Classification of Skin Lesions Using Deep Learning and Machine Learning Methods

Skin lesions are a common disease seen in many people. According to medical literature, the most dangerous type is skin cancer, which is malignant melanoma. In the field of dermoscopy, segmenting and classifying various skin lesions is of great importance, especially for early diagnosis. Computer-aided systems can give very good results in detecting cancerous cells and predicting whether the lesion is benign or malignant. The main purpose of this study is to show that the use of Convolutional Neural Networks (CNN), which is a Deep Learning model, results in high accuracy success in detecting diseases from skin lesion images. Convolutional Neural Network architectures, pretrained EfficientNetB1, ResNet101, ConvNeXtBase, DenseNet201, NasNetLarge models were used for training. In this study, 2 different data sets were used. Experiments were performed on the two-class ISIC-2016 data set and the eight-class ISIC-2019 data set, and the highest accuracy was achieved with 83% test accuracy using the ISIC-2019 data set. In addition, experiments were conducted on machine learning models with 2 data sets and the classification successes were compared by achieving 80% accuracy in the KNN model using the highest accuracy ISIC-2016 data set.