Skin cancer diagnostics: a VGGEnsemble approach

• Main Authors: Arzmi, Mohd Hafiz, P.P. Abdul Majeed, Anwar, Musa, Rabiu Muazu, Mohd Razman, Mohd Azraai, Gan, Hong-Seng, Mohd Khairuddin, Ismail, Ab. Nasir, Ahmad Fakhri
• Format: Book Chapter
• Language: English
• Published: Springer 2023
• Subjects: R Medicine (General)
• Online Access: http://irep.iium.edu.my/103899/1/103899_Skin%20cancer%20diagnostics%20a%20VGGEnsemble%20approach.pdf; http://irep.iium.edu.my/103899/; https://link.springer.com/chapter/10.1007/978-981-19-8937-7_5

The human skin is the largest organ of the human body, and it is highly susceptible to lesions. This study attempts to classify two distinct classes of malignant skin cancers, i.e., Actinic Keratosis (AK) and Basal Cell Carcinoma (BCC), as well as Dermatofibroma (DF), which is benign. A total of 330 dermoscopy images were split into the 70:15:15 ratio for training, testing and validation, respectively. Different VGG-Logistic Regression (LR) pipelines, i.e., VGG16-LR and VGG19-LR, were formulated. In addition, the effect of combining the features extracted from both VGG models, dubbed as VGGEnsemble, was also investigated. It was demonstrated from the study that the ensemble model yielded a better classification accuracy than its standalone versions. Therefore, it could be concluded that the performance of the pipeline is improved through this approach and subsequently could aid the diagnostics of different types of skin diseases by dermatologists.