Deep convolution neural network based automatic multi-class classification of skin cancer from dermoscopic images

Melanoma has been considered as the most fatal category of skin cancer. From past few decades, image processing has shown to be a boon in biomedical field where such cancerous diseases can be diagnosed well in time. Skin Cancer might not be the deadliest but it's necessary to detect it at its early stage in order to save lives of many. So finding a technique which gives high accuracy and early detection is very crucial. Enthusiastic results of image processing in medical field have convinced trained practitioners to rely on the outputs obtained from computer-vision system (CVS). Deep learning technique has impressively advanced and evolved in image classification task in recent years. The proposed research work aims to classify malignant types of skin cancer using image -processing based deep convolutional neural network (DCNN) for human health livelihood. This research work automatically classifies melanoma and non-melanoma skin cancer types; by using different pre-trained DCNN models via transfer learning technique. Comparative analysis done on the findings obtained by training different pre-trained models with and without dropout shows the efficiency of the proposed research work. The strategic framework designed for the classification of the proposed biomedical image shows novel outcome when relative analysis was obtained with and without the use of stochastic gradient descent (SGD). High classification indicates the viability of this work in real-time application.