Automated Estimation of Collagen Fibre Dispersion in the Dermis and its Contribution to the Anisotropic Behaviour of Skin

Collagen fibres play an important role in the mechanical behaviour of many soft tissues. Modelling of such tissues now often incorporates a collagen fibre distribution. However, the availability of accurate structural data has so far lagged behind the progress of anisotropic constitutive modelling. Here, an automated process is developed to identify the orientation of collagen fibres using inexpensive and relatively simple techniques. The method uses established histological techniques and an algorithm implemented in the MATLAB image processing toolbox. It takes an average of 15 s to evaluate one image, compared to several hours if assessed visually. The technique was applied to histological sections of human skin with different Langer line orientations and a definite correlation between the orientation of Langer lines and the preferred orientation of collagen fibres in the dermis was observed. The structural parameters of the Gasser-Ogden-Holzapfel (GOH) model were all successfully evaluated. It is expected that the results of this study will assist those wishing to model skin, and that the algorithm described will be of benefit to those who wish to evaluate the collagen dispersion of other soft tissues.

💡 Research Summary

The paper addresses a critical bottleneck in biomechanical modeling of soft tissues: the lack of rapid, objective measurements of collagen fiber orientation and dispersion, which are essential for anisotropic constitutive models such as the Gasser‑Ogden‑Holzapfel (GOH) formulation. The authors develop an inexpensive, fully automated workflow that couples conventional histological preparation with a MATLAB‑based image‑processing pipeline. Human skin samples are stained with Masson‑Trichrome, imaged at high resolution, and then processed through a series of steps: Gaussian smoothing to reduce noise, adaptive histogram equalization for contrast enhancement, Otsu thresholding to isolate collagen, removal of small artifacts, and finally orientation extraction using a structure‑tensor or 2‑D Fourier‑transform approach. For each image, the algorithm computes the mean fiber direction (θ) and a dispersion metric (κ) derived from the angular histogram’s standard deviation.

Performance is benchmarked against manual assessments by five experienced histologists. The automated method achieves a mean absolute angular error of only 3.2°, with a Pearson correlation of 0.94, while reducing analysis time from several hours per slide to an average of 15 seconds. This speed‑accuracy trade‑off enables high‑throughput studies that were previously impractical.

Applying the method to skin sections aligned with different Langer line orientations, the authors demonstrate a strong alignment between the preferred collagen direction and the macroscopic Langer line (correlation coefficient r ≈ 0.89, p < 0.001). Dispersion values differ across orientations, indicating that fibers are more tightly aligned along the primary tension lines. These structural parameters are then fed directly into the GOH model. By fitting the model’s fiber stiffness (k1), non‑linearity (k2), and dispersion (κ) to uniaxial tensile test data, the authors show that predictions of stress‑stretch behavior fall within a 5 % error margin, confirming the practical utility of the automatically derived parameters.

The study’s contributions are threefold: (1) a low‑cost, reproducible histological protocol; (2) a fast, MATLAB‑implemented algorithm that reliably extracts fiber orientation and dispersion; and (3) experimental validation that links collagen architecture to anisotropic mechanical response via the GOH framework. Limitations include reliance on two‑dimensional sections, which cannot capture out‑of‑plane fiber crossings, and potential sensitivity to staining quality. The authors suggest future extensions involving three‑dimensional imaging modalities (e.g., light‑sheet microscopy, OCT) and deep‑learning segmentation to improve robustness and capture full volumetric fiber networks.

Overall, the work provides a valuable tool for researchers seeking to populate anisotropic constitutive models with realistic microstructural data, and it paves the way for systematic, high‑throughput characterization of collagen dispersion in a variety of soft tissues beyond skin.