Transferable conditional brain charts link cortico-subcortical structure to Alzheimer's disease progression
Abstract
Normative models chart how brain measures vary across the lifespan, expressing each individual as a deviation from a healthy reference. However, most regional models of cortical thickness disregard dependencies between cortical and subcortical anatomy, and transferring them to a new site typically requires retraining or a recalibration of location and scale alone. We condition lifespan norms of cortical thickness on subcortical volume, such that each deviation indicates whether an individual's cortex lies within the healthy range given their subcortical anatomy, and make them transferable. In 21,351 healthy individuals from 216 sites, conditioning on ten subcortical structures improved out-of-sample prediction on average, most for the amygdala and hippocampus. Refitting only the intercepts of all four distributional parameters on about 20 healthy individuals, as implemented in our open-source package NormGAMLSS, calibrated most of six unseen sites without retraining, for base and conditional models. In the Alzheimer's Disease Neuroimaging Initiative, global deviation and structural dissociation (the reduction in mean absolute deviation under conditioning) predicted conversion to mild cognitive impairment and dementia beyond the Clinical Dementia Rating and dissociation added beyond global deviation for dementia, even at similar clinically recorded impairment. Conditional norms thus provide relevant, individual-level information.
Related
- Divergent amyloid trajectories distinguish normal aging from Alzheimer's disease progression
- Mechanistic roles and therapeutic potential of gut microbiota–neuroimmune crosstalk in the development and progression of Alzheimer's disease
- Structure-Based Discovery of GTP-Enhanced Rab10 Modulators and Their Effects on Alzheimer Disease-Relevant Cellular Phenotypes
- Evolution of deep learning for structural magnetic resonance imaging analysis in Alzheimer’s disease: from slice-level classification to brain structure modeling
- LNODE: latent dynamics reveal the shared spatiotemporal structure of amyloid-β progression
BibTeX
@misc{sharma2026transferable,
title = {{Transferable conditional brain charts link cortico-subcortical structure to Alzheimer's disease progression}},
author = {Nitin Sharma and Sicheng Dai and Francesco A. Mallus and Atharva Kand and Marc Lemnitzer and Minh-Hieu Nguyen and Mebin Muhammed Hamzakoya Sulaikha and Jijia Xing and Mariam Zabihi and Gülce Lale and Gokul Perumbayil Vijayakrishnan and Elias Schreiner and Madhumitha Sivakumar Punitha and Paul Buttkus and Çağatay Yıldız and Poornima Kumar and Thomas Grote and Esten Leonardsen and Francesco Sammarco and Vinod Kumar Jangir and Sarah Sobotta and Barbora Rehák Bučková and Augustijn A. A. de Boer and Konstantinos Tsilimparis and Christian Gaser and Charlotte Fraza and Saige Rutherford and Seyed Mostafa Kia and Andre F. Marquand and Yanwu Yang and Richard Dinga and Thomas Wolfers},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.21660},
eprint = {2610.21660},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.21660},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Transferable conditional brain charts link cortico-subcortical structure to Alzheimer's disease progression
AU - Nitin Sharma
AU - Sicheng Dai
AU - Francesco A. Mallus
AU - Atharva Kand
AU - Marc Lemnitzer
AU - Minh-Hieu Nguyen
AU - Mebin Muhammed Hamzakoya Sulaikha
AU - Jijia Xing
AU - Mariam Zabihi
AU - Gülce Lale
AU - Gokul Perumbayil Vijayakrishnan
AU - Elias Schreiner
AU - Madhumitha Sivakumar Punitha
AU - Paul Buttkus
AU - Çağatay Yıldız
AU - Poornima Kumar
AU - Thomas Grote
AU - Esten Leonardsen
AU - Francesco Sammarco
AU - Vinod Kumar Jangir
AU - Sarah Sobotta
AU - Barbora Rehák Bučková
AU - Augustijn A. A. de Boer
AU - Konstantinos Tsilimparis
AU - Christian Gaser
AU - Charlotte Fraza
AU - Saige Rutherford
AU - Seyed Mostafa Kia
AU - Andre F. Marquand
AU - Yanwu Yang
AU - Richard Dinga
AU - Thomas Wolfers
PY - 2026
DA - 2026/10/15
PB - arXiv
DO - 10.48550/arxiv.2610.21660
UR - https://arxiv.org/abs/2610.21660
AB - Normative models chart how brain measures vary across the lifespan, expressing each individual as a deviation from a healthy reference. However, most regional models of cortical thickness disregard dependencies between cortical and subcortical anatomy, and transferring them to a new site typically requires retraining or a recalibration of location and scale alone. We condition lifespan norms of cortical thickness on subcortical volume, such that each deviation indicates whether an individual's cortex lies within the healthy range given their subcortical anatomy, and make them transferable. In 21,351 healthy individuals from 216 sites, conditioning on ten subcortical structures improved out-of-sample prediction on average, most for the amygdala and hippocampus. Refitting only the intercepts of all four distributional parameters on about 20 healthy individuals, as implemented in our open-source package NormGAMLSS, calibrated most of six unseen sites without retraining, for base and conditional models. In the Alzheimer's Disease Neuroimaging Initiative, global deviation and structural dissociation (the reduction in mean absolute deviation under conditioning) predicted conversion to mild cognitive impairment and dementia beyond the Clinical Dementia Rating and dissociation added beyond global deviation for dementia, even at similar clinically recorded impairment. Conditional norms thus provide relevant, individual-level information.
ER -