Automated Analysis of Collagen Histology in Ageing Skin

Traditionally, expert analysis is required to evaluate pathological changes manifested in tissue biopsies. This is a highly-skilled process, notwithstanding issues of limited throughput and inter-operator variability, thus the application of image analysis algorithms to this domain may drive innovation in disease diagnostics. There are a number of problems facing the development of objective, unsupervised methods in morphometry that must be overcome. In the first instance, we decided to focus on one aspect of skin histopathology, that of collagen structure, as changes in collagen organisation have myriad pathological sequelae, including delayed wound healing and fibrosis. Methods to quantify incremental loss in structure are desirable, particularly as subclinical changes may be difficult to assess using existing criteria. For example, collagen structure is known to change with age, and through the calculation of foci distances in ellipses derived from the Fourier scatter, we were able to measure a decrease in collagen bundle thickness in picrosirius stained skin with age. Another key indicator of skin physiology is new collagen synthesis, which is necessary to maintain a healthy integument. To investigate this phenomenon, we developed a colourbased image segmentation method to discriminate newly-synthesised from established collagen revealed by Herovici’s polychrome staining. Our scheme is adaptive to variations in hue and intensity, and our use of K-means clustering and intensity-based colour filtering informed the segmentation and quantification of red (indicating old fibres) and blue pixels (indicating new fibres). This allowed the determination of the ratio of young to mature collagen fibres in the dermis, revealing an age-related reduction in new collagen synthesis. These automated colour and frequency domain methods are tractable to high-throughput analysis and are independent of operator variability.