Subtype and Stage Inference with Timescales

Abstract

Neurodegenerative conditions typically have highly heterogeneous trajectories, with variability in both the spatial and temporal progression of neurological changes. Disentangling the variability in spatiotemporal progression patterns offers major benefits for patient stratification and disease understanding but is a complex methodological challenge. Here we present Temporal Subtype and Stage Inference (T-SuStaIn), a technique that uniquely integrates distinct ideas from unsupervised learning: disease progression modelling, clustering, and hidden Markov modelling. We formulate T-SuStaIn mathematically and devise an algorithm for inferring the model parameters and uncertainty. We demonstrate that the combination of disease progression modelling, clustering, and hidden Markov modelling uniquely enables the discovery of subtypes distinguished not just by ordering of abnormality accumulation, but also timescale. We apply T-SuStaIn to longitudinal volumetric imaging data from the Alzheimer’s Disease Neuroimaging Initiative, deriving spatiotemporal Alzheimer’s disease subtypes together with their timelines of evolution and associated uncertainty. T-SuStaIn has broad utility across a range of longitudinal clustering problems, both in neurodegenerative conditions and more widely in progressive diseases.

L. M. Aksman—Joint first author.

Data used in preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wp-content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf.