Deep-E2M2IT: Deep learning-based combined electromagnetic impedance tomography for abdominal fat segmentation

Abstract Abdominal fat has been an important predictor of all-cause mortality. Due to its low cost, ease of use and non-ionizing properties, abdominal adipose content screening using bioimpedance has received increasing research attention in recent years. In this study, we investigate a deep learning-based combined electromagnetic impedance tomography (Deep-E2M2IT) for abdominal fat segmentation. A physics-encoded neural network (PeNN) is designed to predict the segmentation mask for both subcutaneous and visceral fat using data collected through both electric/magnetic excitations and electric/magnetic measurements. Numerical experiments demonstrate the proposed method can accurately predict the distribution of subcutaneous and visceral fat.