Machine Learning-Assisted Equivalent Circuit Characterization for Electrical Impedance Spectroscopy Measurements of Bone Fractures.

Monitoring bone healing after osteosynthesis in the lower limb provides insight into the need for additional medical intervention. Electrical impedance spectroscopy (EIS), via an implantable device, has recently been studied as a remote and radiation-free surveillance method for fracture patients. To address ambiguities in conventional EIS characterizations of physiological and pathological features, this article presents a machine learning (ML)-assisted EIS interpretation method, tailored specifically for cases where an intramedullary nail (IM nail) is applied. Building on the equivalent circuit model method, we propose modeling the electrical properties of the tissue under test (TUT) using generic distributed elements. A convolutional neural network (CNN) is then deployed to prefit the measured EIS with the aim of identifying the dominating elements, addressing the issue of initial guesswork, and revealing their physiological meaning. This method is demonstrated through the analysis of in vivo EIS measurements of the rabbit tibia postsurgery. By aligning fitting curves to closely match the measured spectra, the relevant components in the equivalent circuit model can be consistently identified, and their local correlation with the bone states is characterized. Our method exhibits potential for the quantitative analysis of EIS and could pave the way for its future research and application.