MULTI-CLASS SKIN CANCER IMAGE CLASSIFICATION WITH TEXTURE AND COLOR ANALYSIS USING SUPPORT VECTOR MACHINE

• Main Author: Varrel Putra Kusuma, Al
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/70898
• Institution: Institut Teknologi Bandung

Skin cancer is one of the most common types of skin cancer found in the world. This disease has a high dependency with the speed and accuracy of the diagnosis to be treated. When late treatment occurs, skin cancer is difficult to treat and can end in death. Today, diagnosis is made using dermatoscope, but price and access to this tool are often difficult, so this solution is not effective enough in the community. With the recent development of technology, especially in the field of machine kearning and computer vision, dermatoscope could be replicated so that it can easily accessed using smartphone or PC. In this thesis, a system is built to detect skin cancer that is easily accessed by everyone. A patient can take his own photo of the skin that is indicated to have skin cancer. Then the patient can send a picture of his picture through the website interface that is built. Furthermore, the system will conduct an analysis of these images using machine learning to classify the types of skin cancers they suffer. With this system, the technology can be more easily accessed, both by patients and health workers quickly and easily, which is expected to encourage faster treatment to be carried out. The results of this study are the construction of a skin cancer classification system with a support vector machine model. From the measurement results, SVM performance is best compared to other classification models. This model has an overall accuracy and f1-score of 78%. As for the specific performance of a disease, the f1-score obtained is 92% for melanoma, 75% for basal cell carcinoma and 56% for squamous cell carcinoma.