MELANOMA CLASSIFICATION USING AUTOMATIC REGION GROWING FOR IMAGE SEGMENTATION
Melanoma or skin cancer is one of the most common cancer in the world and can be fatal if not diagnosed early. Many methods have been developed to perform the segmentation process of melanoma classification, including region growing. However region growing method has disadvantages, such as there is a threshold parameter that must be set and the seed parameter that must be manually determined by the user. In this research, we proposed a system for melanoma classification that use automatic region growing to perform image segmentation. The analysis of interclass variance of the overall intensity of melanoma image is implemented to obtain the seed point and threshold parameter values that can provide optimal segmentation results for each image automatically. Then several features are extracted from the melanoma object and classification is performed to classify benign and malignant melanoma. The average accuracy, sensitivity, and specificity of the automatic region growing method on 12 test images were 97.6%, 94.8%, and 98.7%, respectively. Based on the experimental results, the automatic region growing method gives better segmentation results than the region growing method because the threshold value used is adaptive in accordance with the grayscale information of the input image and because the proposed method is able to provide several seed points automatically. The classification result of 30 images of benign melanoma and 30 images of malignant melanoma give 83.3%, 80.0%, and 86.7% average value of accuracy, sensitivity, and specificity, respectively.