Brain age prediction: A comparison between machine learning models using region- and voxel-based morphometric data

• Type of Material: Article
• Authors: Lea Baecker, King's College London; Jessica Dafflon, King's College London; Pedro F da Costa, King's College London; Rafael Garcia-Dias, King's College London; Sandra Vieira, King's College London; Cristina Scarpazza, King's College London; Vince Calhoun, Emory University; João R Sato, Universidade Federal do ABC, São Paulo; Andrea Mechelli, King's College London; Walter HL Pinaya, King's College London
• Language: English
• Date: 2021-03-19
• Publisher: WILEY
• Publication Version: Final Published Version
• Copyright Statement: © 2021 The Authors. Human Brain Mapping published by Wiley Periodicals LLC.
• License: Creative Commons Attribution 4.0 International
• Final Published Version (URL): http://dx.doi.org/10.1002/hbm.25368
• Title of Journal or Parent Work: HUMAN BRAIN MAPPING
• Volume: 42
• Issue: 8
• Start Page: 2332
• End Page: 2346
• Grant/Funding Information: Fundação de Amparo à Pesquisa do Estado de São Paulo, Grant/Award Numbers: 2018/04654‐9, 2018/21934‐5; National Institutes of Health, Grant/Award Numbers: R01DA049238, R01MH118695; Wellcome Trust, Grant/Award Number: 208519/Z/17/Z
• Keywords: DISORDERS; Life Sciences & Biomedicine; Neuroimaging; machine learning; healthy ageing; DIFFEOMORPHIC IMAGE REGISTRATION; Neurosciences & Neurology; SEGMENTATION; CLASSIFICATION; SCHIZOPHRENIA; Radiology, Nuclear Medicine & Medical Imaging; Science & Technology; regression analysis; MRI SCANS; Neurosciences; biological ageing; support vector machine
• Research Categories: Engineering, Biomedical; Mathematics; Psychology, General

Abstract

Brain morphology varies across the ageing trajectory and the prediction of a person's age using brain features can aid the detection of abnormalities in the ageing process. Existing studies on such “brain age prediction” vary widely in terms of their methods and type of data, so at present the most accurate and generalisable methodological approach is unclear. Therefore, we used the UK Biobank data set (N = 10,824, age range 47–73) to compare the performance of the machine learning models support vector regression, relevance vector regression and Gaussian process regression on whole-brain region-based or voxel-based structural magnetic resonance imaging data with or without dimensionality reduction through principal component analysis. Performance was assessed in the validation set through cross-validation as well as an independent test set. The models achieved mean absolute errors between 3.7 and 4.7 years, with those trained on voxel-level data with principal component analysis performing best. Overall, we observed little difference in performance between models trained on the same data type, indicating that the type of input data had greater impact on performance than model choice. All code is provided online in the hope that this will aid future research.