Comparative Study of Melanoma Disease Classification Using Deep Learning

Melanoma is one of the deadliest and fastest-growing diseases on the globe, taking many lives every year. The early identification of melanoma through dermoscopy images can notably enhance the chances of survival. However, due to factors including the absence of contrast between the lesions and the skin and the visual similarity between melanoma and non-malignant lesions, reliable melanoma differentiation is extremely difficult. Therefore, the accuracy and productivity of pathologists can be significantly increased by implementing a trustworthy automated method for the detection of skin tumours. This study introduces a method that employs deep learning models for cancer detection. Furthermore, we evaluate and analyze the following six deep learning approaches: Inception-ResNetV2, CNN, VGG16, EfficientNet, Densenet201, and MobileNetV2. Two different datasets, ISIC and MNIST, were used to evaluate the suggested deep learning frameworks. The experimental results demonstrate the promising accuracy of our frameworks. This survey highlights significant datasets, benchmark challenges, and evaluation metrics related to skin lesion analysis, offering a thorough overview of the field.