An uncertainty-aware, shareable, and transparent neural network architecture for brain-age modeling

Tim Hahn; Jan Ernsting; Nils R Winter; Vincent Holstein; Ramona Leenings; Marie Beisemann; Lukas Fisch; Kelvin Sarink; Daniel Emden; Nils Opel; Ronny Redlich; Jonathan Repple; Dominik Grotegerd; Susanne Meinert; Jochen G Hirsch; Thoralf Niendorf; Beate Endemann; Fabian Bamberg; Thomas Kröncke; Robin Bülow; Henry Völzke; Oyunbileg von Stackelberg; Ramona Felizitas Sowade; Lale Umutlu; Börge Schmidt; Svenja Caspers; Harald Kugel; Tilo Kircher; Benjamin Risse; Christian Gaser; James H Cole; Udo Dannlowski; Klaus Berger

doi:10.1126/sciadv.abg9471

An uncertainty-aware, shareable, and transparent neural network architecture for brain-age modeling

Tim Hahn, Jan Ernsting, Nils R Winter, Vincent Holstein, Ramona Leenings, Marie Beisemann, Lukas Fisch, Kelvin Sarink, Daniel Emden, Nils Opel, Ronny Redlich, Jonathan Repple, Dominik Grotegerd, Susanne Meinert, Jochen G Hirsch, Thoralf Niendorf, Beate Endemann, Fabian Bamberg, Thomas Kröncke, Robin BülowHenry Völzke, Oyunbileg von Stackelberg, Ramona Felizitas Sowade, Lale Umutlu, Börge Schmidt, Svenja Caspers, Harald Kugel, Tilo Kircher, Benjamin Risse, Christian Gaser, James H Cole, Udo Dannlowski, Klaus Berger

Neuroimaging

Research output: Contribution to journal › Article › peer-review

10 Citations (Scopus)

Abstract

The deviation between chronological age and age predicted from neuroimaging data has been identified as a sensitive risk marker of cross-disorder brain changes, growing into a cornerstone of biological age research. However, machine learning models underlying the field do not consider uncertainty, thereby confounding results with training data density and variability. Also, existing models are commonly based on homogeneous training sets, often not independently validated, and cannot be shared because of data protection issues. Here, we introduce an uncertainty-aware, shareable, and transparent Monte Carlo dropout composite quantile regression (MCCQR) Neural Network trained on N = 10,691 datasets from the German National Cohort. The MCCQR model provides robust, distribution-free uncertainty quantification in high-dimensional neuroimaging data, achieving lower error rates compared with existing models. In two examples, we demonstrate that it prevents spurious associations and increases power to detect deviant brain aging. We make the pretrained model and code publicly available.

Original language	English
Article number	abg9471
Journal	Science Advances
Volume	8
Issue number	1
DOIs	https://doi.org/10.1126/sciadv.abg9471
Publication status	Published - 5 Jan 2022

Access to Document

10.1126/sciadv.abg9471Licence: CC BY-NC

Cite this

Hahn, T., Ernsting, J., Winter, N. R., Holstein, V., Leenings, R., Beisemann, M., Fisch, L., Sarink, K., Emden, D., Opel, N., Redlich, R., Repple, J., Grotegerd, D., Meinert, S., Hirsch, J. G., Niendorf, T., Endemann, B., Bamberg, F., Kröncke, T., ... Berger, K. (2022). An uncertainty-aware, shareable, and transparent neural network architecture for brain-age modeling. Science Advances, 8(1), Article abg9471. https://doi.org/10.1126/sciadv.abg9471

@article{73f0ca90afb14f2aaa0fd37cd651ad0f,

title = "An uncertainty-aware, shareable, and transparent neural network architecture for brain-age modeling",

abstract = "The deviation between chronological age and age predicted from neuroimaging data has been identified as a sensitive risk marker of cross-disorder brain changes, growing into a cornerstone of biological age research. However, machine learning models underlying the field do not consider uncertainty, thereby confounding results with training data density and variability. Also, existing models are commonly based on homogeneous training sets, often not independently validated, and cannot be shared because of data protection issues. Here, we introduce an uncertainty-aware, shareable, and transparent Monte Carlo dropout composite quantile regression (MCCQR) Neural Network trained on N = 10,691 datasets from the German National Cohort. The MCCQR model provides robust, distribution-free uncertainty quantification in high-dimensional neuroimaging data, achieving lower error rates compared with existing models. In two examples, we demonstrate that it prevents spurious associations and increases power to detect deviant brain aging. We make the pretrained model and code publicly available.",

