Melanoma Detection Using 3D CNN from Hyperspectral Images

Most of the deaths related to skin are caused by melanoma, which is responsible for 75% of these deaths. If detected and treated early, it results in a 99% chance of survival. Clinical methods to diagnose melanoma are dermoscopic analysis and histopathological analysis. In existing system, dermoscopic images are used as input. Using these dermoscopic images, some of the features are lost because of its 2-dimensional property. The usage of 2D dermoscopic images reduces accuracy of the results. The dermoscopic images doesn’t give information of Melanoma at an early stage. To rectify that we thought of giving 3D images as input that has a benefit of not losing any feature while feature extraction thus providing more accurate results and detection at an early stage. To achieve the melanoma segmentation with higher accuracy, our system uses a deep convolutional network to analyse hyper-spectral pathology images. A three-dimensional convolutional network is used for the melanoma detection from hyper-spectral images to facilitate the use of the spectral features of three-dimensional hyper-spectral images.