Bioimpedance Measurement under Feature Combination of Time-Frequency Domain and Application for E. coli Detection

Abstract

Electrochemical impedance sensors are widely used for bacterial detection and interpreted by the equivalent circuit model (ECM) to evaluate the concentration changes on the electrode interface. However, this method is mainly based on prior models and focuses on frequency domain analysis which cannot precisely describe unknown dynamic processes and a fractional surface coverage of electrode. Distribution of Relaxation Times (DRT) method, as an alternative to interpret impedance spectra can be applied to solve the problem. In this research, two electrode schemes with different modification steps were designed to collect impedance data. An Ensemble Tree-SHAP model was then adopted to quantitatively analyze the impedance data processed by ECM-DRT method in time-frequency domain and applied for the detection of Escherichia coli (E. coli). The results showed that using a dataset integrated with time-frequency domain features significantly improved the accuracy of bacterial concentration prediction. As a result, the RMSE of untrained concentration with LightGBM model was 6.4 × 1 0 − 4 log CFU ml−1, with the sensor detection limit about 2.69 log CFU ml−1. The research provided a tool combining time-frequency analysis for accurate prediction of bacterial concentration, which is crucial for evaluation of the antibacterial effect of long-term antibiotics on E. coli in future.