Answers are grounded in this paper and cite their sources.
Alcohol use is linked to distributed structural brain alterations, but population-level alcohol-related neurobiology remains difficult to quantify. We tested whether deep learning–derived MRI representations capture reproducible variation in alcohol consumption severity and converge with clinical and biological markers. Representations were extracted from gray-matter volume maps in the German National Cohort (NAKO; N ≈ 28,000). Ridge-regression scores were derived from structural MRI to estimate alcohol consumption severity, measured with the Alcohol Use Disorders Identification Test–Consumption (AUDIT-C). Models were trained in NAKO, externally validated in UK Biobank ( N ≈ 25,000), evaluated in FOR2107 ( N ≈ 1900), and tested against ICD-10 alcohol use disorder (AUD), alcohol-related biomarkers, and cognition. In NAKO, the structural MRI-derived ridge-regression score correlated with measured AUDIT-C severity ( r = 0.11; 95% CI, 0.09–0.12; q FDR = 1.2 × 10 −70 ; MAE = 1.64). This association replicated in UK Biobank ( r = 0.09; 95% CI, 0.07–0.10; q FDR = 4.9 × 10 −41 ; MAE = 1.96) and generalized to AUD, with higher odds in the highest versus lowest brain-score quintile (OR = 2.28; 95% CI, 1.49–3.48; q FDR = 3.9 × 10 −4 ). The score was associated with gamma-glutamyl transferase ( β = 1.17 U/L; q FDR = 0.007), mean corpuscular volume ( β = 14.26 fL; q FDR = 4.4 × 10 −4 ), and slower reaction time ( β = 163.40 ms; q FDR = 3.0 × 10 −5 ). Deep learning–derived structural MRI representations capture weak but reproducible alcohol-related brain variation that converges with AUD diagnosis, peripheral biomarkers, and cognition, supporting their use in population-level addiction neuroscience while underscoring limits for individual-level prediction.
No takes yet. Share an insight, caveat, or question.
Wiegert et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: