Biospeckle laser digital image processing for quantitative and statistical evaluation of the activity of Ciprofloxacin on Escherichia coli K-12

Abstract

Antibiotic susceptibility testing is a necessary step prior to the treatment of clinical infections. A major concern is the time required to obtain a fast and reliable result. The aim of this work is to use Biospeckle laser in a 15min assay for an antimicrobial susceptibility test of Ciprofloxacin in serial two-fold dilutions on Escherichia coli K-12 using Venereal Disease Research Laboratory (VDRL) plates. Analysis of images by video edition is performed on a quantitatively selected region of interest, and processed with ImageJ-ImageDP; and by the construction of time series and analysis with either statistical diagnostics tests or Autoregressive Integrated Moving Average (ARIMA) models. Antimicrobial susceptibility tests are also performed for the purpose of quantitative comparison, showing a profile that is comparable to the result obtained with ImageJ-ImageDP processing after 15min of antibiotic action. Only the time series of the least affected bacteria (low Ciprofloxacin concentration) behaves in an expected manner, being non-independent and mainly non-linear, non-normal, and heteroscedastic. The most affected bacteria (higher Ciprofloxacin concentration) are non-independent and tend to be linear, normal and heteroscedastic. Adjustment to a linear regression identifies both, the culture medium without bacteria and the most affected bacteria, normality identifies the most affected bacteria and heteroscedasticity-homoscedasticity distinguishes the presence-absence of bacteria, respectively. ARIMA models (1,1,1)(1,0,1)11 and (4,1,1)(1,1,1)11 fit the time series of the most affected bacteria while the latter also fits the culture medium without bacteria. The time series of the least affected bacteria are identified by a (7,1,2)(1,0,1)11 model. The non-linear, non-normal and heteroscedastic behavior of this group is probably responsible for its adjustment to a model with a relatively high parameter. The four methods: diagnostic statistical tests, fitting of ARIMA models, ImageJ-ImageDP and antimicrobial susceptibility tests, show similar results, being able to distinguish among the groups of assays with bacteria and Ciprofloxacin below and above the Minimal Inhibitory Concentration.\n\nAuthor SummaryBiospeckle laser patterns occur when a dynamic surface is illuminated. This research describes its application to the activity of Escherichia coli bacteria and the effect of 18 different concentrations of an antibiotic (Ciprofloxacin) in a 15min assay, using VDRL plates where the sample has a relatively small volume and is flat shaped. The assay is performed on an anti-vibration table in a dark room with a laser that sequentially illuminates each of the wells of the plate. A camera takes short 30sec videos with approximately 750 frames and sends them to a computer where image processing takes place. In order to select a segment of 80 successive frames to analyze, the region with the higher variation was identified, punched out and edited as a \"flip-book animation\" with a program named ImageJ and processed with another program named ImageDP that takes the difference between successive frames and is able to describe the speed with which the Biospeckle dots move, expecting to show that the antibiotic affects the bacteria by changing the speed with which they move. Also in each video, within the region of higher differential activity, a pixel was selected to construct a time series which is the successive value of that pixel in 253 frames, representing a recording of 8sec. This was analyzed with two statistical methods: diagnostic statistics tests and ARIMA models, both of which try to demonstrate how the results are organized. All the results, the speed of the dots in the \"flip-book animation\" and the structure of the data of the time series, were comparable to those obtained with traditional antimicrobial susceptibility tests with the same bacteria and antibiotic.