A review of self-supervised imaging phenotypes for genome-wide association studies.

Self-supervised learning for imaging genetics can expand the phenotypic space for GWAS of organ structure and function by learning scalable, data-driven traits from high-content images.

• Why it matters: Expert-defined image-derived phenotypes capture only a fraction of biological variation, limiting discovery in biobank-scale imaging-genomic cohorts.
• What they did: This review summarizes technical frameworks, challenges, and applications of SSL-based imaging phenotypes for GWAS in brain MRI, cardiac MRI, and retinal imaging.
• The result: Early studies suggest SSL-derived traits can complement traditional IDPs, but broader validation across cohorts, ancestries, scanners, and imaging protocols is still needed.