Skip to content

Imaging and neuropathology

TL;DR — Amyloid PET began with Pittsburgh Compound-B, which showed 1.5–1.9-fold higher cortical retention in mild AD than controls and an inverse relationship with glucose metabolism strongest in parietal cortex (r = −0.72, P=0.0001) (Klunk 2004, PMID 14991808), and was validated against autopsy: florbetapir visual reads agreed with post-mortem amyloid status in 96% of 29 end-of-life participants (Clark 2011, PMID 21245183). Tau PET was validated the same way, with flortaucipir predicting Braak V–VI pathology at sensitivity 92.3–100% but specificity of only 52.0–92.0% across five independent readers (Fleisher 2020, PMID 32338734) — a sensitivity/specificity asymmetry that matters when the scan gates treatment. Quantitative comparability required a common unit: the Centiloid scale anchors any tracer and analysis pipeline to 0 (young controls ≤45 years) and 100 (typical AD) (Klunk 2015, PMID 25443857). Neuropathology has its own standard, the NIA-AA "ABC" score combining Thal amyloid phase (A), Braak tangle stage (B) and CERAD neuritic plaque score (C), explicitly allowing AD neuropathologic change without cognitive impairment and requiring assessment of Lewy body, vascular, hippocampal sclerosis and TDP-43 copathology (Montine 2012, PMID 22101365; Hyman 2012, PMID 22265587). Amyloid PET changes management — 60.2% of MCI and 63.5% of dementia cases in IDEAS, with the etiologic diagnosis switching away from AD in 25.1% and towards it in 10.5% (Rabinovici 2019, PMID 30938796) — but did not reduce 12-month hospitalisations or ED visits versus matched controls (Rabinovici 2023, PMID 37812437).

Amyloid PET

Milestone Finding
First human amyloid tracer (PiB) 16 patients with mild AD vs 9 controls: retention increased in frontal (1.94-fold, P=0.0001), parietal (1.71-fold, P=0.0002), temporal (1.52-fold, P=0.002), occipital (1.54-fold, P=0.002) cortex and striatum (1.76-fold, P=0.0001); equivalent in white matter, pons and cerebellum; low in both young (21 y) and older (69.5 ± 11 y) controls; inverse correlation with FDG uptake, strongest in parietal cortex (r = −0.72, P=0.0001) (Klunk 2004, PMID 14991808)
Autopsy validation of an ¹⁸F tracer Florbetapir PET a mean 99 days (range 1–377) before death in 29 end-of-life participants: visual read and quantitative cortical uptake correlated with immunohistochemical amyloid (ρ 0.78, 95% CI 0.58–0.89, P<0.001) and silver-stain neuritic plaque score (ρ 0.71, 0.47–0.86, P<0.001); binary agreement 96%; all 74 younger participants (18–50 y) read as negative (Clark 2011, PMID 21245183)
Quantitative standardisation The Centiloid scale rescales any tracer/pipeline to 0 (young controls ≤45 y) and 100 (typical AD), with a defined "standard" PiB method and a procedure for calibrating non-standard methods (Klunk 2015, PMID 25443857)

Two structural limits follow. First, PiB's inverse relation to FDG uptake means amyloid and hypometabolism are anti-correlated regionally, which is exactly the opposite of the tau–FDG relationship (see below) — amyloid does not mark the tissue that is failing. Second, Centiloid change is now the standard readout of anti-amyloid target engagement, and its interpretation was challenged by autopsy of aducanumab-treated patients showing clearance restricted to cortical layer I despite Centiloid reductions of −6% to −81% (Boon 2025, PMID 41109234). See amyloid biology.

Tau PET

Flortaucipir was validated against autopsy in a terminal-illness cohort at 28 sites: among 64 primary-cohort patients (77% with dementia, 22% cognitively normal), visual reads predicted a B3 (Braak V–VI) tau pattern with sensitivity 92.3% (95% CI 79.7–97.3) to 100.0% (91.0–100.0) and specificity 52.0% (33.5–70.0) to 92.0% (75.0–97.8) across five readers; for high AD neuropathologic change, sensitivity 94.7–100.0% and specificity 50.0–92.3%. Prespecified success criteria were met and replicated in a second reader study (Fleisher 2020, PMID 32338734).

