Unraveling the biomechanical properties of collagenous tissues pathologies using synchrotron-based phase-contrast microtomography with deep learning

Abstract

Mechanical stimuli are regulators not only in cells but also of the extracellular matrix activity, with special reference to collagen bundles composition, amount and distribution. Synchrotron-based phase-contrast computed tomography was widely demonstrated to resolve collagen bundles in 3D in several body districts and in both pre-clinical and clinical contexts. In this perspective study we hypothesized, supporting the rationale with synchrotron imaging experimental examples, that deep learning semantic image segmentation can better identify and classify collagen bundles compared to common thresholding segmentation techniques. Indeed, with the support of neural networks and deep learning, it is possible to quantify structures in synchrotron phase-contrast images that were not distinguishable before. In particular, collagen bundles can be identified by their orientation and not only by their physical densities, as was made possible using conventional thresholding segmentation techniques. Indeed, localised changes in fiber orientation, curvature and strain may involve changes in regional strain transfer and mechanical function (e.g., tissue compliance), with consequent pathophysiological implications, including developmental of defects, fibrosis, inflammatory diseases, tumor growth and metastasis. Thus, the comprehension of these kinetics processes can foster and accelerate the discovery of therapeutic approaches for the maintaining or re-establishment of correct tissue tensions, as a key to successful and regulated tissues remodeling/repairing and wound healing.