Distinguishing cancerous from noncancerous cells through analysis of electrical noise

D. C. Lovelady, T. C. Richmond, A. N. Maggi, C.-M. Lo, and D. A. Rabson

Phys. Rev. E 76, 041908 – Published 11 October 2007

Abstract

Since 1984, electric cell-substrate impedance sensing (ECIS) has been used to monitor cell behavior in tissue culture and has proven sensitive to cell morphological changes and cell motility. We have taken ECIS measurements on several cultures of noncancerous and cancerous human ovarian surface epithelial cells. By analyzing the noise in real and imaginary electrical impedance, we demonstrate that it is possible to distinguish the two cell types purely from the signatures of their electrical noise. Our measures include power-spectral exponents, Hurst and detrended fluctuation analysis, and estimates of correlation time; principal-component analysis combines all the measures. The noise from both cancerous and noncancerous cultures shows correlations on many time scales, but these correlations are stronger for the noncancerous cells.

Received 28 July 2007

DOI:https://doi.org/10.1103/PhysRevE.76.041908

Authors & Affiliations

D. C. Lovelady, T. C. Richmond, A. N. Maggi, C.-M. Lo, and D. A. Rabson^*

Department of Physics, University of South Florida, Tampa, Florida 33620, USA

^*Corresponding author. davidra@ewald.cas.usf.edu

Article Text (Subscription Required)

Click to Expand

References (Subscription Required)

Click to Expand

Issue

Vol. 76, Iss. 4 — October 2007

Reuse & Permissions

Access Options

Author publication services for translation and copyediting assistance advertisement

Images

Figure 1
Scheme of data extraction from noise. (a) Time series of resistance for one of the experimental runs. Taking the discrete-time derivative and normalizing to zero mean and unit variance gives the noise, shown in (b). The power spectrum of noise is shown in (c), using half-overlapping windows of 512 points in order to reduce scatter. Fits to the first and last 100 frequencies estimate low- and high-frequency power laws $f^{- α}$ . White noise would have appeared frequency independent $(α = 0)$ . The Fourier transform of the power spectrum gives the autocorrelation (d), which we fitted to a shifted power-law decay to extract the measure $β_{0}$ . As explained in the text, subtle differences in the univariate noise distribution (e) (smoothed) discriminate between cancerous and noncancerous micromotion.Reuse & Permissions
Figure 2
(Color online) Projection along the first two principal components of the 14-dimensional space determined by Tables . Blue open symbols mark the 34 HOSE runs, red crosses the 26 SKOV experiments. As populations, these two sets are distinct, but the overlap of clusters makes it difficult to distinguish individual runs in this type of projection.Reuse & Permissions
Figure 3
Increments of a finite random walk with persistence (left) may mimic certain aspects of the experimental data (right), but with notable differences. The random process has $a = 0.8894$ , so an exponential decay time $τ = 4.00$ . The experiment is a typical capacitance-noise time series of HOSE, with a measured $1 ∕ e$ crossing of $5.7$ . (a) and (b) show the best-fit lines to the first 100 points (excluding zero and the lowest frequency) of the power spectrum; both give slopes $\approx - 1.0$ , but the random data level off noticeably at low frequencies, as would be expected of white noise. Autocorrelation curves (c) and (d) show fits to exponential (dotted line) and shifted power-law (1) (solid) decays. For the random noise, the two fits fall on top of one another, but for the experimental data, a power law fits better than exponential decay.Reuse & Permissions
Figure 4
(Color online) Fractional discrepancies between $H_{pred}$ given by (3) and measured Hurst exponent as functions of measured spectral exponent $α_{low}$ . Near the bottom, plotted with large $+$ symbols, are artificially generated data with known long-time correlations. At top are generated data (large $◇$ ) from random-walk increments with persistence times ranging from $2$ (smaller values of $α$ ) to $7$ (larger values). In the middle are experimental results for HOSE (blue $\circ$ ) and SKOV (red $\times$ ). Most of the experimental data look more like the correlated data than the uncorrelated, but a few overlap with uncorrelated noise; all of these are SKOV.Reuse & Permissions

Physical Review E

covering statistical, nonlinear, biological, and soft matter physics