Date of Award
Electrical and Computer Engineering
Deep Learning;Melanoma;Segmentation;Skin Cancer;Superpixel;VGG-19
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Skin cancer, characterized by the abnormal growth of skin cells, is a globally prevalent and serious condition. Despite advancements in digital diagnosis techniques, many existing skin cancer detection methods often fail to achieve satisfactory accuracy levels. In our first study, we propose a novel superpixel-based segmentation method that significantly surpasses the traditional k-means clustering in performance. After segmentation, we utilize Convolutional Neural Networks such as VGG16, ResNet50, DenseNet, Xception, Inception, and Mobilenet for feature extraction and classification. We also conducted a comparative analysis with other existing techniques. Our results suggest that our superpixel-based segmentation method notably enhances the accuracy of skin cancer detection. This makes it a promising tool for boosting the precision of automated skin cancer diagnosis systems. In the second study, we enhanced the performance of VGG19 by integrating Batch Normalization and Global Average Pooling layers. Our proposed model outperforms the standard VGG19 in classifying skin cancer. Both of our studies utilized the ISIC 2019 dataset.
Mohammadi Aydoghmishi, Faezeh, "Skin Cancer Detection by Deep Learning Algorithms" (2023). Electronic Theses and Dissertations. 9203.