Introduction 34 9 22 1 2 8 13 16 20 12 14 22 36 30 5 28 3 6 18 21 11 17 26 27 37 6 28 30 23 33 35 While single cell observations provide a more fundamental insight into matrix remodeling than macroscopic studies, a concern is efficiency. Time-lapsed video microscopy of single cells forms a labor-intensive and time-consuming experiment. Meaningful experiments require the comparison of large data sets, including different cell types, various matrix compositions, or the use of biochemical or molecular interventions. A fully automated, motorized microscopy setup is required that scans series of individual cells and their surrounding matrix. In addition, each individual cell should automatically be kept in focus during remodeling experiments, which take up 24 h or more. Finally, stable algorithms are needed for the automated analysis of geometrical reorganization with minimal user input. The aim of this study is to develop an automated technique that can be employed to obtain and analyze local matrix remodeling by individual cells. The system that we present allows for monitoring of ∼75 cells in parallel, using time-lapse video microscopy and computer-controlled stage positioning. In addition, we present and evaluate a new algorithm for automated detection of collagen matrix deformation around these cells. Methods Cell culturing and collagen matrix preparation Smooth muscle cells, obtained from mesenteric small arteries, were cultured in Leibovitz medium with 10% (v/v) heat-inactivated fetal calf serum. Cells from passages three to nine were used in experiments. 2 2 Automated microscopic imaging x y z original enhanced \documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym} 
\usepackage{amsfonts} 
\usepackage{amssymb} 
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}
$$ {\text{FI}} = \frac{{{\text{SD}}_{{{\text{original}}}} }} {{{\text{SD}}_{{{\text{enhanced}}}} }} $$\end{document} z z x y Gel dynamics analysis 1 Fig. 1 white box yellow box green boxes 1  I 1 t t 0 I 2 t t 0 t I 1 I 2 I 2 I 1 I 1 I 2 1 n cc 1 r 5 Table 1 Settings for matrix compaction analysis by decomposition cross-correlation Decomposition stage Cross-correlation window n cc 1 768 × 768 96 2 384 × 384 48 3 192 × 192 24 4 96 × 96 24 5 48   48 24 6 24 × 24 12 7 12 × 12 6 I 1 I 2 I 1 n cc Validation The method described above (decomposition CC) was validated on several test series against a straightforward cross-correlation analysis (direct CC), with settings according to decomposition stage 1. 2 Table 2 Characteristics of validation images: CC was tested by addition of Gaussian white noise with mean 0.0 and increasing variance levels, in several cases the image was resized or translated Index Scaling (%) Gaussian white noise variance RD image RD increase (%) RD noise Horizontal shift (pixels) a 1 0.0000 0.250   0 b 0.97 0.0000 0.251 0.5  0 c 0.97 0.0005 0.260 4.1 0.104 0 d 0.97 0.0010 0.269 7.6 0.146 0 e 0.97 0.0020 0.285 14.1 0.207 0 f 0.97 0.0030 0.301 20.2 0.253 0 g 0.97 0.0040 0.316 26.3 0.293 0 h 0.97 0.0050 0.329 31.7 0.327 0 i 0.97 0.0005 0.259 4.1 0.108 60 j 0.97 0.0030 0.303 20.2 0.265 60 RD Results Parallel recording of matrix compaction movies n xy z Validation of the collagen compaction analysis Analysis based on cross-correlation of images at a series of decomposition stages was compared with straightforward cross-correlation. This was performed on images simulating matrix compaction, subsequently followed by a challenge of increasing amounts of image noise and artificial translation. 2 2 r 2 Fig. 2 Top left right 2 2 red asterisks Middle 2 Bottom 2  3 2 Fig. 3 2  2 2 4 Fig. 4 Top left Red asterisks Top right Bottom 2 2  Collagen compaction by individual smooth muscle cells 5 Fig. 5 Top left red asterisk Top right yellow circle green Bottom  Discussion This study aimed at developing an automated technique for obtaining and analyzing matrix remodeling by individual cells. Emphasis was put on construction of an automatic, reliable algorithm for assessment of a detailed matrix displacement field. Especially, refinement of a cross-correlation based image analysis with a decomposition scheme was investigated. While “classic” direct CC sufficed for pairs of images with high correlation and low noise, this was no longer the case when substantial matrix remodeling occurred within the time frame between two consecutive images. This resulted in failing of CC at spots of high geometrical reorganization. However, when using decomposition CC, the gross displacements at these positions could be estimated by analysis of parent images with larger dimensions. This way, at a RD increase of 20.2% the average displacement error was lowered threefold in decomposition CC as compared to direct CC. 4 I 2 I 1 n cc 1 I 2 1 1 19 20 29 35 6 28 30 11 23 35 23 33 26 23 33 5 10 21 30 4 15 32 7 24 25 31 37 38 15