- Email: [email protected]
WHITE MATTER HYPERINTENSITIES ARE ASSOCIATED WITH HIPPOCAMPAL ATROPHY RATES AFTER ADJUSTING FOR OTHER VASCULAR MARKERS IN PREDEMENTIA DISEASE STAGES
Poster Presentations: Wednesday, July 19, 2017
P1547
Neuroscience Institute, Singapore, Singapore; 7Cerebral Imaging Centre Douglas Research Centre, Verdun, QC, Canada; 8icometrix NV, Leuven, Belgium; 9Douglas Hospital Research Centre, Verdun, QC, Canada; 10 Reference Center for Biological Markers of Dementia (BIODEM), Institute Born-Bunge, University of Antwerp, Antwerp, Belgium; 11 Department of Neurology and Memory Clinic, Hospital Network Antwerp (ZNA) Middelheim and Hoge Beuken, Antwerp, Belgium; 12Institute Born-Bunge, University of Antwerp, Antwerp, Belgium; 13Douglas Mental Health University Institute, Montreal, QC, Canada; 14McConnell Brain Imaging Centre - McGill University, Montreal, QC, Canada. Contact e-mail: [email protected]
should be considered in studies aiming to validate classifications such as A/T/N. Specifically, whereas GMV could be an effective marker for enrichment of a 4-year clinical trial, WMV could better enrich a 2-year one.
Background: A working group has recently proposed a theoretical classification scheme for AD biomarkers: Amyloid/Tau/Neurodegeneration (“A/T/N”). The critical next step is validating this scheme’s neurodegeneration biomarkers (CSF total tau (tTau), FDG-PET, and structural MRI). Notably, it is postulated that neurodegeneration markers measure different processes within the neurodegenerative spectrum: CSF tTau is postulated to measure predominantly neuronal injury, FDG-PET synaptic degeneration, and structural MRI brain cell loss. Therefore, this study aimed to test whether neurodegeneration markers significantly improve prediction when added to the combined A and T markers, and also to compare their different predictive powers of progression to dementia over typical clinical trial periods of 2 and 4 years. Methods: Data from MCI subjects from ADNI (N¼473) was used, all having baseline CSF biomarkers, structural MRI, FDG-PET in AD-related regions, and cognitive and clinical follow-up up to 4 years. Stratification according to “A” and “T” status was applied, based on CSF Ab42 and pTau181. The MSmetrix software was used to quantify whole brain, grey matter (GMV), and white matter volumes (WMV) on MRI. Logistic and linear models were used to assess prediction of progression to dementia and cognitive decline, respectively. All analyses were adjusted for age, gender, education, and APOE ε4 carrier status. Results: We found that neurodegeneration markers predicted progression to dementia only in MCI A+T+ subjects. Clinical status and cognitive decline over 2 years in MCI A+T+ was predicted by all neurodegeneration markers but CSF tTau, whereas 4-year progression was predicted by all but WMV. Interestingly, GMV best predicted progression over 4 years, while WMV best predicted progression in 2 years time (Fig. 1). Conclusions: Together, these results support the A/T/N system as the association between the three categories is highly important to predict progression. Additionally, we showed that the relative predictive powers of different neurodegeneration markers strongly depend on time-to-progression to dementia. Importantly, this suggests that time-to-progression to dementia
Hugh Pemberton1, Cassidy M. Fiford1, Phoebe Walsh1, Owen T. Carmichael2, Geert Jan Biessels3, Carole H. Sudre1,4, M. Jorge Cardoso1,4, Frederik Barkhof5,6, Josephine Barnes1, the ADNI Investigators, 1UCL Institute of Neurology, London, United Kingdom; 2 Pennington Biomedical Research Center, Baton Rouge, LA, USA; 3 University Medical Center Utrecht, Utrecht, Netherlands; 4Translational Imaging Group, University College London, London, United Kingdom; 5 Radiology & Nuclear Medicine, VU University Medical Center, Amsterdam, Netherlands; 6Institutes of Neurology and Healthcare Engineering, UCL, London, United Kingdom. Contact e-mail: h.pemberton@ ucl.ac.uk
Figure 1. Grey matter volume best predicted progression over 4 years, while white matter volume best predicted progression over 2 years.
