Use this URL to cite or link to this record in EThOS:
Title: EMED : Enhancement of Melanoma Early Diagnosis
Author: Clawson, Kathy M. J.
ISNI:       0000 0004 2720 2761
Awarding Body: University of Ulster
Current Institution: Ulster University
Date of Award: 2009
Availability of Full Text:
Access from EThOS:
Malignant melanoma is an aggressive cutaneous cancer originating in melanocytes, the cells which produce skin pigmentation. When treating malignant melanoma prompt diagnosis is paramount, as the depth of vertical invasion (Breslow thickness) is inversely correlated with survival rates. The development of computerised systems for assisted recognition of malignant melanoma can aid enhanced early diagnosis of this disease and help reduce the subjectivity associated with clinical diagnostic procedures. We describe and evaluate key elements of a prototype system for classifying benign pigmented skin lesions and cutaneous malignant melanoma, with emphasis placed on the extraction of existing clinical criteria. The system utilises image processing methods to enable specific features to be defined. Statistical- and intelligence- based classification techniques are then employed to automate identification of significant features, with the aim of reducing feature analysis variability. After lesion boundary determination, a total of 77 parameters for describing the visual appearance (shape asymmetry, border irregularity, colour and differential distribution, and dimension) of pigmented skin lesions are extracted from digital epiluminescence microscopy images and assessed using a double-blind study. A variety of classification methodologies (specifically C5.0 induction, classification and regression induction, and a neural network model) are applied to assess the diagnostic capability of these parameters and to compare system performance against the diagnostic accuracy achieved by two experts using clinical visual inspection methodologies. It is concluded that the parameters generated may have high discriminative value when differentiating between benign and malignant lesions. When using our full feature set on a data set comprising 30 lesions, diagnostic accuracy achieved via the automated classifiers constituted a significant improvement when compared with clinical diagnostic accuracy achieved by our experts.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available