Early Skin Cancer Detection Using the SLICE-3D Dataset: A Transfer Learning Model and DCGAN Approach to Address Data Imbalance

Early detection of skin cancer is crucial for improving patient outcomes, as the disease progresses rapidly when left untreated. Recent advancements in artificial intelligence have revolutionized the field of early detection, giving clinicians more accurate and efficient diagnostic tools. In this paper, two convolutional neural network-based classifiers using transfer learning are proposed to improve early skin cancer detection. These models were trained and tested on the novel ISIC-2024 dataset. To mitigate the class imbalance in this Dataset, a Generative Adversarial Network (DCGAN) is adopted to synthesize malignant samples. Additionally, the pre-trained VGG-16 and MobileNetV2 models were fine-tuned to improve feature learning and classification performance. Our MobileNetV2-based model outperformed the VGG16-based model, achieving an accuracy of 96.87%, a precision of 98.97%, and a recall of 94.7%. These results highlight the impact of deep learning in early skin cancer detection, and most importantly, they lead to better patient outcomes.