jack Jr. has consulted for Lily and serves on an independent data monitoring board for Roche and as a speaker for Eisai, but he receives no personal compensation from any commercial entity. He receives research support from NIH and the Alexander Family Alzheimer's Disease Research Professorship of the Mayo Clinic. Ronald C. Petersen serves as a consultant for Roche, Inc., Merck, Inc., Biogen, Inc., Eisai, Inc., Genentech, Inc., and Nestle, Inc.; served on a DSMB for Genentech; receives royalties from Oxford University Press and UpToDate; and receives NIH funding. Melissa E. Murray served as a consultant for AVID Radiopharmaceuticals, received grant funding from Eli Lilly and Company and the Rainwater Charitable Foundation, and is supported by the NIH (NIA). Val J. Lowe serves as a consultant for Bayer Schering Pharma, Piramal Life Sciences, Life Molecular Imaging, Eisai Inc., AVID Radiopharmaceuticals, Eli Lilly and Company, PeerView Institute for Medical Education, and Merck Research and receives research support from GE Healthcare, Siemens Molecular Imaging, AVID Radiopharmaceuticals, and the NIH (NIA, NCI). Author disclosures are available in the supporting information.

CONSENT STATEMENT

All participants provide informed consent.

Supporting information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information