The reader-dependent specificity is the operational weakness. A modality with 95% sensitivity and 50% specificity in the hands of some readers is a good rule-out and a poor rule-in — relevant because tau PET is being proposed as a prognostic gate, and because 9.8% of cognitively unimpaired people in a 21-cohort pooled analysis were tau-PET positive (Moscoso 2025, PMID 40522652).

Tau PET's distinctive value is topographic. It mirrors clinical phenotype (posterior in PCA, left-lateralised in logopenic aphasia, medial temporal in amnestic presentations), correlates strongly and negatively with FDG uptake (r = −0.49 ± 0.07, P<0.001) where amyloid does not (r = 0.16 ± 0.09), and predicts the location of future atrophy at the single-patient level over 15 months where amyloid PET does not (Ossenkoppele 2016, PMID 26962052; La Joie 2020, PMID 31894103). Fuller treatment on tau biology and spread.

Structural and metabolic MRI/PET

Structural MRI contributes in three ways: excluding structural mimics (subdural collection, tumour, normal-pressure hydrocephalus), quantifying the AD atrophy signature, and — mandatorily in the treatment era — establishing baseline microhaemorrhage and siderosis burden before anti-amyloid therapy and monitoring for ARIA.

The AD cortical signature is a set of limbic and association-cortex regions that thin early. Measured in 49 people with CDR 0.5, an aggregate thickness measure of nine such regions predicted progression to mild AD dementia over ~2.5 years with 83% sensitivity and 65% specificity, and CDR sum-of-boxes correlated with temporal and parietal thickness at baseline (Bakkour 2009, PMID 19109536). This is respectable prognostic performance for a structural measure and poor compared with what fluid biomarkers now achieve; MRI's contemporary role is anatomical and safety-related rather than diagnostic.

FDG-PET shows temporoparietal and posterior cingulate hypometabolism in AD. Its quantitative relationship to the molecular markers is informative: hypometabolism tracks tau closely and amyloid weakly (PMID 26962052), and the original PiB study already found the amyloid–FDG relationship to be inverse (PMID 14991808).

Does imaging change anything?

Question Study Answer
Does amyloid PET change management? IDEAS, 11,409 Medicare beneficiaries with MCI or dementia of uncertain aetiology meeting appropriate-use criteria, assessed by 946 dementia specialists at 595 sites Composite management change in 60.2% (95% CI 59.1–61.4) of MCI and 63.5% (62.1–64.9) of dementia cases, both exceeding the prespecified 30% threshold (P<0.001). Etiologic diagnosis changed from AD to non-AD in 25.1% (24.3–25.9) and from non-AD to AD in 10.5% (10.0–11.1). Amyloid PET was positive in 55.3% of MCI and 70.1% of dementia (Rabinovici 2019, PMID 30938796)
Does it change outcomes? IDEAS participants (n=12,684) matched to Medicare controls who had not had amyloid PET 12-month hospitalisation 24.0% vs 25.1% (relative reduction −4.49%, 97.5% CI −9.09 to 0.34); ED visits 44.8% in both (−0.12%, −3.19 to 3.05). Both fell short of the prespecified ≥10% relative reduction. Hospitalisation was lower among amyloid-positive than amyloid-negative participants (21.4% vs 25.7%; adjusted OR 0.83, 95% CI 0.78–0.89) (Rabinovici 2023, PMID 37812437)

The pair is instructive: a diagnostic test can substantially change what clinicians do without changing what happens to patients, at least on health-care-utilisation endpoints over one year, in an era before disease-modifying treatment was available. The second study is non-randomised with matched controls, so residual confounding is unresolved; the lower hospitalisation among amyloid-positive participants most likely reflects that amyloid-negative patients have other, more acutely destabilising illnesses.

