{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,18]],"date-time":"2026-10-18T00:03:37Z","timestamp":1792281817836,"version":"4.4.0"},"reference-count":38,"publisher":"MIT Press","license":[{"start":{"date-parts":[[2026,2,18]],"date-time":"2026-02-18T00:00:00Z","timestamp":1771372800000},"content-version":"vor","delay-in-days":48,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"name":"Fondo de Innovaci\u00f3n Tecnol\u00f3gica de Buenos Aires, Ministerio de Producci\u00f3n, Ciencia e Innovaci\u00f3n Tecnol\u00f3gica de la Provincia de Buenos Aires, Argentina","award":["FITBA-B06"],"award-info":[{"award-number":["FITBA-B06"]}]},{"DOI":"10.13039\/501100003074","name":"Agencia Nacional de Promoci\u00f3n Cient\u00edfica y Tecnol\u00f3gica, Argentina","doi-asserted-by":"crossref","award":["PICT-PRH-2022-01"],"award-info":[{"award-number":["PICT-PRH-2022-01"]}],"id":[{"id":"10.13039\/501100003074","id-type":"DOI","asserted-by":"crossref"}]},{"name":"IBRO-Wellcome Neuroscience Capacity Accelerator for Mental Health"}],"content-domain":{"domain":["direct.mit.edu"],"crossmark-restriction":true},"short-container-title":[],"published-print":{"date-parts":[[2026,3,12]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>Brain age prediction from T1-weighted MRI and its associated brain age gap (BAG) has emerged as a promising neuroimaging biomarker for assessing deviations from normative aging. However, the robustness, bias, and interpretability of existing models across external datasets remain poorly understood, limiting clinical translation. In this study, we evaluated four publicly available brain age models (ENIGMA, DeepBrainNet, Pyment, and BrainAgeNeXt) across four independent MRI datasets (ADNI, UNSAM Long COVID, and two OpenNeuro cohorts), comprising 1,634 subjects with diverse demographic and clinical profiles. Models were tested using their original preprocessing pipelines, and performance was assessed using mean absolute error (MAE), mean error (ME), and BAG variability metrics, with additional analyses of biases related to age, dataset, ethnicity, and education. Interpretability was evaluated using Layer-wise Relevance Propagation, and anatomical correlates were explored using BrainChart-derived centile scores. Group-level comparisons were performed between cognitively normal (CN) individuals and patients with Mild Cognitive Impairment (MCI), Alzheimer\u2019s disease (AD), or Long COVID (LC). Models based on 3D convolutional neural networks (Pyment and BrainAgeNeXt) outperformed the DeepBrainNet 2D CNN and the ENIGMA ridge regression model in both accuracy (MAE: 3.9\u20133.7 vs. 6.2\u201312.4 years respectively) and stability (ASTD: 3.2\u20132.9 vs. 4.6\u20138.3 years). Dataset-specific BAG differences were largely explained by age distributions, whereas ethnicity showed a statistically significant but small effect on BAG in some models. Relevance maps highlighted the lateral ventricles as the most consistently relevant anatomical region, with additional cerebellar contributions emerging in older adults for BrainAgeNeXt. Group-level analyses confirmed elevated BAG in MCI and AD patients compared to CN, while no significant differences were observed in Long COVID participants. These findings suggest that, while BAG is a promising biomarker for group-level analyses, current models are required to address age and demographic biases to enable individual-level clinical application.<\/jats:p>","DOI":"10.1162\/imag.a.1164","type":"journal-article","created":{"date-parts":[[2026,2,18]],"date-time":"2026-02-18T19:54:27Z","timestamp":1771444467000},"update-policy":"https:\/\/doi.org\/10.1162\/mitpressjournals.corrections.policy","source":"Crossref","is-referenced-by-count":3,"title":["Bias and generalizability of brain age prediction models: A multi-cohort evaluation with anatomical and interpretability insights"],"prefix":"10.1162","volume":"4","author":[{"given":"Lautaro J.","family":"Aguzin Parrilli","sequence":"first","affiliation":[{"name":"Centro Universitario de Im\u00e1genes M\u00e9dicas (CEUNIM), Escuela de Ciencia y Tecnolog\u00eda, Universidad Nacional de Gral. San Mart\u00edn, Buenos Aires, Argentina"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-6085-484X","authenticated-orcid":false,"given":"Martin A.","family":"Belzunce","sequence":"additional","affiliation":[{"name":"Centro Universitario de Im\u00e1genes M\u00e9dicas (CEUNIM), Escuela de Ciencia y Tecnolog\u00eda, Universidad Nacional de Gral. San Mart\u00edn, Buenos Aires, Argentina"},{"name":"Instituto de Ciencias F\u00edsicas (ICIFI UNSAM-CONICET), Escuela de Ciencia y Tecnolog\u00eda, Universidad Nacional de Gral. San Mart\u00edn (UNSAM), Buenos Aires, Argentina"},{"name":"Consejo Nacional de Investigaciones Cient\u00edficas y Tecnol\u00f3gicas (CONICET), Buenos Aires, Argentina"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"281","published-online":{"date-parts":[[2026,3,12]]},"reference":[{"key":"2026031215553501700_IMAG.a.1164-b1","article-title":"Software for dataset-wide XAI: From local explanations to global insights with Zennit, CoRelAy, and ViRelAy (Version 2)","author":"Anders","year":"2021","journal-title":"arXiv"},{"issue":"7","key":"2026031215553501700_IMAG.a.1164-b2","doi-asserted-by":"crossref","first-page":"e0130140","DOI":"10.1371\/journal.pone.0130140","article-title":"On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation","volume":"10","author":"Bach","year":"2015","journal-title":"PLoS One"},{"issue":"7","key":"2026031215553501700_IMAG.a.1164-b3","doi-asserted-by":"crossref","first-page":"2312","DOI":"10.1093\/brain\/awaa160","article-title":"MRI signatures of brain age and disease over the lifespan based on a deep brain network and 14 468 individuals worldwide","volume":"143","author":"Bashyam","year":"2020","journal-title":"Brain"},{"key":"2026031215553501700_IMAG.a.1164-b4","doi-asserted-by":"crossref","first-page":"102063","DOI":"10.1016\/j.nicl.2019.102063","article-title":"Bias-adjustment in neuroimaging-based brain age frameworks: A robust scheme","volume":"24","author":"Beheshti","year":"2019","journal-title":"NeuroImage: Clinical"},{"issue":"7906","key":"2026031215553501700_IMAG.a.1164-b5","doi-asserted-by":"crossref","first-page":"525","DOI":"10.1038\/s41586-022-04554-y","article-title":"Brain charts for the human lifespan","volume":"604","author":"Bethlehem","year":"2022","journal-title":"Nature"},{"issue":"1","key":"2026031215553501700_IMAG.a.1164-b6","doi-asserted-by":"crossref","first-page":"450","DOI":"10.1186\/s12883-024-03959-8","article-title":"Cognitive impact and brain structural changes in long COVID patients: A cross-sectional MRI study two years post infection in a cohort from Argentina","volume":"24","author":"Cataldo","year":"2024","journal-title":"BMC Neurology"},{"issue":"12","key":"2026031215553501700_IMAG.a.1164-b7","doi-asserted-by":"crossref","first-page":"681","DOI":"10.1016\/j.tins.2017.10.001","article-title":"Predicting age using neuroimaging: Innovative brain ageing biomarkers","volume":"40","author":"Cole","year":"2017","journal-title":"Trends in Neurosciences"},{"key":"2026031215553501700_IMAG.a.1164-b8","doi-asserted-by":"crossref","first-page":"115","DOI":"10.1016\/j.neuroimage.2017.07.059","article-title":"Predicting brain age with deep learning from raw imaging data results in a reliable and heritable biomarker","volume":"163","author":"Cole","year":"2017","journal-title":"NeuroImage"},{"issue":"17","key":"2026031215553501700_IMAG.a.1164-b9","doi-asserted-by":"crossref","first-page":"6139","DOI":"10.1002\/hbm.26502","article-title":"Prediction of brain age using structural magnetic resonance imaging: A comparison of accuracy and test\u2013retest reliability of publicly available software packages","volume":"44","author":"D\u00f6rfel","year":"2023","journal-title":"Human Brain Mapping"},{"issue":"7907","key":"2026031215553501700_IMAG.a.1164-b10","doi-asserted-by":"crossref","first-page":"697","DOI":"10.1038\/s41586-022-04569-5","article-title":"SARS-CoV-2 is associated with changes in brain structure in UK Biobank","volume":"604","author":"Douaud","year":"2022","journal-title":"Nature"},{"key":"2026031215553501700_IMAG.a.1164-b11","doi-asserted-by":"crossref","first-page":"119637","DOI":"10.1016\/j.neuroimage.2022.119637","article-title":"OpenBHB: A large-scale multi-site brain MRI data-set for age prediction and debiasing","volume":"263","author":"Dufumier","year":"2022","journal-title":"NeuroImage"},{"issue":"8","key":"2026031215553501700_IMAG.a.1164-b12","doi-asserted-by":"crossref","first-page":"3829","DOI":"10.1038\/s41380-019-0626-7","article-title":"Brain-age in midlife is associated with accelerated biological aging and cognitive decline in a longitudinal birth cohort","volume":"26","author":"Elliott","year":"2021","journal-title":"Molecular