A Comparative Analysis Towards Melanoma Classification Using Transfer Learning by Analyzing Dermoscopic Images

• DOI: 10.48550/arxiv.2312.01212
• Publication Date: 2023-01-01
• Authors: Md. Fahim Uddin; Nafisa Tafshir; Mohammad Monirujj...
• Keywords: Transfer of learning; Deep Neural Networks; Expansive

Melanoma is a sort of skin cancer that starts in the cells known as melanocytes. It more dangerous than other types because it can spread to organs. be fatal if spreads parts body. Early detection key cure, but requires skills skilled doctors diagnose it. This paper presents system combines deep learning techniques with established transfer methods enable lesions classification and diagnosis melanoma lesions. Using Convolutional Neural Networks, method for categorizing images into benign malignant this research (CNNs). Researchers used 'Deep Learning' train an expansive number photos & essentially get expected result neural networks need trained huge parameters dermoscopic are sensitive very hard classify. paper, has been emphasized building models less complexity comparatively better accuracy limited datasets partially fewer so predict at ease from input correctly possible within devices computational power. The dataset obtained ISIC Archive. Multiple pre-trained ResNet101, DenseNet, EfficientNet, InceptionV3 have implemented using complete comparative analysis every model achieved good accuracy. Before training models, data augmented by multiple improve Moreover, results previous state-of-the-art approaches adequate melanoma. Among these architectures, DenseNet performed others which gives validation 96.64%, loss 9.43% test set 99.63%.