Standardized MR-based localization and quantification of adipose tissue compartments in large cohorts of healthy individuals

Abstract

Obesity is a global public health challenge and a heterogeneous chronic disease associated with disorders of, for example, the endocrine and cardiovascular system. In 2022, 2.5 billion adults were overweight, of which 890 million were living with obesity; the global prevalence has almost tripled since 1975. Starting in the 1950s, research has shown that the distribution of adipose tissue in the body plays a crucial role in characterizing different phenotypes of obesity and that subphenotypes of abdominal obesity are associated with metabolic iseases. This necessitates a closer examination of the distribution of adipose tissue. Non-invasive imaging techniques such as magnetic resonance imaging (MRI) are particularly suitable for this purpose. They are also used in large cohort studies such as NAKO, aiming to investigate the causes of chronic population-wide diseases. Retrospective analyses of large cohort studies as well as the consecutive evaluation of adipose tissue compartments from MRI in ongoing cross-sectional and intervention studies are dependent on standardized, reliable, precise, objective and time-saving methods for image analysis. The aim of this work is to contribute to the standardization of MR-based localization and quantification of adipose tissue compartments by adapting and evaluating an existing deep learning-based image segmentation tool (nnU-Net) for the standardized quantification of visceral and subcutaneous adipose tissue and to demonstrate its applicability on image datasets of more than 11,000 participants of the NAKO. Furthermore, a method for the standardized evaluation of the fat fraction and its subregional distribution in the hematopoietic bone marrow of the vertebral bodies is developed. In addition, a method for correcting confounded measurements of fat fraction in the liver, muscles and bone marrow of the vertebral bodies is proposed. The results show that the developed segmentation model can be used effectively for automatic localization and quantification as well as for characterizing the distribution of visceral and subcutaneous adipose tissue along the body axis. Furthermore, it was shown that the developed method for analyzing the fat fraction in bone marrow provides robust and precise results and enables the investigation of group-specific distribution patterns within the vertebral bodies. In addition, the comparison of commercially available two-point and multi-echo Dixon MRI in bone marrow shows higher systematic errors in the quantification of fat fraction compared to liver and muscle. This error can be corrected using linear models derived from the data. The presented approaches for standardized MR-based analysis of body composition and their applications in a large population-based cohort study contribute to a better understanding of the influence of adipose tissue distribution on metabolic diseases, allow the identification of specific phenotypes associated with cardiometabolic risk and thus provide an essential contribution to risk stratification. In future work, the proposed methods for adipose tissue quantification are to be applied to the more than 30,000 MR image data sets of the NAKO. Furthermore, the methodology is to be extended to additional (ectopic) fat compartments and used, for example, to quantify epi-/pericardial adipose tissue or the fat fraction in the pancreas.