Data-driven clustering of atrophy patterns in Alzheimer’s disease

Background To enable early detection and treatment of Alzheimer’s disease, a better understanding of the heterogeneous disease progression is needed. Longitudinal MRI data has been utilized by previous studies to uncover patterns of brain atrophy. In this work, we aimed to improve upon previous methods by combining a continuous modeling of disease stages with a data-driven clustering algorithm. Methods We developed an algorithm that primarily stages subjects on a continuous latent-time disease scale and subsequently finds clusters of atrophy patterns in a completely data-driven manner. The core of the algorithm is a set of non-linear mixed-effects models, which includes both subject-specific random effects and cluster-specific fixed effects. We applied our algorithm to longitudinal MRI data from the ADNI and OASIS databases and analyzed the found clusters rigorously by studying biomarkers, indicators of co-pathologies, and connections with previously established subtypes. We also utilized the BioFINDER-2 cohort as an independent validation dataset. Results We found four clusters of longitudinal atrophy patterns; A : Progressive atrophy, B : Slow progression, C : Delayed atrophy and D : High atrophy. The D (High atrophy) exhibited higher amounts of atrophy, primarily in the typical regions affected in AD. The delayed atrophy group C remained relatively spared until the final stages of the disease, where we saw rapid neurodegeneration. The B (Slow progression) group displayed the least atrophy, and the A (Progressive atrophy) group showed a typical AD-like progression. However, the clusters could only be somewhat reproduced in perturbations of the discovery cohort. In the validation cohort, two clusters were found, which to a certain degree resembled some characteristics of the ADNI/OASIS clusters. Conclusion Our method combines a continuous disease staging model with a data-driven clustering algorithm, which mitigates some limitations of previous attempts at modeling brain atrophy patterns in Alzheimer’s disease. We found four clusters of atrophy patterns, which expressed different clinical profiles. However, the clusters are likely mixes of previously found AD subtypes, and are produced from continuous progression patterns rather than discrete subtypes.

Authors

Institutions

Publication Details

Journal
Alzheimer s Research & Therapy
Published
2026-11-19
DOI
https://doi.org/10.1186/s13195-026-02261-8
Primary Topic
Dementia and Cognitive Impairment Research
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Data-driven clustering of atrophy patterns in Alzheimer’s disease

Sebastian Palmqvist, Erik Stomrud, Lars Lau Rakêt, Filip Winzell et al.
Alzheimer s Research & Therapy
Dementia and Cognitive Impairment Research
article

Data-driven clustering of atrophy patterns in Alzheimer’s disease

Sebastian Palmqvist, Erik Stomrud, Lars Lau Rakêt, Filip Winzell, Niklas Mattsson, Nicola Spotorno, Kalle Åström, Anders Heyden, Shorena Janelidze, Ida Arvidsson, Niels Christian Overgaard
article en

Abstract

Background To enable early detection and treatment of Alzheimer’s disease, a better understanding of the heterogeneous disease progression is needed. Longitudinal MRI data has been utilized by previous studies to uncover patterns of brain atrophy. In this work, we aimed to improve upon previous methods by combining a continuous modeling of disease stages with a data-driven clustering algorithm. Methods We developed an algorithm that primarily stages subjects on a continuous latent-time disease scale and subsequently finds clusters of atrophy patterns in a completely data-driven manner. The core of the algorithm is a set of non-linear mixed-effects models, which includes both subject-specific random effects and cluster-specific fixed effects. We applied our algorithm to longitudinal MRI data from the ADNI and OASIS databases and analyzed the found clusters rigorously by studying biomarkers, indicators of co-pathologies, and connections with previously established subtypes. We also utilized the BioFINDER-2 cohort as an independent validation dataset. Results We found four clusters of longitudinal atrophy patterns; A : Progressive atrophy, B : Slow progression, C : Delayed atrophy and D : High atrophy. The D (High atrophy) exhibited higher amounts of atrophy, primarily in the typical regions affected in AD. The delayed atrophy group C remained relatively spared until the final stages of the disease, where we saw rapid neurodegeneration. The B (Slow progression) group displayed the least atrophy, and the A (Progressive atrophy) group showed a typical AD-like progression. However, the clusters could only be somewhat reproduced in perturbations of the discovery cohort. In the validation cohort, two clusters were found, which to a certain degree resembled some characteristics of the ADNI/OASIS clusters. Conclusion Our method combines a continuous disease staging model with a data-driven clustering algorithm, which mitigates some limitations of previous attempts at modeling brain atrophy patterns in Alzheimer’s disease. We found four clusters of atrophy patterns, which expressed different clinical profiles. However, the clusters are likely mixes of previously found AD subtypes, and are produced from continuous progression patterns rather than discrete subtypes.

Alzheimer s Research & Therapy
Lund University (SE), Skåne University Hospital (SE)
Dementia and Cognitive Impairment Research
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon