Automated Segmentation of the Hypothalamus: Associations of Age and Socioeconomic Status with Subunit Volumes in Cognitively Healthy Adults

Abstract

Background: The hypothalamus regulates sleep-wake cycles, energy balance, and autonomic and neuroendocrine function, yet its small size and poorly contrasted boundaries have long hindered in vivo volumetry. Deep-learning segmentation now allows the hypothalamus to be divided into functionally meaningful subunits. We examined whether hypothalamic subunit volumes are associated with age and socioeconomic status (SES) in cognitively healthy adults. Methods: From the OASIS-1 cross-sectional dataset (n = 416, 18-96 years), 133 participants with a Clinical Dementia Rating of 0 and complete segmentation output were analysed (33-94 years). A FreeSurfer-based convolutional neural network, run on a cloud virtual machine, segmented five subunits per hemisphere. Volumes were normalised to estimated total intracranial volume (eTIV) and related to age and SES (Hollingshead index) using Pearson correlations with Benjamini-Hochberg false discovery rate (FDR) correction; partial correlations adjusted for sex (age) and for age and sex (SES) served as sensitivity analyses. Results: Older age was associated with lower normalised volume of the right posterior subunit (r = -.456, q < .001) and the left posterior subunit (r = -.274, q = .007); both associations persisted after adjustment for sex. Lower SES showed a trend towards larger left anterior-inferior volume (r = .242, uncorrected p = .005) that did not survive FDR correction (q = .053). Conclusions: Age-related volumetric differences in the healthy hypothalamus appear to be concentrated in its posterior portion, a region housing arousal-promoting circuitry. Automated subunit segmentation offers a scalable tool for studying the hypothalamus in ageing and in relation to environmental exposures; the cross-sectional associations reported here should not be interpreted as causal or as within-person change.