Robust feature extraction from impedimetric signals using wavelet packet decomposition with application to cytotoxicity testing

Impedance-based cytotoxicity testing provides a simple, real-time and non-invasive assay technique that interferes minimally with cell morphology and function. In this study, impedimetric measurements of ECV304 cells with increasing concentrations of hydrogen peroxide were performed over 3 h, using D sensor connected to a lock-in amplifier at frequencies between 200 and 830 kHz. The output signals were analysed and decomposed using a two-level wavelet packet decomposition to obtain low- and higher-frequency coefficients of each signal. Subsequently, the sum of the negative and positive values and the difference values of each sub-band were calculated to acquire features of the signal coefficients. The patterns of features were compared with the microscopy images showing morphological changes of the cells for the four hydrogen peroxide concentrations and correlations between the extracted features and the morphological changes were identified. This demonstrated that wavelet packet decomposition was an effective feature extraction technique, which could provide a robust input vector for toxin clustering and classification purposes.