P4-524
WHITE MATTER HYPERINTENSITIES ARE ASSOCIATED WITH HIPPOCAMPAL ATROPHY RATES AFTER ADJUSTING FOR OTHER VASCULAR MARKERS IN PREDEMENTIA DISEASE STAGES
Background: Hippocampal volume change is an important diagnostic and disease-tracking biomarker in Alzheimer’s disease (AD). As a proxy for global cerebral vascular disease (CVD), white matter hyperintensities (WMH) have been found to be predictive of hippocampal atrophy rates in controls and late mild cognitive impairment (LMCI). In this preliminary work we extend this to early MCI (EMCI) and assess whether WMH has predictive value over other CVD markers: lacunes and microbleeds (CMB). Methods: Data were from the Alzheimer’s Disease Neuroimaging Initiative (Go and 2). We measured hippocampal atrophy rates using automated techniques (Similarity and Truth Estimation for Propagated Segmentations – STEPS, and symmetric Boundary Shift Integral - BSI) over a 12 month interval. WMH were measured on baseline images using a novel automated algorithm (Bayesian Model Selection BaMoS). We counted CMBs and lacunes manually at baseline. For CMBs, results were categorised into 0, 1, 2-4 and >4 CMBs. Linear regression analyses were used to assess whether each CVD marker was associated with subsequent hippocampal change. For CMBs and lacunes, the reference groups against which those who had CMBs and lacunes were compared, were those with none. Separate models were fitted for each CVD marker and diagnostic group. Further models were fitted with all CVD markers as predictors. Other covariates included gender and intracranial volume (TIV). Results: Log2WMH was predictive of hippocampal atrophy rates in controls, EMCI and LMCI (see table). An overall test showed CMB category was predictive of hippocampal atrophy rates in controls. We found no evidence for a relationship of lacunes with hippocampal atrophy rates. Log2WMH was independently predictive of hippocampal atrophy rates in controls and LMCI, and at trend level in EMCI. Having 1 vs. 0 CMB was independently predictive of hippocampal rates in controls and AD. Conclusions: We found that CVD markers predict hippocampal atrophy in all diagnostic groups: WMH predicted atrophy in controls through to LMCI and CMB numbers predicted hippocampal tissue loss in controls and AD. The fact that many of these findings were independent of one another suggests that different aspects of vascular disease may contribute to disease progression.
P1548
Poster Presentations: Wednesday, July 19, 2017
Table of Results
Controls
EMCI
LMCI
AD
N with HBSI HBSI left+right, ml/year MMSE score, /30 Age, years % Male WMH, ml median (IQR) Log2WMH % CMB category*: 0, 1, 2-4, >4 CMBs % Lacunes: 0, 1, 2 lacunes Associations of log2WMH with HBSI Associations of CMB category with HBSI
113 0.05 (0.11) 29 (1) 74 (6) 50 1.9 (3.5) 0.9 (1.9) 78,19,2,1
106 0.06 (0.10) 28 (2) 70 (7) 58 1.9 (4.5) 0.9 (1.9) 82,10,8,1
96 0.11(0.13) 27 (2) 72 (8) 50 1.8 (3.9) 0.9 (1.9) 90,5,3,2
36 0.19 (0.15) 23 (2) 75 (8) 69 3.8 (8.1) 1.8 (2.0) 64,22,8,6
97,3,0 0.02 (0.00, 0.03) p[0.008
