Use this URL to cite or link to this record in EThOS:
Title: Computer aided classification of histopathological damage in images of haematoxylin and eosin stained human skin
Author: Haggerty, Juliana
ISNI:       0000 0004 6060 4631
Awarding Body: Newcastle University
Current Institution: University of Newcastle upon Tyne
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Access from Institution:
Excised human skin can be used as a model to assess the potency, immunogenicity and contact sensitivity of potential therapeutics or cosmetics via the assessment of histological damage. The current method of assessing the damage uses traditional manual histological assessment, which is inherently subjective, time consuming and prone to intra-observer variability. Computer aided analysis has the potential to address issues surrounding traditional histological techniques through the application of quantitative analysis. This thesis describes the development of a computer aided process to assess the immune-mediated structural breakdown of human skin tissue. Research presented includes assessment and optimisation of image acquisition methodologies, development of an image processing and segmentation algorithm, identification and extraction of a novel set of descriptive image features and the evaluation of a selected subset of these features in a classification model. A new segmentation method is presented to identify epidermis tissue from skin with varying degrees of histopathological damage. Combining enhanced colour information with general image intensity information, the fully automated methodology segments the epidermis with a mean specificity of 97.7%, a mean sensitivity of 89.4% and a mean accuracy of 96.5% and segments effectively for different severities of tissue damage. A set of 140 feature measurements containing information about the tissue changes associated with different grades of histopathological skin damage were identified and a wrapper algorithm employed to select a subset of the extracted features, evaluating feature subsets based their prediction error for an independent test set in a Naïve Bayes Classifier. The final classification algorithm classified a 169 image set with an accuracy of 94.1%, of these images 20 were an unseen validation set for which the accuracy was 85.0%. The final classification method has a comparable accuracy to the existing manual method, improved repeatability and reproducibility and does not require an experienced histopathologist.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (D.Eng.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available