Identification of Common Blood Metabolic Derangements Using Magnetic Resonance Signatures of the Pancreas and Liver

Abstract

Aim: To investigate whether common disturbances of glucose and lipid metabolism can be automatically identified from magnetic resonance signatures of the pancreas and liver. Methods: In this proof-of-principle study, 100 individuals with a history of pancreatitis-a relatively homogeneous population at risk for metabolic derangements-underwent magnetic resonance assessment on the same 3.0 Tesla scanner. Automated measurements of fat fraction and water proton transverse relaxation time (R2 water) in the pancreas and liver were obtained. Fasting blood samples were analysed for high-density lipoprotein (HDL) cholesterol, low-density lipoprotein (LDL) cholesterol, triglycerides, glucose and insulin. Associations between magnetic resonance signatures and blood metabolic measures were assessed using generalised additive models adjusted for age, sex and body mass index. Results: In fully adjusted models, HDL dyslipidaemia was significantly associated with intra-pancreatic fat (p = 0.015) and intra-hepatic fat (p = 0.047), LDL dyslipidaemia with pancreas R2 water (p = 0.009), and triglyceride dyslipidaemia with intra-hepatic fat (p = 0.046). Lower HOMA-beta was significantly associated with intra-pancreatic fat (p = 0.001), intra-hepatic fat (p = 0.004), pancreas R2 water (p = 0.031) and liver R2 water (p = 0.014). Higher HOMA-IR was significantly associated with pancreas R2 water (p = 0.016). Conclusions: Automated magnetic resonance signatures of pancreatic and hepatic tissue composition were significantly associated with clinically relevant disturbances in lipid metabolism and indices of glucose homeostasis. These findings support the feasibility of opportunistic, automated detection of abnormal blood metabolic parameters using high-resolution cross-sectional imaging.