94,5,1 0.01 (0.00, 0.02) p[0.02
98,2,0 0.02 (0.01, 0.04) p[0.002
97,0,3 0.01 (-0.02, 0.03) p¼0.6
0 vs. 1 CMB: 0.07 (0.01, 0.12), p[ 0.02 0 vs. 2-4 CMBs: 0.06 (-0.10, 0.21), p¼0.5 0 vs. >4 CMBs: 0.20 (-0.02, 0.43) p¼0.08 Overall test: P[0.04 0 vs. 1 lacune: -0.04 (-0.17, 0.09) p¼0.5
0 vs. 1 CMB: 0.03 (-0.04, 0.09), p¼0.4 0 vs. 2-4 CMBs: -0.02 (-0.09, 0.06), p¼0.7 0 vs. >4 CMBs: -0.08 (-0.28, 0.12), p¼0.4 Overall test: p¼0.6 0 vs. 1 lacune: 0.04 (-0.05, 0.14), p¼ 0.3 0 vs. 2 lacunes: 0.12 (-0.08, -0.32) p¼0.2 Overall test: P¼ 0.3 log2WMH: 0.01 (-0.00, 0.02) P¼0.09
0 vs. 1 CMB: 0.09 (-0.04, 0.21), p¼0.2 0 vs. 2-4 CMBs: 0.05 (-0.11, 0.22), p¼0.5 0 vs. >4 CMBs: 0.09 (-0.10, 0.29), p¼0.4 Overall test: p¼0.4 0 vs. 1 lacune: -0.10 (-0.29, 0.10), p¼0.3
0 vs. 1 CMB: 0.14 (0.02, 0.26), p[0.02 0 vs. 2-4 CMBs: 0.02 (-0.17, 0.20), p¼0.9 0 vs. >4 CMBs: 0.04 (-0.18, 0.27), p¼0.7 Overall test: p¼0.1 0 vs. 2 lacunes: 0.03 (-0.28, 0.34), p¼0.8
log2WMH: 0.02 (0.01, 0.04) p[ 0.009
0 vs. 1 CMB: 0.15 (0.02, 0.27), p[0.02
Associations of number of lacunes with HBSI
Significant or trend-level independent associations with HBSI
log2WMH: 0.01 (0.00, 0.02) p[0.03 0 vs. 1 CMB: 0.06, (0.01, 0.12) p[0.03
Results are reported as mean (SD) unless otherwise indicated and for associations these are beta, (95% CI) and p value. Betas represent a one unit increase in the predictor variable on the outcome (mls / year loss hippocampal volume). * data missing in 3 controls, 2 EMCI. Bold indicates significant at p<0.05.
P4-525
DATA-DRIVEN TAU-PET COVARIANCE NETWORKS ENHANCE PREDICTION OF RETROSPECTIVE COGNITIVE CHANGE IN ALZHEIMER’S DISEASE
Jacob W. Vogel1, Niklas Mattsson2,3, Yasser Iturria Medina1, Olof Strandberg2, Michael Sch€oll2,4, Christian Dansereau5,6, Sylvia Villeneuve7, Wiesje M. van der Flier8, Philip Scheltens8, Pierre Bellec5,6, Alan C. Evans1, Oskar Hansson2,3, Rik Ossenkoppele8, 1 Montreal Neurological Institute, McGill University, Montreal, QC, Canada; 2Clinical Memory Research Unit, Lund University, Lund, Sweden; 3 Memory Clinic, Sk ane University Hospital, Lund, Sweden; 4Wallenberg Centre for Molecular and Translational Medicine and the Department of Psychiatry and Neurochemistry, University of Gothenburg, Gothenburg, Sweden; 5Department of Computer Science and Operations Research, Universite de Montreal, Montreal, QC, Canada; 6Centre de Recherche de l’Institut Universitaire de Geriatrie de Montreal, University of Montreal, Montreal, QC, Canada; 7Department of Psychiatry, McGill University, Montreal, QC, Canada; 8Alzheimer Center and Department of Neurology, VU University Medical Center, Amsterdam Neuroscience, Amsterdam, Netherlands. Contact e-mail: [email protected] Background: Positron emission tomography (PET) studies investigating filamentous tau pathology in vivo have used theory-derived regions-of-interest (ROIs) based on post-mortem studies. However, hypothesis-free data-driven approaches may yield ROIs optimized specifically for tau-PET data, resulting in better clinical utility. Methods: Demographic information, MMSE scores, AV1451-PET images and T1-weighted magnetic resonance images closest to PET were downloaded from the ADNI LONI website for 90 individuals (43 controls, 37 MCI, 10 AD dementia). AV1451 images were coregistered to each subject’s T1 image and intensity normalized to the cerebellar
