Welcome to the AD/PD™ 2022 Interactive Program
The conference will officially run on Central European Time (CET) - Barcelona Time
To convert the conference times to your local time Click Here
CLASSIFICATION OF DIFFERENT CAUSES OF DEMENTIA USING DEEP LEARNING TECHNIQUES ON EEG MEASUREMENTS
Abstract
Aims
Dementia is a neurocognitive syndrome that is caused by a variety of brain diseases, among which Alzheimer's disease (AD) is the most common. Other dementia types are vascular dementia (VAD), diffuse Lewy body dementia (DLB), frontotemporal lobe dementia (FTLD), Creutzfeldt Jakob disease (CJD), progressive supranuclear palsy (PSP), and mixed dementia (MXD). Electroencephalography (EEG) biomarkers are used to differentiate these dementia types, although often with limited success. The aim of this study is to investigate whether deep learning techniques can be used on EEG to create a classifier that can correctly diagnose the type of dementia.
Methods
The EEG data for this work includes measurements of 53 AD, 6 VAD, 11 DLB, 19 FTLD, 4 CJD, 2 PSP, and 7 MXD patients, together with the data from 83 healthy controls (HC). Deep learning is applied on the preprocessed epochs to build the classification model. The evaluation is done using holdout for the train-test split (50-33), and cross-validation for the validation set. The evaluation metrics are the AUPRC, accuracy, recall, and F1-score.
Results
The multi-class classification between AD, CJD, DLB, FTLD, MXD, and HC patients resulted in an AUPRC, accuracy, recall, and F1-score of respectively 85.3%, 90.9%, 65.1%, and 62.1%. The Dementia vs. No Dementia classification of this model resulted in a perfect classification of the 17 dementia patients and 16 HC of the test set.
Conclusions
The different type of dementia can be classified with a high accuracy using EEG deep learning.
EFFECTS OF AGE ON REGIONAL R1 MAY INDICATE DIFFERENTIAL CORTICAL MYELINATION IN DEMENTIA AND MCI COMPARED TO COGNITIVELY UNIMPAIRED INDIVIDUALS
Abstract
Aims
Alzheimer’s disease (AD) dementia risk increases with age; prior studies suggest that age-associated myelin degeneration may underlie the regional patterns of AD pathology that develop in AD. Quantitative R1 is highly sensitive to myelin content, as well as other tissue microstructures. Here we tested whether age-associated regional R1 differed among individuals with and without clinical diagnosis of mild cognitive impairment (MCI) and dementia.
Methods
Participants(N=341) underwent MPnRAGE MRI and cognitive assessment for diagnosis of unimpaired(N=291), MCI(N=27) or dementia(N=23). MPnRAGE-derived T1-weighted images and R1 maps were processed using FreeSurfer to estimate the pial and white matter surfaces (Figure 1). Mean R1 was extracted for each region of interest (ROI) in the Desikan-Killiany atlas. Multiple linear regression models were employed to test age-by-diagnosis interactions on R1 for each ROI, controlling for sex and APOE status.
Results
Participants with dementia showed significantly lower R1 with older age in the bilateral cuneus, precuneus, and superior parietal lobules as compared to MCI and unimpaired individuals. In entorhinal cortices, age-related increases in R1 were observed among individuals with dementia compared to MCI and unimpaired individuals. Differences in age-related R1 were also observed in the banks of the superior temporal sulci, whereby participants with MCI showed higher myelin content with older age (Figure 2) as compared to dementia and unimpaired individuals.
Conclusions
Brain regions typically affected by AD-related pathology showed significantly different age-related associations with cortical R1 between diagnostic groups, which could indicate differential myelination patterns. Future analyses will examine longitudinal trajectories of R1 in association with regional AD pathology.
CORTICAL NETWORK MODULARITY CHANGES ALONG THE COURSE OF FRONTOTEMPORAL AND ALZHEIMER’S DEMENTING DISEASES
Abstract
Aims
Cortical network modularity underpins cognitive functions, so we hypothesized its progressive derangement along the course of frontotemporal (FTD) and Alzheimer’s (AD) diseases.
Methods
EEG was recorded in 18 FTD, 18 AD, and 20 healthy controls (HC). In the FTD and AD patients, the EEG recordings were performed at the prodromal stage of dementia, at the onset of dementia, and three years after the onset of dementia. HC underwent three EEG recordings at 2-3-year time interval. Information flows underlying EEG activity recorded at electrode pairs were estimated by means of Mutual Information (MI) analysis. The functional organization of the cortical network was modelled by means of the Graph theory analysis on MI adjacency matrices.
Results
Graph theory analysis showed that the main hub of HC (Parietal area) was lost in FTD patients at onset of dementia, substituted by provincial hubs in frontal leads. No changes in global network organization were found in AD.
