Implementasi EfficientNetV2 untuk Identifikasi Melanoma melalui Citra Dermoskopi

Abstract

Melanoma is one of the deadliest types of skin cancer because it can spread throughout the body through the blood vessel and lymph system. A common symptom of melanoma is the appearance of black, brown or red spots on the skin, so the symptoms resemble a nevus or mole. Early detection and accurate diagnosis are very important in treating melanoma. Melanoma diagnosis is carried out using dermoscopy or histopathology methods which require biopsy and are quite time consuming and expensive. This research uses the EfficientNetV2 architecture which is a development of the EfficientNet architecture which has been proven to be efficient and effective in image processing, especially image classification. This architecture introduces several modifications to improve model efficiency and accuracy. Model training uses the ISIC 2020 dataset which contains 2100 data, divided into 1470 training data, 420 validation data, and 210 test data. The data goes through preprocessing stages which consist of resizing, hair removal, segmentation, normalization and augmentation. After going through preprocessing, the data is forwarded to the EfficientNetV2 architecture for feature extraction and model training using 40 epochs and a batch size of 20. The results of using the EfficientNetV2 architecture show an accuracy of 93.33%. Based on these accuracy results, it can be concluded that the system is very good at identifying melanoma through dermoscopic images.