Classifying skin lesion images into primary morphologies

• Main Author: Macatangay, Jules Matthew A.
• Format: text
• Language: English
• Published: Animo Repository 2016
• Online Access: https://animorepository.dlsu.edu.ph/etd_masteral/5271
• Institution: De La Salle University

Skin lesions are widely common irregularities in skin. Most of the research done by computer scientists on processing images of skin lesions focus on skin cancer malignancy. Less research focus has been on classifying skin lesions into their corresponding skin diseases. The classification of skin lesions into skin diseases is di cult given the large number of skin diseases that exist. It may be suitable to rest classify skin lesions by more general categories to reduce complexity. One such general categorization scheme is through the morphology of skin lesions. Morphology can serve as a viable means of categorizing skin lesions as it is descriptive of a skin lesions structure and appearance. Thus, this research aims to model a system that classifies skin lesions into the primary morphologies in dermatological nomenclature. This was accomplished by applying methods in skin malignancy and skin disease research into the problem of classification by morphology. Based on the results, further research is needed to have a deeper analysis of classification by morphology, especially as the research is exploratory. Feature Selection provided no significant increase, and although dropping color channel features provided a boost in performance, certain color channel features may need to be opted in. For this research, Multilayer Perceptron provided the best output based on Cohen's Kappa, falling at 0.413 and 0.436 for the 4 class test and 3 class test, respectively.