Design of a complex bioimpedance spectrometer using DFT and undersampling for neural networks diagnostics

Electrical impedance spectroscopy offers many applications in the medical field due the fast response, non-invasiveness and low cost. One promising area is the use of this method for diagnostics. This paper describes the design and experimental evaluation of a multifrequencial complex bioimpedance analyzer. Impedance amplitude and phase were calculated using Discrete Fourier Transform (DFT) and high frequency signals were measured with undersampling. The prototype was able to measure values from 1 Ω to 50 kΩ (frequency range from 50 Hz to 500 kHz). The accuracy of the technique was compared with a commercial equipment. The analysis of passive components resulted in a mean error of 2.9% for the magnitude and 0.69 degrees for the phase. Besides, an initial study for head and neck cancer detection through neural networks is shown. One used bioimpedance values as well as gender, age and body mass index as inputs. The network used 120 training and 40 validation data and was able to simulate 77.5% of the two types of diagnostic correctly.