Skin cancer detection using convolutional neural network

Skin cancer is a predominant type of cancer. Unrepaired deoxyribonucleic acid (DNA) in skin cells causes genetic abnormalities or mutations on the skin, resulting in skin cancer. Skin cancer tends to spread slowly to other regions of the body, so it’s easier to treat if caught early. Convolution neural network using AlexNet model is developed for early prediction of skin cancer. Skin cancer is detected and differentiated from melanoma using lesion criteria such as symmetry, color, size, and shape. Dataset with 2750 images are taken with consists of nevus, seborrheic keratoses and melanoma images. Initially preprocessing is done on images by removing air bubbles, noise and artifacts. The dataset is then augmented to address the issue of a lack of appropriate training data or an uneven class balance within the datasets, as well as to enhance the dataset’s size without adding new photos. To boost the training process efficiency, the freshly formed pictures may be utilized to pre-train the specified neural network. Features are extracted from the dataset that helps to identify and recognize the pattern of a large number of datasets. The model is then trained using the AlexNet architecture, which includes convolutional, pooling, ReLU correction, and fully-connected layers. Accuracy of training and validation is 97.8 % and 98.3 %.