Coherency image analysis to quantify collagen architecture: implications in scar assessment

Abstract

An important histological difference between normal, uninjured dermis and scar tissue such as that found in keloid scars is the pattern (morphological architecture) in which the collagen is deposited and arranged. In the uninjured dermis, collagen bundle architecture appears randomly organized (or in a basket weave formation), whereas in pathological conditions such as keloid scar tissue, collagen bundles are often found in whorls or in a hypotrophic scar collagen is more densely packed in a parallel configuration. In the case of skin, a scar disables the dermis, leaving it weaker, stiff and with a loss of optimal functionality. The absence of objective and quantifiable assessments of collagen orientation is a major bottleneck in monitoring progression of scar therapeutics. In this article, a novel quantitative approach for analyzing collagen orientation is reported. The methodology is demonstrated using collagen produced by cells in a model scar environment and examines collagen remodeling post-TGFβ stimulation in vitro. The method is shown to be reliable and effective in identifying significant coherency differences in the collagen deposited by human keloid scar cells. The technique is also compared for analysing collagen architecture in rat sections of normal, scarred skin and tendon tissue. Results demonstrate that the proposed computational method provides a fast and robust way of analyzing collagen orientation in a manner surpassing existing methods. This study establishes this methodology as a preliminary means of monitoring in vitro and in tissue treatment modalities which are expected to alter collagen morphology.

Fig. 1. Confocal microscopy image of deposited collagen under ‘crowding’ conditions without and with TGFβ stimulation. (A) Unstimulated control of collagen deposited by human keloid fibroblasts in the scar-in-a-jar model and (B) collagen deposited by human keloid fibroblasts in the scar-in-a-jar model with the addition of TGFβ stimulation.

Fig. 2. Comparing the computed ‘directionality’ histograms of the unstimulated control and the TGFβ stimulated sample. Computed histograms of: (A) the stimulated control with the directionality plugin, (B) the stimulated control with the coherency analysis in the OrientationJ plugin, (C) the treated sample with the directionality plugin and (D) the treated sample with the coherency analysis in the OrientationJ plugin.

Fig. 3. Hue, Saturation and Brightness (HSB) maps generated with the OrientationJ plugin. (A) Colour maps of the unstimulated control and (B) the TGFβ stimulated sample. These maps are generated using the OrientationJ plugin, with a particular orientation angle of the collagen assigned to a colour and the saturation of that colour, the local coherency of the image.

Fig. 4. Image region selection used for quantitation. Region of interest (ROI) selection for (A) the unstimulated control and (B) the TGFβ treated sample. Ellipses must be used for structure tensor coherency analysis and the amount of collagen in the ROIs was maximized, avoiding large holes/defects.

Fig. 5. Testing the COI technique with Fourier analysis of the test images. Fast Fourier Transforms (FFTs) with an RGB LUT applied of (A) the unstimulated control and (B) the TGFβ treated sample. The corresponding measurement of the length and width of the FFT ellipses are shown for (C) the unstimulated control and (D) the TGFβ treated sample.

Fig. 6. Comparing coherency analysis with COI analysis. (A) COI analysis of both the unstimulated control and the TGFβ treated sample and (B) coherency analysis of both the unstimulated control and the TGFβ treated sample. As the COI measurement is a function of the entire sample, the difference cannot be statistically quantified, but with the multiple sampling of the image with coherency analysis, the difference can be statistically agreed upon (**p < 0.0001), data displayed as mean ± SD and statistically assessed with a one-way ANOVA followed by a Bonferroni comparison test.

Fig. 7. Coherency analysis on keloid and Dupuytren scar cells. Unstimulated and TGFβ stimulated (A) keloid and (B) Dupuytren cells were compared with this method of coherency analysis and found to be significantly different. This demonstrates the robustness of the methodology for analysing an assortment of scar cells with varying amounts of deposited collagen, i.e. large excesses of collagen as produced by stimulated keloid and Dupuytren cells. Dupuytren n = 4, *p < 0.001. Keloid n = 6, **p < 0.0001, data displayed as mean ± SD and statistically assessed with a one-way ANOVA followed by a Bonferroni comparison test.

Fig. 8. Coherency analysis on rat tissue sections. Coherency analysis comparing control skin with scar following a burn injury to that of rat tendon. The collagen fibre alignment in tendon is significantly different to that of control skin however there was no significant difference evident when comparing control skin with normal scarring in the rat, n = 10 images per sample, **p < 0.0001, data displayed as mean ± SD and statistically assessed with a one-way ANOVA followed by a Bonferroni comparison test.