Conclusions
Results showed that the parietal “main hub” of HC and AD patients was lost in the FTD patients at the dementia onset, substituted by frontal “provincial hubs” and local “small worlds”. No change in global network organization was found in AD patients during the disease progression.
Despite a progressive cognitive impairment during the FTD and AD progression, only the FTD patients showed a derangement in the cortical network modularity, possibly due to dysfunctions in frontal functional connectivity.
RELATIONSHIP BETWEEN CSF ABETA42/40 AND [18F]FLUTEMETAMOL PET-DERIVED CENTILOID UNITS IN SUBJECTIVE COGNITIVE DECLINE AND MILD COGNITIVE IMPAIRMENT
Abstract
Aims
To apply the Centiloid PET quantitation Atlas to quantify [18F]flutemetamol brain images and investigate the relationship between Centiloid and CSF Aβ42/40 in patients with Subjective Cognitive Decline (SCD) or Mild Cognitive Impairment (MCI).
Methods
The Centiloid atlas was applied to a cohort of 387 [18F]flutemetamol images obtained from subjects from the Swedish BioFINDER I study, following the methods described by Klunk et.al. (Alzheimer’s & Dementia, 2015). Subjects, comprising Controls (n=128), SCD (n=114) and MCI (n=145) were included for analysis based on the availability of CSF Aβ42/40 information. CSF Aβ were determined using the ADx analysis platform.
Results
Mean (±SD) CSF Aβ42/40 ratios were 0.146 (±0.032) for Subjects in CL bands ranging from -20(±5) CL to 20(±5) CL. Subjects in the 30(±5) CL band were significantly higher (p=0.0155) in average CSF Aβ42/40 ratios (0.138 ± 0.042) than Subjects in the 40(±5) CL band (0.100 ± 0.036). The average CSF Aβ42/40 ratios value for all Subjects ranging from 50(±5) Cl to >135 CL was 0.073 (±0.019).
Conclusions
The comparison of [18F]flutemetamol image quantitation and CSF Aβ42/40 indicates that CSF Aβ42/40 shows amyloid positivity at approximately 30 CL units. Recent comparisons between visual rating and Centiloids from PET imaging indicate visual positivity beginning at approximately 20 CL units – indicating that PET imaging to be competitive with CSF Aβ42/40 measures for the detection of early amyloid.
EVALUATING THE SENSITIVITY OF CENTILOID QUANTIFICATION TO PIPELINE DESIGN
Abstract
Aims
To evaluate the impact of pipeline design for Centiloid quantification of amyloid PET scans.
Methods
A total of 32 Centiloid pipelines were created on SPM12 using different combinations of 2 target and 4 Reference Regions (RR) (standard GAAIN VOIs vs subject-based), and 2 analysis spaces (Native vs MNI) and calibrated to render Centiloid values(CL) for 2 tracers (18F-Flutemetamol and 18F-Florbetaben). A total of 329 amyloid PET images from AMYPAD study were quantified using CL pipelines (Table1). The impact of the pipeline design factors was assessed using Generalized Estimating Equation (GEE) analyses. The model also included the following individual morphometric variables: total intracranial volume (TIV) and grey and white matter volumes in the reference and target regions.
Results
Reference region selection and its definition have a significant impact on CL values (Table2). There is no statistically significant relationship for CL as a function of tracer, space, target. Regarding the reference regions (Table3), the pons provided the lowest mean CL(18.3±3.2) compared to the whole-cerebellum(34.3±3.0) and the cerebellum grey-matter (37.8±2.9). Using the whole-cerebellum + Brainstem as the reference region resulted in a mean difference of 4.8±0.3 and 8.22±1.0 CL in comparison to whole-cerebellum and cerebellum grey-matter respectively. Using GAAIN reference regions provided slightly higher mean CL values (ΔCL:1.8±0.7) compared to a subject-based method(Table4).
Conclusions
While CL quantification is robust against differences in individual morphometric measures, analysis space, tracer, and target VOI, the choice of reference region can significantly influence the resulting CL values, with the pons showing the highest discrepancies relative to other reference regions.
REACTIVE ASTROCYTES AS IMAGED WITH 11C-DED IN PATIENTS WITH DIFFERENT DEMENTIA DISORDERS
Abstract
Aims
Neuroinflammation, which receives ever increasing interest in neurodegenerative diseases, comprises a heterogeneous cascade of events that are thought to be related to the downstream neurodegeneration. The aim of this study was to evaluate the binding of 1C-Deuterium-L-Deprenyl PET (DED) as a measure of reactive astrocytes in patients with different dementia disorders, and to assess its association with other disease biomarkers.
Methods
Twelve patients with a clinical diagnosis of semantic variant of primary progressive aphasia (svPPA) and behavioral variant of frontotemporal dementia (bvFTD) were recruited. All patients had a cerebrospinal fluid biomarker profile that was inconsistent with Alzheimer’s disease (AD). The imaging protocol includes 11C-DED-PET, 18F-FDG-PET, and a 3D T1 MRI. A group of amyloid-beta positive patients with AD (n=20) that underwent similar investigations were used for comparison. Age-adjusted w-scores were created for assessing the load of individual 11C-DED binding relative to that of healthy controls (HC; n=14).
