Deep Learning of the Retina Enables Phenome- and Genome-Wide Analyses of the Microvasculature

Abstract

Background: The microvasculature, the smallest blood vessels in the body, has key roles in maintenance of organ health as well as tumorigenesis. The retinal fundus is a window for human in vivo non-invasive assessment of the microvasculature. Large-scale complementary machine learning-based assessment of the retinal vasculature with phenome-wide and genome-wide analyses may yield new insights into human health and disease. Methods: We utilized 97,895 retinal fundus images from 54,813 UK Biobank participants. Using convolutional neural networks to segment the retinal microvasculature, we calculated fractal dimension (FD) as a measure of vascular branching complexity, and vascular density. We associated these indices with 1,866 incident ICD-based conditions (median 10y follow-up) and 88 quantitative traits, adjusting for age, sex, smoking status, and ethnicity. Results: Low retinal vascular FD and density were significantly associated with higher risks for incident mortality, hypertension, congestive heart failure, renal failure, type 2 diabetes, sleep apnea, anemia, and multiple ocular conditions, as well as corresponding quantitative traits. Genome-wide association of vascular FD and density identified 7 and 13 novel loci respectively, which were enriched for pathways linked to angiogenesis (e.g., VEGF, PDGFR, angiopoietin, and WNT signaling pathways) and inflammation (e.g., interleukin, cytokine signaling). Conclusions: Our results indicate that the retinal vasculature may serve as a biomarker for future cardiometabolic and ocular disease and provide insights on genes and biological pathways influencing microvascular indices. Moreover, such a framework highlights how deep learning of images can quantify an interpretable phenotype for integration with electronic health records, biomarker, and genetic data to inform risk prediction and risk modification.

GWAS of Retinal Vessel Tortuosity Identifies 173 Novel Loci Revealing Genes and Pathways Associated with Vascular Pathomechanics and Cardiometabolic Diseases

“…These studies demonstrated that GWAS has great potential to reveal genes with a key role for modulating vascular properties and potential pathomechanisms, yet, due to their limited power, revealed at best the tip of the iceberg. Indeed, a recently published study [49] performed a GWAS on vascular density and fractal dimension extracted from 97 895 images across 54 813 participants from the UK Biobank [54], finding 7 and 13 loci associated with these traits, respectively. The most significant association with the vascular fractal dimension was found for a locus containing the OCA2 gene (p∼10 −80 ), while for the vascular density, the strongest associations were for MEF2C and GNB3.…”

“…In recent years, genome-wide association studies (GWAS) have been used to link genes with phenotypes extracted from fundus images, such as vessel size [39,40], optic disc morphology [41,42], vascular density [43], fractal dimensions [43] and vessel tortuosity [44]. The diameter of the retinal microvasculature was associated with genes TEAD1, TSPAN10, GNB3 and OCA2 [39].…”

“…The diameter of the retinal microvasculature was associated with genes TEAD1, TSPAN10, GNB3 and OCA2 [39]. A recently published study [43] on vascular density and fractal dimensions reported 7 and 13 single nucleotide polymorphism (SNPs) associated with these traits respectively, including OCA2, MEF2C and GNB3. Retinal vessel tortuosity has been associated with SNPs that map to the genes ACTN4 and COL4A2 [44] Tortuosity of the vasculature was reported in the context of CAD [45] and connective tissue disease [46].…”

Artificial intelligence enabled retinal vasculometry for prediction of circulatory mortality, myocardial infarction and stroke

“…40 A recent study using the UK Biobank data source in fewer participants (54,813 vs 65,144 in this study), showed that retinal vessel density and fractal dimensions (extracted from the entire image after deep learning vessel segmentation without distinction between arterioles and venules) were associated with other health outcomes, including overall mortality, hypertension and congestive heart failure, but did not report on risk predictionperformance. 41 Moreover, there was no consistent evidence of associations with incident circulatory disease, and cerebrovascular disease and associations with incident MI were null. 41 Another study in a sub-set of UK Biobank participants (n=5,663) with both retinal and cardiovascular magnetic resonance imaging used deep learning /AI approaches to estimate structural cardiac indices as intermediaries for predicting MI.…”

“…41 Moreover, there was no consistent evidence of associations with incident circulatory disease, and cerebrovascular disease and associations with incident MI were null. 41 Another study in a sub-set of UK Biobank participants (n=5,663) with both retinal and cardiovascular magnetic resonance imaging used deep learning /AI approaches to estimate structural cardiac indices as intermediaries for predicting MI. 42 However, given their approach, specific retinal features of importance remain unclear.…”