Melanoma (skin cancer) diagnosis via image with deep learning technique (using AlexNet and Googlenet architecture)

Melanoma (skin cancer) is one of the most common types of cancer worldwide, and particularly the melanoma subtype is highly dangerous due to its high malignancy potential. If not diagnosed early, melanoma can spread rapidly and lead to fatal outcomes, making it a significant public health threat. Early diagnosis of melanoma greatly improves treatment success and patient survival rates. However, traditional diagnostic methods such as dermatological examinations, expert observation, and biopsy procedures can be time-consuming and costly. Deep learning models, trained on large datasets, provide high accuracy in automated diagnosis and classification. In this study, a classification system for melanoma diagnosis was developed using AlexNet and GoogleNet architectures in the MATLAB environment. The aim of our study is to develop a deep learning-based approach for melanoma (skin cancer) diagnosis and to analyze the accuracy of results obtained with AlexNet and GoogleNet architectures. This thesis compares the performance of deep learning methods, particularly AlexNet and GoogleNet, highlighting the advantages of each model and demonstrating that both can be effective tools in skin cancer diagnosis. The AlexNet model achieved a classification accuracy of 90.19%, while the GoogleNet model achieved 88.38%. Additionally, the thesis includes sensitivity and specificity statistics, as well as Type I and Type II error analysis results. This study offers a new perspective on artificial intelligence applications in medical imaging. Furthermore, this thesis aims to provide healthcare professionals with a supportive tool while showcasing the potential of deep learning and image processing techniques in the healthcare field.