author = "Tim Hahn and Jan Ernsting and Winter, {Nils R} and Vincent Holstein and Ramona Leenings and Marie Beisemann and Lukas Fisch and Kelvin Sarink and Daniel Emden and Nils Opel and Ronny Redlich and Jonathan Repple and Dominik Grotegerd and Susanne Meinert and Hirsch, {Jochen G} and Thoralf Niendorf and Beate Endemann and Fabian Bamberg and Thomas Kr{\"o}ncke and Robin B{\"u}low and Henry V{\"o}lzke and {von Stackelberg}, Oyunbileg and Sowade, {Ramona Felizitas} and Lale Umutlu and B{\"o}rge Schmidt and Svenja Caspers and Harald Kugel and Tilo Kircher and Benjamin Risse and Christian Gaser and Cole, {James H} and Udo Dannlowski and Klaus Berger",

year = "2022",

month = jan,

day = "5",

doi = "10.1126/sciadv.abg9471",

language = "English",

volume = "8",

journal = "Science Advances",

issn = "2375-2548",

publisher = "American Association for the Advancement of Science",

number = "1",

}

Hahn, T, Ernsting, J, Winter, NR, Holstein, V, Leenings, R, Beisemann, M, Fisch, L, Sarink, K, Emden, D, Opel, N, Redlich, R, Repple, J, Grotegerd, D, Meinert, S, Hirsch, JG, Niendorf, T, Endemann, B, Bamberg, F, Kröncke, T, Bülow, R, Völzke, H, von Stackelberg, O, Sowade, RF, Umutlu, L, Schmidt, B, Caspers, S, Kugel, H, Kircher, T, Risse, B, Gaser, C, Cole, JH, Dannlowski, U & Berger, K 2022, 'An uncertainty-aware, shareable, and transparent neural network architecture for brain-age modeling', Science Advances, vol. 8, no. 1, abg9471. https://doi.org/10.1126/sciadv.abg9471

TY - JOUR

T1 - An uncertainty-aware, shareable, and transparent neural network architecture for brain-age modeling

AU - Hahn, Tim

AU - Ernsting, Jan

AU - Winter, Nils R

AU - Holstein, Vincent

AU - Leenings, Ramona

AU - Beisemann, Marie

AU - Fisch, Lukas

AU - Sarink, Kelvin

AU - Emden, Daniel

AU - Opel, Nils

AU - Redlich, Ronny

AU - Repple, Jonathan

AU - Grotegerd, Dominik

AU - Meinert, Susanne

AU - Hirsch, Jochen G

AU - Niendorf, Thoralf

AU - Endemann, Beate

AU - Bamberg, Fabian

AU - Kröncke, Thomas

AU - Bülow, Robin

AU - Völzke, Henry

AU - von Stackelberg, Oyunbileg

AU - Sowade, Ramona Felizitas

AU - Umutlu, Lale

AU - Schmidt, Börge

AU - Caspers, Svenja

AU - Kugel, Harald

AU - Kircher, Tilo

AU - Risse, Benjamin

AU - Gaser, Christian

AU - Cole, James H

AU - Dannlowski, Udo

AU - Berger, Klaus

PY - 2022/1/5

Y1 - 2022/1/5

N2 - The deviation between chronological age and age predicted from neuroimaging data has been identified as a sensitive risk marker of cross-disorder brain changes, growing into a cornerstone of biological age research. However, machine learning models underlying the field do not consider uncertainty, thereby confounding results with training data density and variability. Also, existing models are commonly based on homogeneous training sets, often not independently validated, and cannot be shared because of data protection issues. Here, we introduce an uncertainty-aware, shareable, and transparent Monte Carlo dropout composite quantile regression (MCCQR) Neural Network trained on N = 10,691 datasets from the German National Cohort. The MCCQR model provides robust, distribution-free uncertainty quantification in high-dimensional neuroimaging data, achieving lower error rates compared with existing models. In two examples, we demonstrate that it prevents spurious associations and increases power to detect deviant brain aging. We make the pretrained model and code publicly available.

AB - The deviation between chronological age and age predicted from neuroimaging data has been identified as a sensitive risk marker of cross-disorder brain changes, growing into a cornerstone of biological age research. However, machine learning models underlying the field do not consider uncertainty, thereby confounding results with training data density and variability. Also, existing models are commonly based on homogeneous training sets, often not independently validated, and cannot be shared because of data protection issues. Here, we introduce an uncertainty-aware, shareable, and transparent Monte Carlo dropout composite quantile regression (MCCQR) Neural Network trained on N = 10,691 datasets from the German National Cohort. The MCCQR model provides robust, distribution-free uncertainty quantification in high-dimensional neuroimaging data, achieving lower error rates compared with existing models. In two examples, we demonstrate that it prevents spurious associations and increases power to detect deviant brain aging. We make the pretrained model and code publicly available.

UR - http://www.scopus.com/inward/record.url?scp=85122879648&partnerID=8YFLogxK

U2 - 10.1126/sciadv.abg9471

DO - 10.1126/sciadv.abg9471

M3 - Article

SN - 2375-2548

VL - 8

JO - Science Advances

JF - Science Advances

IS - 1

M1 - abg9471

ER -

An uncertainty-aware, shareable, and transparent neural network architecture for brain-age modeling

Abstract

Access to Document

Other files and links

Fingerprint

Cite this