gray matter. Images were then spatially normalized to the MNI152 template, stacked into a 4D image and masked with a liberal cortical mask including subcortical regions. This image stack was entered into a previously validated voxelwise clustering algorithm, which includes a procedure for defining optimal clustering solutions. The algorithm produces ROIs that represent patterns of AV1451 signal covariance across subjects. Linear mixed models were run assessing relationships between AV1451 signal within these ROIs and retrospective change in MMSE scores (575 observations). Corrected Aikake’s information criterion (AICc) was used to compare models using our data-derived ROIs to models using theory-derived ROIs from three tau-PET studies published in 2016. All models were adjusted for age, gender and education. Results: The clustering algorithm identified a six-cluster solution to be the best solution (Figure 1). These clusters included a medial/inferior temporal ROI, a temporo-parietal predominant ROI, a frontal ROI, a dorsal ROI, a “partial-volume” ROI encircling the cortex, and an “off-target” ROI encompassing the subcortex, medial occipital lobe and the hippocampus. Across all ROIs tested, the model using the data-driven temporo-parietal ROI to predict retrospective change in MMSE (Figure 2) showed the best model fit (AICc¼2312.56). With the exception of the stage IV ROI from Cho et al. (AICc¼2320.54), no other ROIs demonstrated competitive model fit (DAICc > 10; Figure 3). Conclusions: Our clustering approach yielded a data-driven temporo-parietal ROI that is known to be highly AD specific and was identified as the best predictor of retrospective cognitive decline. The results suggest that hypothesisfree data-derived ROIs may offer enhanced predictive utility compared to theory-driven ROIs by utilizing information specific to tau-PET signal.
P1547
Neuroscience Institute, Singapore, Singapore; 7Cerebral Imaging Centre Douglas Research Centre, Verdun, QC, Canada; 8icometrix NV, Leuven, Belgium; 9Douglas Hospital Research Centre, Verdun, QC, Canada; 10 Reference Center for Biological Markers of Dementia (BIODEM), Institute Born-Bunge, University of Antwerp, Antwerp, Belgium; 11 Department of Neurology and Memory Clinic, Hospital Network Antwerp (ZNA) Middelheim and Hoge Beuken, Antwerp, Belgium; 12Institute Born-Bunge, University of Antwerp, Antwerp, Belgium; 13Douglas Mental Health University Institute, Montreal, QC, Canada; 14McConnell Brain Imaging Centre - McGill University, Montreal, QC, Canada. Contact e-mail: [email protected]
should be considered in studies aiming to validate classifications such as A/T/N. Specifically, whereas GMV could be an effective marker for enrichment of a 4-year clinical trial, WMV could better enrich a 2-year one.
