Akademik Veri Yönetim Sistemi
2022 Innovations in Intelligent Systems and Applications Conference, ASYU 2022, Antalya, Türkiye, 7 - 09 Eylül 2022
© 2022 IEEE.Skin cancer is one of the most common types of cancer worldwide. Early diagnosis of malignant skin cancer is important in terms of removing the tumor completely from the body and increasing the chance of survival of the patient. Segmentation is an important process used in medical image analysis, such as detecting diseases, determining the disease region and monitoring morphological changes in the disease region. In this study, skin lesions on dermoscopic images are segmented using deep learning. For this purpose, U-net model with an encoder and decoder structure is used as the segmentation method. In the encoder part of the model, EfficientNetB7 model is proposed as the backbone structure in order to extract the features from the images. The proposed U-Net model with EfficientNetB7 backbone was trained and tested on ISIC 2017 dataset containing 2750 dermoscopic images. According to the experimental results, 0.889 sensitivity, 0.994 specificity, 0.945 accuracy, 0.884 dice constant and 0.809 Jaccard index value were reached. According to the Jaccard index value, which is frequently used in the evaluation of segmentation performance, a better result was obtained than the existing methods in the literature.