Psychiatry"},{"issue":"9","key":"2026031215553501700_IMAG.a.1164-b13","doi-asserted-by":"crossref","first-page":"e0184661","DOI":"10.1371\/journal.pone.0184661","article-title":"MRIQC: Advancing the automatic prediction of image quality in MRI from unseen sites","volume":"12","author":"Esteban","year":"2017","journal-title":"PLoS One"},{"issue":"3","key":"2026031215553501700_IMAG.a.1164-b14","doi-asserted-by":"crossref","first-page":"883","DOI":"10.1016\/j.neuroimage.2010.01.005","article-title":"Estimating the age of healthy subjects from T1-weighted MRI scans using kernel methods: Exploring the influence of various parameters","volume":"50","author":"Franke","year":"2010","journal-title":"NeuroImage"},{"key":"2026031215553501700_IMAG.a.1164-b15","article-title":"Machine learning with multi-site imaging data: An empirical study on the impact of scanner effects (version 1)","author":"Glocker","year":"2019","journal-title":"arXiv"},{"issue":"4","key":"2026031215553501700_IMAG.a.1164-b16","doi-asserted-by":"crossref","first-page":"224","DOI":"10.1002\/hbm.10123","article-title":"Three-dimensional maximum probability atlas of the human brain, with particular reference to the temporal lobe","volume":"19","author":"Hammers","year":"2003","journal-title":"Human Brain Mapping"},{"issue":"9","key":"2026031215553501700_IMAG.a.1164-b17","doi-asserted-by":"crossref","first-page":"5124","DOI":"10.1038\/s41380-020-0754-0","article-title":"Brain aging in major depressive disorder: Results from the ENIGMA major depressive disorder working group","volume":"26","author":"Han","year":"2021","journal-title":"Molecular Psychiatry"},{"key":"2026031215553501700_IMAG.a.1164-b18","doi-asserted-by":"crossref","first-page":"101967","DOI":"10.1016\/j.compmedimag.2021.101967","article-title":"Uncertainty estimation and explainability in deep learning-based age estimation of the human brain: Results from the German National Cohort MRI study","volume":"92","author":"Hepp","year":"2021","journal-title":"Computerized Medical Imaging and Graphics"},{"key":"2026031215553501700_IMAG.a.1164-b19","doi-asserted-by":"crossref","first-page":"119504","DOI":"10.1016\/j.neuroimage.2022.119504","article-title":"Towards the interpretability of deep learning models for multi-modal neuroimaging: Finding structural changes of the ageing brain","volume":"261","author":"Hofmann","year":"2022","journal-title":"NeuroImage"},{"key":"2026031215553501700_IMAG.a.1164-b20","doi-asserted-by":"crossref","first-page":"791222","DOI":"10.3389\/fnagi.2022.791222","article-title":"Linking brain age gap to mental and physical health in the Berlin aging study II","volume":"14","author":"Jawinski","year":"2022","journal-title":"Frontiers in Aging Neuroscience"},{"issue":"3","key":"2026031215553501700_IMAG.a.1164-b21","doi-asserted-by":"crossref","first-page":"1118","DOI":"10.1002\/hbm.26144","article-title":"Benchmarking the generalizability of brain age models: Challenges posed by scanner variance and prediction bias","volume":"44","author":"Jirsaraie","year":"2023","journal-title":"Human Brain Mapping"},{"key":"2026031215553501700_IMAG.a.1164-b22","doi-asserted-by":"crossref","DOI":"10.1162\/imag_a_00487","article-title":"BrainAgeNeXt: Advancing brain age modeling for individuals with multiple sclerosis","volume":"3","author":"La Rosa","year":"2025","journal-title":"Imaging Neuroscience"},{"key":"2026031215553501700_IMAG.a.1164-b23","doi-asserted-by":"crossref","first-page":"119210","DOI":"10.1016\/j.neuroimage.2022.119210","article-title":"Deep neural networks learn general and clinically relevant representations of the ageing brain","volume":"256","author":"Leonardsen","year":"2022","journal-title":"NeuroImage"},{"key":"2026031215553501700_IMAG.a.1164-b24","article-title":"A ConvNet for the 2020s (version 2)","author":"Liu","year":"2022","journal-title":"arXiv"},{"issue":"7","key":"2026031215553501700_IMAG.a.1164-b25","doi-asserted-by":"crossref","first-page":"712","DOI":"10.1093\/cercor\/9.7.712","article-title":"Patterns of age-related shrinkage in cerebellum and brainstem observed in vivo using three-dimensional MRI volumetry","volume":"9","author":"Luft","year":"1999","journal-title":"Cerebral Cortex"},{"key":"2026031215553501700_IMAG.a.1164-b26","doi-asserted-by":"crossref","first-page":"1","DOI":"10.1016\/j.dsp.2017.10.011","article-title":"Methods for interpreting and understanding deep neural