IC‐P‐183: AN UNSUPERVISED DATA‐DRIVEN APPROACH TO DEFINE WHOLE‐BRAIN PATTERNS OF RATE OF TAU ACCUMULATION

Background: Given the central role that accumulation of tau pathology is thought to play in the clinical progression especially in early disease stages, the ability to quantify the spread of tau is of great interest.Due to observed heterogeneity in patterns of tau deposition,though, stratification of individuals into homogenous prognostic sub-cohorts is critical to better understanding the natural course of this pathology progression,and ultimately the clinical disease progression.The classical group-level analyses of longitudinal tau-PET,where from brain regions are assessed independently,yet,rely on pre-defined group separations largely defined by prior knowledge on AD-related factors.As this approach might underplay the spatially and temporally-rich information provided by tau-PET,we propose an unsupervised data-driven whole-brain pattern analysis approach.Methods: We used longitudinal flortaucipir-PET scans to investigate distinct patterns of tau accumulation over time in healthy(N1⁄470;up to 2.8y follow-up) and MCI(N1⁄443;up to 2.9y follow-up).Flortaucipir-burden,adjusted for average cerebellar-uptake and corrected for partial-volume-effects,were calculated separately for twelve Braak1-3,fourteen Braak4,sixteen Braak5,and ten Braak6 regions for each timepoint.We performed a data-driven longitudinal cluster analysis as unsupervised profiling of flortaucipir-progression.A mixture of multivariate generalized linear mixed models was developed with random-intercept to analyze flortaucipir-burden from all brain regions and all time-points simultaneously(i.e.,multivariate featurespace) while accounting for potential inter-regional dependence and irregular time sampling. Results: Two distinct flortaucipir-progression patterns were identified in the Braak1/3-regions where sub-groups differ in flortaucipir-rate only in bilateral fusiform and lingual regions and differ in baseline flortaucipir-burden in the remaining Braak1/3-regions.Advanced age, female gender,amyloid-positivity,and greater CDR were associated with this flortaucipir-progression pattern distinction at Braak1/3 stages.Two distinct flortaucipir-progression patterns were identified in the Braak4 regions where sub-groups differ in flortaucipir-rate in bilateral inferior temporal and bilateral middle temporal regions and differ in baseline flortaucipir burden in bilateral temporal pole. Advanced age, amyloid-positivity, and greater CDR were associated with this flortaucipir-progression pattern distinction at Braak4 stage.Only one cluster was identified both for Braak5 and Braak6 stages with a non-significant flortaucipir-rate from baseline.Participants in distinct Braak1/3 and Braak4 progression clusters significantly differed in baseline cognition(LIMM,LDEL,TrailsA,Trails-B) and in rates of cognitive decline(LIMM,LDEL). Conclusions: The proposed unsupervised approaches aims to overcome challenges related to participant heterogeneity in tau pathology that might confound studies on clinical and pathological disease progression.