Melanoma detection from dermoscopy images using Nasnet Mobile with Transfer Learning
Mustafa Çakmak, Mehmet Emin Tenekeci
Abstract
Early detection of melanoma, which is the most dangerous type of skin lesion, offers timely treatment. Specialists determine the melanoma from dermoscopic images. It is distinguished manually from 7 different types of skin lesions. Detection and classification of diseases from medical images has become widespread with the development of computers and algorithms. In this study, deep learning methods, which are the latest technology, are used for the diagnosis of melanoma. Since the training of deep learning methods requires a lot of data and there is not enough labeled data in medical applications, model training is carried out with the transfer learning method. In this study, Nasnet Mobile architecture is classified by retraining using the HAM10000 skin lesion dataset provided by ISIC 2018. The data set is divided into two for training and testing. In order to increase the number of samples for training data and to eliminate the imbalance, data increase was carried out. In order to see the effect of the data augmentation process, the training and testing process was carried out before and after the data augmentation. The results obtained as a result of training and testing were evaluated with accuracy, precision, recall and f1-score metrics. The accuracy rate obtained with the Nasnet Mobile model was increased to 89.20% before data increase and 97.90% after data increase. The obtained results have been compared with similar studies and its effectiveness has been shown.