networks","volume":"73","author":"Montavon","year":"2018","journal-title":"Digital Signal Processing"},{"issue":"1","key":"2026031215553501700_IMAG.a.1164-b27","doi-asserted-by":"crossref","first-page":"27","DOI":"10.61186\/nl.3.1.27","article-title":"Neuroimaging findings of COVID-19: A systematic review on longitudinal studies","volume":"3","author":"Nabizadeh","year":"2024","journal-title":"Neurology Letters"},{"key":"2026031215553501700_IMAG.a.1164-b28","doi-asserted-by":"crossref","first-page":"101871","DOI":"10.1016\/j.media.2020.101871","article-title":"Accurate brain age prediction with lightweight deep neural networks","volume":"68","author":"Peng","year":"2021","journal-title":"Medical Image Analysis"},{"issue":"1","key":"2026031215553501700_IMAG.a.1164-b29","doi-asserted-by":"crossref","first-page":"4581","DOI":"10.1038\/s41467-022-32186-3","article-title":"Addressing fairness in artificial intelligence for medical imaging","volume":"13","author":"Ricci Lara","year":"2022","journal-title":"Nature Communications"},{"key":"2026031215553501700_IMAG.a.1164-b30","article-title":"BOLD variability during cognitive control for an adult lifespan sample [Dataset]","author":"Rieck","year":"2024","journal-title":"Openneuro"},{"key":"2026031215553501700_IMAG.a.1164-b31","article-title":"MedNeXt: Transformer-driven scaling of ConvNets for medical image segmentation (version 5)","author":"Roy","year":"2023","journal-title":"arXiv"},{"key":"2026031215553501700_IMAG.a.1164-b32","doi-asserted-by":"crossref","first-page":"109370","DOI":"10.1016\/j.compeleceng.2024.109370","article-title":"A review of explainable artificial intelligence in healthcare","volume":"118","author":"Sadeghi","year":"2024","journal-title":"Computers and Electrical Engineering"},{"issue":"4","key":"2026031215553501700_IMAG.a.1164-b33","doi-asserted-by":"crossref","first-page":"943","DOI":"10.1007\/s12021-022-09572-9","article-title":"How machine learning is powering neuroimaging to improve brain health","volume":"20","author":"Singh","year":"2022","journal-title":"Neuroinformatics"},{"key":"2026031215553501700_IMAG.a.1164-b34","article-title":"Attention is all you need (version 7)","author":"Vaswani","year":"2017","journal-title":"arXiv"},{"key":"2026031215553501700_IMAG.a.1164-b35","doi-asserted-by":"crossref","first-page":"101879","DOI":"10.1016\/j.media.2020.101879","article-title":"Detect and correct bias in multi-site neuroimaging datasets","volume":"67","author":"Wachinger","year":"2021","journal-title":"Medical Image Analysis"},{"issue":"5","key":"2026031215553501700_IMAG.a.1164-b36","doi-asserted-by":"crossref","first-page":"561","DOI":"10.1016\/j.jalz.2016.10.006","article-title":"The Alzheimer\u2019s disease neuroimaging initiative 3: Continued innovation for clinical trial improvement","volume":"13","author":"Weiner","year":"2017","journal-title":"Alzheimer\u2019s & Dementia"},{"issue":"8","key":"2026031215553501700_IMAG.a.1164-b37","doi-asserted-by":"crossref","first-page":"1238","DOI":"10.3390\/tomography10080093","article-title":"Machine learning and deep learning approaches in lifespan brain age prediction: A comprehensive review","volume":"10","author":"Wu","year":"2024","journal-title":"Tomography"},{"key":"2026031215553501700_IMAG.a.1164-b38","article-title":"Structural (t1) images of 113 young healthy adults; study of effects of chronotype on brain structure [Dataset]","author":"Zareba","year":"2022","journal-title":"Openneuro"}],"container-title":["Imaging Neuroscience"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/direct.mit.edu\/imag\/article-pdf\/doi\/10.1162\/IMAG.a.1164\/2582230\/imag.a.1164.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/direct.mit.edu\/imag\/article-pdf\/doi\/10.1162\/IMAG.a.1164\/2582230\/imag.a.1164.pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,3,12]],"date-time":"2026-03-12T19:55:46Z","timestamp":1773345346000},"score":1,"resource":{"primary":{"URL":"https:\/\/direct.mit.edu\/imag\/article\/doi\/10.1162\/IMAG.a.1164\/135350\/Bias-and-generalizability-of-brain-age-prediction"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026]]},"references-count":38,"URL":"https:\/\/doi.org\/10.1162\/imag.a.1164","relation":{},"ISSN":["2837-6056"],"issn-type":[{"value":"2837-6056","type":"electronic"}],"subject":[],"published-other":{"date-parts":[[2026]]},"published":{"date-parts":[[2026]]},"article-number":"IMAG.a.1164"}}