Background: A working group has recently proposed a theoretical classification scheme for AD biomarkers: Amyloid/Tau/Neurodegeneration (“A/T/N”). The critical next step is validating this scheme’s neurodegeneration biomarkers (CSF total tau (tTau), FDG-PET, and structural MRI). Notably, it is postulated that neurodegeneration markers measure different processes within the neurodegenerative spectrum: CSF tTau is postulated to measure predominantly neuronal injury, FDG-PET synaptic degeneration, and structural MRI brain cell loss. Therefore, this study aimed to test whether neurodegeneration markers significantly improve prediction when added to the combined A and T markers, and also to compare their different predictive powers of progression to dementia over typical clinical trial periods of 2 and 4 years. Methods: Data from MCI subjects from ADNI (N¼473) was used, all having baseline CSF biomarkers, structural MRI, FDG-PET in AD-related regions, and cognitive and clinical follow-up up to 4 years. Stratification according to “A” and “T” status was applied, based on CSF Ab42 and pTau181. The MSmetrix software was used to quantify whole brain, grey matter (GMV), and white matter volumes (WMV) on MRI. Logistic and linear models were used to assess prediction of progression to dementia and cognitive decline, respectively. All analyses were adjusted for age, gender, education, and APOE ε4 carrier status. Results: We found that neurodegeneration markers predicted progression to dementia only in MCI A+T+ subjects. Clinical status and cognitive decline over 2 years in MCI A+T+ was predicted by all neurodegeneration markers but CSF tTau, whereas 4-year progression was predicted by all but WMV. Interestingly, GMV best predicted progression over 4 years, while WMV best predicted progression in 2 years time (Fig. 1). Conclusions: Together, these results support the A/T/N system as the association between the three categories is highly important to predict progression. Additionally, we showed that the relative predictive powers of different neurodegeneration markers strongly depend on time-to-progression to dementia. Importantly, this suggests that time-to-progression to dementia
Hugh Pemberton1, Cassidy M. Fiford1, Phoebe Walsh1, Owen T. Carmichael2, Geert Jan Biessels3, Carole H. Sudre1,4, M. Jorge Cardoso1,4, Frederik Barkhof5,6, Josephine Barnes1, the ADNI Investigators, 1UCL Institute of Neurology, London, United Kingdom; 2 Pennington Biomedical Research Center, Baton Rouge, LA, USA; 3 University Medical Center Utrecht, Utrecht, Netherlands; 4Translational Imaging Group, University College London, London, United Kingdom; 5 Radiology & Nuclear Medicine, VU University Medical Center, Amsterdam, Netherlands; 6Institutes of Neurology and Healthcare Engineering, UCL, London, United Kingdom. Contact e-mail: h.pemberton@ ucl.ac.uk
Figure 1. Grey matter volume best predicted progression over 4 years, while white matter volume best predicted progression over 2 years.
P4-524
WHITE MATTER HYPERINTENSITIES ARE ASSOCIATED WITH HIPPOCAMPAL ATROPHY RATES AFTER ADJUSTING FOR OTHER VASCULAR MARKERS IN PREDEMENTIA DISEASE STAGES
Background: Hippocampal volume change is an important diagnostic and disease-tracking biomarker in Alzheimer’s disease (AD). As a proxy for global cerebral vascular disease (CVD), white matter hyperintensities (WMH) have been found to be predictive of hippocampal atrophy rates in controls and late mild cognitive impairment (LMCI). In this preliminary work we extend this to early MCI (EMCI) and assess whether WMH has predictive value over other CVD markers: lacunes and microbleeds (CMB). Methods: Data were from the Alzheimer’s Disease Neuroimaging Initiative (Go and 2). We measured hippocampal atrophy rates using automated techniques (Similarity and Truth Estimation for Propagated Segmentations – STEPS, and symmetric Boundary Shift Integral - BSI) over a 12 month interval. WMH were measured on baseline images using a novel automated algorithm (Bayesian Model Selection BaMoS). We counted CMBs and lacunes manually at baseline. For CMBs, results were categorised into 0, 1, 2-4 and >4 CMBs. Linear regression analyses were used to assess whether each CVD marker was associated with subsequent hippocampal change. For CMBs and lacunes, the reference groups against which those who had CMBs and lacunes were compared, were those with none. Separate models were fitted for each CVD marker and diagnostic group. Further models were fitted with all CVD markers as predictors. Other covariates included gender and intracranial volume (TIV). Results: Log2WMH was predictive of hippocampal atrophy rates in controls, EMCI and LMCI (see table). An overall test showed CMB category was predictive of hippocampal atrophy rates in controls. We found no evidence for a relationship of lacunes with hippocampal atrophy rates. Log2WMH was independently predictive of hippocampal atrophy rates in controls and LMCI, and at trend level in EMCI. Having 1 vs. 0 CMB was independently predictive of hippocampal rates in controls and AD. Conclusions: We found that CVD markers predict hippocampal atrophy in all diagnostic groups: WMH predicted atrophy in controls through to LMCI and CMB numbers predicted hippocampal tissue loss in controls and AD. The fact that many of these findings were independent of one another suggests that different aspects of vascular disease may contribute to disease progression.
