Multimodal population brain imaging in the UK Biobank prospective epidemiological study

Abstract

Medical imaging has enormous potential for early disease prediction, but is impeded by the difficulty and expense of acquiring datasets prior to symptom onset. UK Biobank aims to address this problem directly by acquiring high quality, consistently acquired imaging data from 100,000 predominantly healthy participants, with health outcomes tracked over coming decades. The brain imaging includes structural, diffusion and functional modalities. Along with body and cardiac imaging, genetics, lifestyle measures, biological phenotyping and health records, this is expected to enable discovery of imaging markers of a broad range of diseases at their earliest stages, as well Users may view, print, copy, and download text and data-mine the content in such documents, for the purposes of academic research, subject always to the full Conditions of use

Joint prediction of multiple scores captures better individual traits from brain images

“…Neuroimaging studies often use low-rank multi-output models like CCA (Hotelling, 1936) and partial least squares (PLS) (Krishnan et al, 2011) to link imaging based features to other 115 blocks of data: to cross-predict fMRI and EEG (Deligianni et al, 2014), to explain genetic outcomes (Floch et al, 2012), behavioral and clinical scores (Monteiro et al, 2016;Miller et al, 2016;Smith et al, 2015), or a different imaging modality (Avants et al, 2010;Sui et al, 2012). Reduced rank regression 120 (Vounou et al, 2010;Izenman, 1975) is a related multi-output linear-regression.…”

“…they capture co-variations between brain functional 45 connectivity and a set of lifestyle, demographic, and clinical questionnaires, that are then mixed into one compound variable (Smith et al, 2015). On UK biobank's 5000-subjects cohort, Miller et al (2016) found 9 basic modes of signal variation bridging brain images and traits, that mix information of very 50 different nature. CCA relates multiple blocks of data (imaging features on one hand, behavioral scores on the other hand) through a latent factor model.…”

Data-Driven Extraction of a Nested Model of Human Brain Function

“…Eighty-seven contrast images from 17 NeuroVault collections met these criteria and were included. Included in this total are two task contrasts from the United Kingdom Biobank study (Miller et al, 2016) (further references to "NeuroVault" contrasts include these two contrasts).…”

Studying neuroanatomy using MRI

“…In many ways still the gold standard, it is hard to scale to multiple scans (especially for modern studies running into the thousands of participants) 5,6 and multiple brain regions. Thus, the primary use of manual volumetry is in smaller studies with a single, focused hypothesis, as well as for the creation of datasets to be incorporated into automatic segmentation algorithms.…”

“…Population neuroscience integrates epidemiology, genetics and neuroscience to identify influences shaping the human brain from conception onwards 5,148,149 . As we discussed elsewhere 150 , such efforts face three key challenges: (1) an infinite combination of factors influencing the brain from within (genes and their regulation) and the outside (social and physical environment); 2) presence of developmental cascades that carry such influences from one time point to the next (e.g., prenatal to postnatal), from one organ to another (e.g., cardiometabolic to brain), and from one level of organization to a different one (e.g., behavior to gene regulation and vice versa); and (3) structural and functional complexity of the human brain.…”