Deep convolutional neural networks and transfer learning for classifying malignant and benign skin cancer

Over 123,000 occurrences of skin cancer cases are identified globally each year. The ability to identify malignant skin lesions as early as possible would be a huge benefit to clinicians from reliable automatic skin cancer screening systems. In the last five years, it appears that methods based on deep learning perform better than conventional methods for classifying images. Techniques based on machine learning are less practical because they need thousands of labelled images for each class to be trained. We propose a precise method to distinguish benign from malignant skin lesions using deep convolutional neural networks with transfer learning (DCNNT). The first step in our process is to eliminate noise and artefacts from the input images. In the second step, the input images must be normalized and features needed for classification extracted. The third is to augment the data with additional images, which improves classification accuracy. We employ transfer learning models, such as Densenet121, MobileNet, ResNet50, and VGG19 to assess the performance of our suggested DCNNT model. On a dataset comparing benign and malignant tumours, we obtained the highest training and testing accuracy with DCNN and DenseNet transfer learning model, with 98.16% training and 92.42% testing. Our DCNNT model is dependable and robust. We discovered that our suggested model performed better than other model.