P1548
Poster Presentations: Wednesday, July 19, 2017
Table of Results
Controls
EMCI
LMCI
AD
N with HBSI HBSI left+right, ml/year MMSE score, /30 Age, years % Male WMH, ml median (IQR) Log2WMH % CMB category*: 0, 1, 2-4, >4 CMBs % Lacunes: 0, 1, 2 lacunes Associations of log2WMH with HBSI Associations of CMB category with HBSI
113 0.05 (0.11) 29 (1) 74 (6) 50 1.9 (3.5) 0.9 (1.9) 78,19,2,1
106 0.06 (0.10) 28 (2) 70 (7) 58 1.9 (4.5) 0.9 (1.9) 82,10,8,1
96 0.11(0.13) 27 (2) 72 (8) 50 1.8 (3.9) 0.9 (1.9) 90,5,3,2
36 0.19 (0.15) 23 (2) 75 (8) 69 3.8 (8.1) 1.8 (2.0) 64,22,8,6
97,3,0 0.02 (0.00, 0.03) p[0.008
94,5,1 0.01 (0.00, 0.02) p[0.02
98,2,0 0.02 (0.01, 0.04) p[0.002
97,0,3 0.01 (-0.02, 0.03) p¼0.6
0 vs. 1 CMB: 0.07 (0.01, 0.12), p[ 0.02 0 vs. 2-4 CMBs: 0.06 (-0.10, 0.21), p¼0.5 0 vs. >4 CMBs: 0.20 (-0.02, 0.43) p¼0.08 Overall test: P[0.04 0 vs. 1 lacune: -0.04 (-0.17, 0.09) p¼0.5
0 vs. 1 CMB: 0.03 (-0.04, 0.09), p¼0.4 0 vs. 2-4 CMBs: -0.02 (-0.09, 0.06), p¼0.7 0 vs. >4 CMBs: -0.08 (-0.28, 0.12), p¼0.4 Overall test: p¼0.6 0 vs. 1 lacune: 0.04 (-0.05, 0.14), p¼ 0.3 0 vs. 2 lacunes: 0.12 (-0.08, -0.32) p¼0.2 Overall test: P¼ 0.3 log2WMH: 0.01 (-0.00, 0.02) P¼0.09
0 vs. 1 CMB: 0.09 (-0.04, 0.21), p¼0.2 0 vs. 2-4 CMBs: 0.05 (-0.11, 0.22), p¼0.5 0 vs. >4 CMBs: 0.09 (-0.10, 0.29), p¼0.4 Overall test: p¼0.4 0 vs. 1 lacune: -0.10 (-0.29, 0.10), p¼0.3
0 vs. 1 CMB: 0.14 (0.02, 0.26), p[0.02 0 vs. 2-4 CMBs: 0.02 (-0.17, 0.20), p¼0.9 0 vs. >4 CMBs: 0.04 (-0.18, 0.27), p¼0.7 Overall test: p¼0.1 0 vs. 2 lacunes: 0.03 (-0.28, 0.34), p¼0.8
log2WMH: 0.02 (0.01, 0.04) p[ 0.009
0 vs. 1 CMB: 0.15 (0.02, 0.27), p[0.02
Associations of number of lacunes with HBSI
Significant or trend-level independent associations with HBSI
log2WMH: 0.01 (0.00, 0.02) p[0.03 0 vs. 1 CMB: 0.06, (0.01, 0.12) p[0.03
Results are reported as mean (SD) unless otherwise indicated and for associations these are beta, (95% CI) and p value. Betas represent a one unit increase in the predictor variable on the outcome (mls / year loss hippocampal volume). * data missing in 3 controls, 2 EMCI. Bold indicates significant at p<0.05.
