Melanoma Classification Through Transfer Learning by the Analysis of Skin Lesion Images

Abstract

Malignant melanoma is a form of skin cancer that develops from melanocytic cells in the human body and can prove to be dangerous if not treated early. With the advent of artificial intelligence, deep learning approaches have been applied for the diagnosis of the disease which in turn shall help medical professionals in the line of treatment. In this paper, transfer learning is used by implementing the pre-trained VGG19 Net model. The model is pre-trained with ImageNet dataset. Certain layers of the pre-trained model are used for the analysis of the dataset of our interest by freezing rest of the convolutional layers. Here, skin lesions images from the ISBI2016 challenge (Gutman et al. in skin lesion analysis toward melanoma detection: a challenge at the international symposium on biomedical imaging, (Gutman D, Codella NCF, Celebi E, Helba B, Marchetti M, Mishra N, Halpern A (2016) skin lesion analysis toward melanoma detection: a challenge at the international symposium on biomedical imaging (ISBI) 2016, hosted by the International Skin Imaging Collaboration (ISIC), [1]) dataset are considered for the classification purpose. Proposed classification method records a commendable validation accuracy of 81.33% along with a testing accuracy of 86.67%, precision of 95.08%, recall of 82.25%, IoU of 78.26% and Dice score of 85.29%.