Appropriate use criteria

The 2013 amyloid-PET criteria restricted appropriate use to patients with objective cognitive impairment where AD was a diagnostic consideration but the aetiology uncertain, and explicitly listed inappropriate uses including asymptomatic people, determining severity, and non-medical purposes (Johnson 2013, PMID 23359661). The 2025 update covers both amyloid and tau PET across 17 clinical scenarios rated by modified Delphi: for amyloid PET, 7 appropriate, 2 uncertain, 8 rarely appropriate; for tau PET, 5 appropriate, 6 uncertain, 6 rarely appropriate (Rabinovici 2025, PMID 39776249). The larger "uncertain" fraction for tau PET is an accurate reflection of where the evidence stands. Fuller guideline treatment on guidelines.

Neuropathology: the reference standard

The NIA-AA neuropathologic guidelines replaced a diagnosis-by-plaque-density approach with an ABC score (Montine 2012, PMID 22101365; Hyman 2012, PMID 22265587):

Component Scheme Source
A — Aβ/amyloid plaque distribution Thal phase 0–5: neocortex → allocortex → diencephalon/striatum/basal forebrain → brainstem → cerebellum Thal 2002, PMID 12084879
B — Neurofibrillary tangle stage Braak stage I–VI: transentorhinal → limbic → isocortical Braak 1991, PMID 1759558
C — Neuritic plaque score CERAD (none / sparse / moderate / frequent) —

The three major revisions the guidelines introduced are, in the authors' own summary: recognition that AD neuropathologic change may occur in the apparent absence of cognitive impairment; the ABC composite; and structured assessment of commonly comorbid Lewy body disease, vascular brain injury, hippocampal sclerosis and TDP-43 inclusions, with recommendations on minimum brain sampling, staining methods and reporting (PMID 22101365).

That third revision is the reason this repository treats copathology as content rather than noise. Clinicopathological synthesis concludes that cognitive severity correlates best with neocortical tangle burden rather than plaque burden (Nelson 2012, PMID 22487856), and LATE neuropathologic change — a TDP-43 proteinopathy with an amnestic phenotype mimicking AD — is present at cognitively relevant levels in roughly 25% of community autopsy brains (Nelson 2019, PMID 31039256). See vascular and metabolic contributions.

Diagnostic accuracy, prognosis and management are three different claims

For short-horizon prognosis in MCI, tau PET currently adds more than structural MRI or amyloid PET. Across 448 people (331 discovery, 117 external validation) followed for a mean 2.0 years, adding tau PET to a base model of age, sex, education and MMSE improved discovery-cohort prediction of all-cause dementia from AUC 0.71 (95% CI 0.65–0.77) to 0.75 (0.70–0.80), and of AD dementia from 0.75 (0.69–0.82) to 0.84 (0.79–0.89); amyloid-PET Centiloids also improved AD-dementia prediction (0.83) but only tau PET replicated for both outcomes in the validation cohort (Groot 2024, PMID 38857029). The increment for all-cause dementia is small, so prognostic value should not be conflated with diagnostic spectacle.

Functional imaging comparisons also depend on the task. Across seven direct-comparison studies, FDG-PET and ASL-MRI had similar overall AUCs (0.864 versus 0.836), but in the MCI subgroup FDG-PET had sensitivity 0.90 versus 0.75, specificity 0.91 versus 0.73 and AUC 0.92 versus 0.80 (P<0.001) (Radmard 2025, PMID 40898829). Across 38 AI-imaging studies, of which 28 moderate-to-high-quality studies were pooled, SROC-AUC was 0.96 (95% CI 0.94–0.98) for FDG-PET and 0.94 (0.92–0.96) for structural MRI (P=0.02); validation strategy and algorithm type were important sources of heterogeneity (Wang 2025, PMID 41061249). These are classification studies, not proof that AI improves patient outcomes.

Management change is a third endpoint. New IDEAS found diagnosis, medication or counselling changed in 59.0% (95% CI 57.6–60.5) of 5,757 diverse Medicare beneficiaries after amyloid PET, including more than 30% in every ethnoracial and typical/atypical subgroup (Windon 2025, PMID 40728069). Neither that association nor an earlier 30-patient case series in which PET revised 10 diagnoses and clarified nine demonstrates better cognition or function (Mitsis 2014, PMID 24484858). Imaging can be analytically accurate, prognostically informative and management-changing while clinical-outcome utility remains unproven.

