Implementing of Transfer Learning Method in the Diagnosis of Skin Diseases with Convolutional Neural Networks

Abstract

Millions of people are diagnosed with skin cancer every year around the world, and many people die from this disease. Early diagnosis is important in skin diseases. For this reason, studies on identifying skin diseases with high accuracy using computer-assisted machine learning-based algorithms have gained importance. Convolutional neural networks are frequently used to detect skin diseases quickly and with high accuracy using medical images. In this study, a method using transfer learning is proposed to classify the HAM10000 dataset with high accuracy. Pre-trained models with the ImageNet dataset were transferred and used for classification of the HAM10000 dataset. To demonstrate the effectiveness of the proposed method, Xception and DenseNet201 convolutional neural network models are used separately. In experimental studies, the number of images in the dataset was increased by real-time data augmentation method. In the study, better classification results were obtained in the Xcepiton model compared to the DenseNet201 model, according to the test accuracy, precision, sensitivity and fl-score criteria. It has been observed that higher performances are obtained when the results in this study are compared with similar studies in the literature.