Automated Computational Techniques for High-throughput Image Analysis of Skin Structure

Osman, Osman Sharif (2015) Automated Computational Techniques for High-throughput Image Analysis of Skin Structure. Doctoral thesis, University of Buckingham.

2015 Osman Sharif Osman.pdf
Available under License Creative Commons Attribution Non-commercial No Derivatives.

Download (14MB) | Preview


Biological image processing and analysis are concerned with enhancing and quantifying features that reflect different pathological states, based on the use of combinations of image processing algorithms. The integration of image processing and analysis techniques to evaluate and assess skin integrity in both human and mouse models is a major theme in this thesis. More specifically, this thesis describes computational systems for high-throughput analysis of skin tissue section images and non-invasive imaging techniques. As the skin is a largest organ in the mammalian body, and is complex in structure, manual quantification and analysis a hard task for the observer to determine an objective result, and furthermore, the analysis is complex in terms of accuracy and time taken. To look at the gross morphology of the skin, I developed high throughput analysis based on an adaptive active contour model to isolate the skin layers and provide quantification methods. This was utilised in a study to evaluate cutaneous morphology in 475 knockout mouse lines provided by the Mouse Genetics Project (MGP) pipeline, that was generated by the Wellcome Trust Sanger Institute (WTSI). This is a major international initiative to provide both functional annotation of the mammalian genome and insight into the genetic basis of disease. I found 53 interesting adipocyte phenotypes, 18 interesting dermal phenotypes and 3 interesting epidermal phenotypes. I also focussed on the analysis of collagen in the dermis of skin images in several ways. For collagen structure analysis, I developed a combined system of Gabor filtering and Fast Fourier Transform FFT. This analysis allowed the detection of subtle changes in collagen organisation. Using similar images, I also measured collagen bundle thickness by computing the maximum frequency of the FFT power spectrum. To assess collagen dynamics, I developed k-means clustering for segmentation based on colour distribution. The use of this approach allowed the measurement of dermal degradation with age and disease, which was not possible by existing means. Obtaining human skin material to facilitate the drug discovery and development process is not an easy task. The manipulation, monitoring and cost of human subjects makes the use of mouse models more suitable for high-throughput screening. Therefore, I have evaluated skin integrity from mouse tissue rather than human skin, however, mouse skin is thinner than human skin and many morphological features are easier to visualise in human skin, which has implications for analysis. Skin moulds can be used to create an impression of the skin surface. Changes in texture of skin can reflect skin conditions. I developed a skin surface structure analysis system to measure the degree of change in texture of the human skin surface. The alterations detected in texture parameters in skin mould impressions reflected changes caused by sun exposure, ageing and many other clinical parameters. I compared my analysis with the existing Beagley-Gibson scoring system to find correlations between automated and manual analysis to inform a decision on the use of optimal methods. By removing subjectivity of manual methods, I was able to develop a robust system to evaluate, for example, damage resulting from UV exposure. My experimental analysis indicated that techniques developed in this thesis were able to analyse both histological samples and skin surface images in high-throughput experiments. They could, therefore, make a contribution to biological image analysis by providing accurate results to help clinical decision making, and facilitate biological laboratory experiments to improve the quality of research in this field, and save time. Overall, my thesis demonstrated that accurate analysis of the skin to gain meaningful biological information requires an automated system that can achieve feature extraction, quantification, analysis and decision making to find interesting phenotypes and abnormalities. This will help the evaluation of the effects of a specific treatment, and answer many biological questions in fields of cosmetic dermatology and drug discovery, and improve our understanding of the genetic basis of disease.

Item Type: Thesis (Doctoral)
Uncontrolled Keywords: image processing, skin, collagen
Subjects: Q Science > QA Mathematics > QA75 Electronic computers. Computer science
R Medicine > RL Dermatology
Divisions: School of Computing
Depositing User: Rachel Pollard
Date Deposited: 29 Apr 2019 10:26
Last Modified: 29 Apr 2019 10:29

Actions (login required)

View Item View Item