How the modalities map onto each other

Layer In-vivo measure Pathological correlate Validation quality
Amyloid deposition Amyloid PET (Centiloid) Thal phase / CERAD plaques Autopsy-validated; 96% binary agreement (PMID 21245183)
Tangle burden and topography Tau PET Braak stage Autopsy-validated with high sensitivity, reader-variable specificity (PMID 32338734)
Neurodegeneration Structural MRI, FDG-PET Neuronal and synaptic loss Non-specific; signature thickness 83% sens / 65% spec for progression (PMID 19109536)
Amyloid + tau, cheaply Plasma p-tau217 and ratios Alzheimer's neuropathologic change See fluid biomarkers
Copathology Largely unmeasured in life Lewy bodies, TDP-43/LATE, vascular injury, hippocampal sclerosis No validated in-vivo marker for LATE (PMID 31039256)

The last row is the field's largest measurement gap: the pathologies that account for a large share of person-specific cognitive loss have no in-vivo marker at all. Plasma assays now compete with PET for the amyloid and tau rows — mass-spec %p-tau217 is clinically equivalent to FDA-approved CSF tests for Aβ-PET and generally superior for tau-PET (Barthélemy 2024, PMID 38382645), and a %p-tau217 plus eMTBR-tau243 model reached C-index 0.91 versus PET-based AA stages (Salvadó 2026, PMID 42189519) — which is why the 2025 PET appropriate-use criteria have to be read against a changing comparator, not against 2013 practice. See fluid biomarkers.

Open questions

  • Can tau-PET specificity be made reader-independent, given the 52–92% range across five readers against the same autopsies (Fleisher 2020, PMID 32338734)?
  • Does Centiloid change measure parenchymal amyloid removal, given layer-I-restricted clearance at large Centiloid reductions (Boon 2025, PMID 41109234; Klunk 2015, PMID 25443857)?
  • Would amyloid PET change outcomes rather than only management now that a disease-modifying treatment exists — IDEAS was conducted before that era (Rabinovici 2019, PMID 30938796; Rabinovici 2023, PMID 37812437)?
  • What is the role of structural MRI when a blood test outperforms it diagnostically — is it now purely a safety and mimic-exclusion modality (Bakkour 2009, PMID 19109536)?
  • Can an in-vivo marker of LATE-NC be developed, and how much apparent AD would it reclassify (Nelson 2019, PMID 31039256)?
  • With tau PET rated "uncertain" in 6 of 17 scenarios, what evidence would move those scenarios to appropriate (Rabinovici 2025, PMID 39776249)?