P4-525
DATA-DRIVEN TAU-PET COVARIANCE NETWORKS ENHANCE PREDICTION OF RETROSPECTIVE COGNITIVE CHANGE IN ALZHEIMER’S DISEASE
Jacob W. Vogel1, Niklas Mattsson2,3, Yasser Iturria Medina1, Olof Strandberg2, Michael Sch€oll2,4, Christian Dansereau5,6, Sylvia Villeneuve7, Wiesje M. van der Flier8, Philip Scheltens8, Pierre Bellec5,6, Alan C. Evans1, Oskar Hansson2,3, Rik Ossenkoppele8, 1 Montreal Neurological Institute, McGill University, Montreal, QC, Canada; 2Clinical Memory Research Unit, Lund University, Lund, Sweden; 3 Memory Clinic, Sk ane University Hospital, Lund, Sweden; 4Wallenberg Centre for Molecular and Translational Medicine and the Department of Psychiatry and Neurochemistry, University of Gothenburg, Gothenburg, Sweden; 5Department of Computer Science and Operations Research, Universite de Montreal, Montreal, QC, Canada; 6Centre de Recherche de l’Institut Universitaire de Geriatrie de Montreal, University of Montreal, Montreal, QC, Canada; 7Department of Psychiatry, McGill University, Montreal, QC, Canada; 8Alzheimer Center and Department of Neurology, VU University Medical Center, Amsterdam Neuroscience, Amsterdam, Netherlands. Contact e-mail: [email protected] Background: Positron emission tomography (PET) studies investigating filamentous tau pathology in vivo have used theory-derived regions-of-interest (ROIs) based on post-mortem studies. However, hypothesis-free data-driven approaches may yield ROIs optimized specifically for tau-PET data, resulting in better clinical utility. Methods: Demographic information, MMSE scores, AV1451-PET images and T1-weighted magnetic resonance images closest to PET were downloaded from the ADNI LONI website for 90 individuals (43 controls, 37 MCI, 10 AD dementia). AV1451 images were coregistered to each subject’s T1 image and intensity normalized to the cerebellar
gray matter. Images were then spatially normalized to the MNI152 template, stacked into a 4D image and masked with a liberal cortical mask including subcortical regions. This image stack was entered into a previously validated voxelwise clustering algorithm, which includes a procedure for defining optimal clustering solutions. The algorithm produces ROIs that represent patterns of AV1451 signal covariance across subjects. Linear mixed models were run assessing relationships between AV1451 signal within these ROIs and retrospective change in MMSE scores (575 observations). Corrected Aikake’s information criterion (AICc) was used to compare models using our data-derived ROIs to models using theory-derived ROIs from three tau-PET studies published in 2016. All models were adjusted for age, gender and education. Results: The clustering algorithm identified a six-cluster solution to be the best solution (Figure 1). These clusters included a medial/inferior temporal ROI, a temporo-parietal predominant ROI, a frontal ROI, a dorsal ROI, a “partial-volume” ROI encircling the cortex, and an “off-target” ROI encompassing the subcortex, medial occipital lobe and the hippocampus. Across all ROIs tested, the model using the data-driven temporo-parietal ROI to predict retrospective change in MMSE (Figure 2) showed the best model fit (AICc¼2312.56). With the exception of the stage IV ROI from Cho et al. (AICc¼2320.54), no other ROIs demonstrated competitive model fit (DAICc > 10; Figure 3). Conclusions: Our clustering approach yielded a data-driven temporo-parietal ROI that is known to be highly AD specific and was identified as the best predictor of retrospective cognitive decline. The results suggest that hypothesisfree data-derived ROIs may offer enhanced predictive utility compared to theory-driven ROIs by utilizing information specific to tau-PET signal.