Deep learning models for neuroimaging have largely been developed for individual tasks, limiting knowledge transfer across applications. Here we introduce GenFAR, a modular deep learning framework that learns general, clinically informed features from brain MRIs. We trained this modular architecture on 49,246 individuals across 11 cohorts, using 17 diverse classification and regression tasks spanning cognition, clinical, diagnosis, demographics, and biomarkers. This yields aggregated, focused feature sets that capture rich, clinically- and biologically-relevant brain representations. We developed a sequential learning approach where tasks progressively build on previously learned representations. Through an analysis of 5,000 task sequences, we identified an optimal sequence length of six tasks and introduced a Donor Score metric to quantify each task's contribution to downstream performance. This analysis revealed five consistently strong donor tasks (Age, AD/MCI, MMSE, Hypertension, Hyperlipidemia) that formed the base of our sequential model. We demonstrated the utility of our learned representation, in various tasks beyond those included in the training set, to serve as the foundation for specialized secondary predictors. We further showed that using the learned feature representation can substantially increase the sample efficiency of secondary deep learning training tasks and models, as well as improve their accuracy.
V. Bashyam, G. Erus, Junhao Wen et al.· 0 citations
Brain iron accumulation is well established in Alzheimer's disease (AD), but whether the two share a genetic basis remains unclear. Here, we performed genome-wide association analyses of brain iron across six iron-rich brain regions, quantified via quantitative susceptibility mapping (QSM) in 38,142 UK Biobank participants. Genetic pleiotropic overlap, genetic correlation and genomic structural equation modelling were used to identify shared genetic architecture between brain iron and risk of AD, and bidirectional Mendelian randomisation was used to test for causal relationships. We identified 116 independent genome-wide significant loci associated with brain iron (r2<0.1; p <5 x 10-8) and validated these signals in an independent cohort using polygenic scores, which explained between 8.9% and 25.4% of variance in regional iron. Genetic pleiotropy and correlation analyses revealed that brain iron and AD share a polygenic basis extending well beyond the canonical APOE locus. This shared liability is specific to basal ganglia structures relevant to cognition, specifically, iron deposition across the caudate nucleus, putamen, and globus pallidus loaded onto a single latent factor that correlated significantly with AD (rg = 0.13, P = 0.019). Multivariate genome-wide association analysis of the shared liability factor between AD and brain iron revealed 27 independent genome-wide significant loci implicating genes that modulate neuroimmune function. Mendelian randomisation provided no evidence of a causal effect in either direction. In summary, we identified region-specific shared genetic liability between brain iron and AD, and a putative neuroimmune mechanism underlying it.
M. R. Rahman, Y. Xia, P. Raniga et al.· medRxiv· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