Results
The patients with svPPA and bvFTD, that have so far completed the imaging protocol, show high 11C-DED binding compared to HC, although the load of binding was heterogeneous across patients. The regional distribution of 11C-DED binding in svPPA and bvFTD differed substantially from that in patients with AD and was consistent with the expected underlying pattern of neurodegeneration in those disorders.
Conclusions
Reactive astrocyte activation appears to be a common feature of different dementia disorders, although the regional pattern of activation differs. Ongoing work evaluates the relationship between patterns of reactive astrocyte activation, synaptic dysfunction, atrophy, and cognitive performance.
DETECTION OF TAU ACCUMULATION IN AMYLOID POSITIVE PRECLINICAL AND EARLY AD USING [18F]RO-948, [18F]MK-6240, [18F]GTP1, AND [18F]FLORTAUCIPIR TAU PET TRACERS
Abstract
Aims
This study aims to evaluate the ability of [18F]RO-948, [18F]MK-6240, [18F]GTP1, and [18F]flortaucipir tau PET tracers to detect longitudinal changes in amyloid positive cognitive unimpaired (Aβ+CU) and mild cognitive impaired (Aβ+MCI) subjects.
Methods
A total of 341 participants underwent tau PET at baseline and after one year using either [18F]RO-948 (n=13 Aβ+CU, 23 Aβ+MCI), [18F]MK-6240 (n=30 Aβ+CU, 37 Aβ+MCI), [18F]GTP1 (n=5 Aβ+CU, 126 Aβ+MCI), or [18F]flortaucipir (n=61 Aβ+CU, 46 Aβ+MCI). Amyloid positivity was assessed using quantitative amyloid PET when available or via either PET visual read or CSF Aβ42/40. Annualized SUVR changes in Braak I/II and Braak III/IV were calculated using a Freesurfer based pipeline, with whole cerebellar cortex as reference region. Cohen’s D longitudinal effect size was computed for each cohort. 95% confidence intervals (CIs) were calculated using a bootstrap approach.
Results
In Aβ+CU (Figure 1), [18F]RO-948 and [18F]MK-6240 had comparable effect sizes in Braak I/II region (0.61±0.27 and 0.72±0.18, respectively), whereas [18F]flortaucipir did not detect any longitudinal change. The [18F]GTP1 Aβ+CU results were not reliable because only 5 subjects, younger and with significantly lower amyloid load were imaged. In Aβ+MCI (Figure 2), all tracers had comparable effect sizes in Braak III/IV region (range: 0.50-0.62).
Conclusions
Though cohort differences precluded cross-cohort evaluations, both [18F]RO-948 and [18F]MK-6240 could detect longitudinal changes in Braak I/II in Aβ+CU after one year, while no changes were detectable with [18F]flortaucipir and not enough data was available for [18F]GTP1. All tracers performed similarly in Aβ+MCI. Head-to-head studies are needed to confirm these results.
SMALL VESSELS-BIG PROBLEMS IN ALZHEIMER’S DISEASE
Abstract
Aims
Alzheimer’s Disease (AD) and vascular dementia (VaD) account for most cases of dementia. Although their etiology remains unclear, the major vascular cause of dementia is Cerebral Small Vessel Disease (CSVD). The diagnosis of CSVD relies on imaging findings on MRI. White matter hyperintensities (WMH) are the foremost common hallmark of CSVD. The aim of our study is to investigate the pathological mechanisms underlying WMH.
Methods
We conducted high-field (7 Tesla) MRI on human postmortem brains (n=24) of elderly people with chronic hypertension during life to assess CSVD burden, focusing on the analysis of WMH by Fazekas score severity and volumetry. After careful evaluation, brain axial biopsies were taken and (immuno-)histochemistry (IHC) was performed to assess neuroinflammation, vascular remodeling, (de-)myelination, and amyloid-β burden. To accurately co-register our MRI findings to IHC data, we developed a custom written MATLAB script to compare changes in WMH on MRI to microscopical changes in IHC.
Results
WMH were characterized by an increase in neuroinflammatory markers compared to surrounding normal appearing white matter (NAWM). Higher burden of WMH positively correlated to an increase of active microglial cells in WMH. Individuals with higher WMH burden showed increased loss of myelin, which was more pronounced in WMH than in NAWM. Blood vessels found in WMH were smaller than those found in NAWM.
Conclusions
Our study provides a unique opportunity to investigate neurodegenerative markers underlying MRI visible lesions at microscopical scale. Understanding CSVD hallmarks’ underlying pathology will shed light on possible biomarkers or potential treatment strategies for CSVD, and subsequently AD.