論文ID: 22-00090
Spatiotemporal local changes have been extracted for evaluating physiological phenomena by an image reconstruction algorithm of electrical impedance tomography (EIT) using sparse Bayesian learning (SBL). The proposed method identifies a region of interest (ROI) by a priori information on conductivity distribution σ of each biological tissue and automatically learns block-sparsity and temporal-correlation in the identified ROI in the form of blocked column vector (BCV). Two types of numerical simulations are conducted to model lymphedema (LE) and venous edema (VE) that cause spatiotemporal local changes in σ of subcutaneous tissue fluids in human calf phantom which consists of three parts: great saphenous vein (GSV) as ROI and subcutaneous adipose tissue (SAT) and muscle as background. From the results, spatiotemporal local changes in σ are clearly extracted only near GSV by the proposed method even in a field where the time-varying σ in the background is large. Furthermore, the accuracy of the proposed method is evaluated under the variant conditions of conductivity ratios of SAT and muscle relative to GSV, i.e., ρGSV/SAT and ρGSV/muscle, respectively, and area ratio accuracy ARAGSV is the highest in the case where ρGSV/SAT = ρGSV/muscle, which achieves ARAGSV = 2.241 regardless of the values of ρGSV/SAT and ρGSV/muscle.
