Skin Cancer Detection and Classification Based on Deep Learning

Melanoma (MEL) is a form of malignant cancer of the skin that is harmful to health. Appropriate analysis of MEL at initial phase is critical for the achievement percentage of comprehensive treatment. Dermoscopic pictures with benign and malignant types of skin cancer can be examined by Computer Vision System (CVS) to rationalize the procedure of skin cancer classification/detection. Hence, in this studies an automated method is proposed and developed that will help to detect and classify the skin cancer disease by processing skin images. For this purpose, the Deep Learning (DL) technique is performed with the assistance of the TensorFlow Back end and the Keras libraries. The HAM -10000 dataset is used for this purpose it consists of 10015 dermoscopic images which is publicly accessible. 70% of the dataset is deployed for training and 24% for testing and 6% for validation of our proposed model. In pre-processing step, data set is modified by converting into grayscale for removing hair and noise from it. Furthermore, thresholding, morphological filtering, Black Hat and Inpaint is applied to enhance the database. After that data augmentation techniques (Tensor flow back and Keras) are used to enhance the size and quality of training dataset. In particular, MobileN et optimizer is used for classification purpose. The proposed technique provides a sophisticated diagnosis and classification accuracy when compared with previous techniques. Top-N accuracy has been used to validate the performance of proposed technique. The proposed method shows the Top-l accuracy of about 90%. Top-l precision of 80 percent is considered equivalent to a best systems and proposed model accomplished the Top-1 precision of 90%. This top efficiency of the classification and the rapidity of meeting to the ultimate outcomes are the features of this study related to other research.