Compliance with Ethical Standards
Ethics approval was obtained from the Austin Health Human Research Ethics Committee (reference LNR/17/405).
Subject identification and demographic data collection
Fifty-one subjects in total with various ante-mortem and post-mortem diagnoses were retrospectively identified by cross-referencing the databases of the Austin Health Molecular Imaging Dementia Research group, the Sydney Brain Bank, and the Victorian Brain Bank. These subjects, with prior informed consent, had undergone both an amyloid PET scan at Austin Health and post-mortem neuropathologic brain evaluation in Melbourne or Sydney between all years recorded in the database (2004 to 2017). Exclusion criteria for these prior studies were history of stroke, significant medical illness, recent cancer, and substance use disorder. Two of the fifty-one cases were excluded due to a diagnosis of familial AD, because different neuropathological processes in this condition, such as a significantly greater density of Aβ plaques in the cerebellum than in sporadic AD [24, 25], may have confounded amyloid PET quantification and interpretation. Data has been published on a proportion of the cases [26, 27].
Amyloid PET Imaging and Centiloid determination
Aβ imaging was performed with either 11C-PiB or 18F-florbetaben (FBB). The methodology for PET imaging with these tracers has been previously described [28, 29]. A 20-minute acquisition was commenced 50 minutes post-injection of 11C-PiB or 90 minutes post-injection of FBB. A transmission scan was performed for attenuation correction. PET images were reconstructed using a 3D row-action maximum likelihood algorithm (RAMLA). The standard Centiloid cortical and whole cerebellar volumes of interest template were applied to the summed and spatially normalized PET images in order to obtain standardized uptake value ratios (SUVR). For this study we used the CapAIBL software package, which when compared to standard approach, has the benefit of not requiring a corresponding MRI to quantify the PET scan [30, 31]. This package has been validated against the standard Centiloid method that uses the public domain software program SPM8 to spatially normalize each subject’s MRI and then apply those parameters to spatially normalize the amyloid PET scan . The SUVR were transformed into Centiloid units by linear transformation using the PET tracer specific equations published for conversion of Centiloid method SUVR to Centiloid units with a minor correction applied for the CapAIBL registration [13, 28, 29, 32].
Neuropathological evaluation was performed at the Victorian Brain Bank (Melbourne, Australia) and Sydney Brain Bank (Neuroscience Research Australia, Sydney, Australia) to determine a global C score from inferior temporal regions of fixed brain hemispheres based on the Consortium for Establish a Registry for Alzheimer’s Disease (CERAD) neuropathologic assessment guidelines . Frequency of neuritic plaques per 100x microscopic field were categorised as none, sparse, moderate or frequent with corresponding C scores of 0, 1, 2 or 3 respectively, as described in published guidelines . ADNC classification was also obtained as defined by the NIA-AA 2012 criteria . The ADNC rating uses the Thal amyloid plaque distribution, the Braak neurofibrillary tangle stage, and the CERAD neuritic plaque score, to classify AD neuropathologic change as not, low, intermediate or high.
One amyloid PET expert reader (author CR), blinded to CL values and neuropathological data, visually interpreted all scans using MedView v12 software, viewing images in greyscale and rainbow colour scale. The method used to visually read amyloid PET has been previously described . Scans were classified positive when cortical activity was equal to or greater than white matter activity in one or more lobes.
The clinicopathological diagnosis for each case factored in both neuropathological assessment and antemortem clinical diagnosis. Neuropathological diagnosis was made in accordance with published guidelines , and included morphological examination with immunohistochemistry analyses for Aβ, tau, TDP43, and alpha-synuclein in several brain regions. There were 17 AD and 32 non-AD cases. Non AD cases included diagnoses of frontotemporal dementia (n=12), normal controls (n=3), dementia with Lewy bodies (n=3), Parkinson’s disease dementia with concurrent diffuse Lewy bodies (n=3), hippocampal sclerosis (n=2), Creutzfeldt-Jakob disease (n=2), progressive supranuclear palsy (n=2), motor neuron disease (n=1), hippocampal ischaemia (n=1), corticobasal degeneration (n=1), multisystem atrophy (n=1), and a case of mixed AD and dementia with Lewy bodies (n =1). This last case was included in all analyses, except for the ‘Centiloid results in clinicopathological AD diagnosis’ analysis.
Three aspects of CL performance were investigated. Firstly, CL values were compared with dichotomized neuropathological C score categories using two different approaches: “high vs low” plaque density (“high” = moderate and frequent, and “low” = none and sparse), and; “any vs none” (“any” = sparse, moderate and frequent, and “none” = none). A Youden Index  was used to determine the optimal CL thresholds from receiver operator characteristic curves. CL values were also compared with binary ADNC classification of “unlikely AD” (ADNC scores of not or low) vs “likely AD” (ADNC scores of intermediate or high). Secondly, values were compared with visual read (positive or negative). Thirdly, CL values were compared with cases of AD as determined by clinicopathological diagnosis using descriptive statistics. To assess for the contribution of interval from PET scan to time to death, analyses were repeated using adjusted CL values, after applying a sigmoidal adjustment derived from our previous work . We derived that CL increases very slowly below 20 CL, then accelerates to a maximum of 5 CL increase per year for the almost linear mid-section (40 to 110 CL) of the sigmoid curve that best describes amyloid accumulation over time, before slowing again at higher CL values. Consequently, each individual CL value was adjusted to that expected at the time of death based on the average rate of increase for the CL level at the time of the scan and the duration between the scan and post mortem examination.