References

  1. Klunk WE, et al. Imaging brain amyloid in Alzheimer's disease with Pittsburgh Compound-B. Ann Neurol. 2004;55:306-19. PMID 14991808.
  2. Klunk WE, et al. The Centiloid Project: standardizing quantitative amyloid plaque estimation by PET. Alzheimers Dement. 2015;11:1-15. PMID 25443857.
  3. Clark CM, et al. Use of florbetapir-PET for imaging beta-amyloid pathology. JAMA. 2011;305:275-83. PMID 21245183.
  4. Fleisher AS, et al. Positron emission tomography imaging with [18F]flortaucipir and postmortem assessment of Alzheimer disease neuropathologic changes. JAMA Neurol. 2020;77:829-839. PMID 32338734.
  5. Montine TJ, et al. National Institute on Aging-Alzheimer's Association guidelines for the neuropathologic assessment of Alzheimer's disease: a practical approach. Acta Neuropathol. 2012;123:1-11. PMID 22101365.
  6. Hyman BT, et al. National Institute on Aging-Alzheimer's Association guidelines for the neuropathologic assessment of Alzheimer's disease. Alzheimers Dement. 2012;8:1-13. PMID 22265587.
  7. Rabinovici GD, et al. Association of amyloid positron emission tomography with subsequent change in clinical management among Medicare beneficiaries with mild cognitive impairment or dementia. JAMA. 2019;321:1286-1294. PMID 30938796.
  8. Rabinovici GD, et al. Amyloid positron emission tomography and subsequent health care use among Medicare beneficiaries with mild cognitive impairment or dementia. JAMA Neurol. 2023;80:1166-1173. PMID 37812437.
  9. Rabinovici GD, et al. Updated appropriate use criteria for amyloid and tau PET: a report from the Alzheimer's Association and Society for Nuclear Medicine and Molecular Imaging Workgroup. Alzheimers Dement. 2025;21:e14338. PMID 39776249.
  10. Johnson KA, et al. Appropriate use criteria for amyloid PET: a report of the Amyloid Imaging Task Force, the Society of Nuclear Medicine and Molecular Imaging, and the Alzheimer's Association. J Nucl Med. 2013;54:476-90. PMID 23359661.
  11. Bakkour A, et al. The cortical signature of prodromal AD: regional thinning predicts mild AD dementia. Neurology. 2009;72:1048-55. PMID 19109536.
  12. Braak H, et al. Neuropathological stageing of Alzheimer-related changes. Acta Neuropathol. 1991;82:239-59. PMID 1759558.
  13. Thal DR, et al. Phases of Aβ-deposition in the human brain and its relevance for the development of AD. Neurology. 2002;58:1791-800. PMID 12084879.
  14. Moscoso A, et al. Frequency and clinical outcomes associated with tau positron emission tomography positivity. JAMA. 2025;334:229-242. PMID 40522652.
  15. Ossenkoppele R, et al. Tau PET patterns mirror clinical and neuroanatomical variability in Alzheimer's disease. Brain. 2016;139:1551-67. PMID 26962052.
  16. La Joie R, et al. Prospective longitudinal atrophy in Alzheimer's disease correlates with the intensity and topography of baseline tau-PET. Sci Transl Med. 2020;12. PMID 31894103.
  17. Boon BDC, et al. Neuropathological changes and amyloid-related imaging abnormalities in Alzheimer's disease treated with aducanumab versus untreated: a retrospective case-control study. Lancet Neurol. 2025;24:931-944. PMID 41109234.
  18. Nelson PT, et al. Correlation of Alzheimer disease neuropathologic changes with cognitive status: a review of the literature. J Neuropathol Exp Neurol. 2012;71:362-81. PMID 22487856.
  19. Nelson PT, et al. Limbic-predominant age-related TDP-43 encephalopathy (LATE): consensus working group report. Brain. 2019;142:1503-1527. PMID 31039256.
  20. Barthélemy NR, et al. Highly accurate blood test for Alzheimer's disease is similar or superior to clinical cerebrospinal fluid tests. Nat Med. 2024;30:1085-1095. PMID 38382645.
  21. Salvadó G, et al. Plasma eMTBR-tau243 and %p-tau217 for biological staging of Alzheimer disease. JAMA Neurol. 2026;83:654-666. PMID 42189519.
  22. Groot C, et al. Tau positron emission tomography for predicting dementia in individuals with mild cognitive impairment. JAMA Neurol. 2024;81:845-856. PMID 38857029.
  23. Radmard M, et al. Functional imaging in Alzheimer's disease: a systematic review and meta-analysis of ASL-MRI and FDG-PET. J Neuroimaging. 2025;35:e70080. PMID 40898829.
  24. Wang B, et al. Comparative diagnostic accuracy of AI-assisted fluorine-18 fluorodeoxyglucose positron emission tomography versus structural magnetic resonance imaging in Alzheimer disease. JMIR Aging. 2025;8:e76981. PMID 41061249.
  25. Windon CC, et al. Amyloid PET and clinical management in a diverse, cognitively impaired population: the New IDEAS Study. Alzheimers Dement. 2025;21:e70504. PMID 40728069.
  26. Mitsis EM, et al. A consecutive case series experience with florbetapir PET imaging in an urban dementia center. Mol Neurodegener. 2014;9:10. PMID 24484858.