Skin lesion classification in dermoscopic images using stacked Convolutional Neural Network

Skin lesion detection and classification is always observed as a difficult problem to solve. Manual detection of skin lesions via visual image inspection can be time-consuming and tedious. Automatic diagnosis and classification are considered a critical problem to solve because of the involvement of many factors like different image sizes, hairs in the image, bad color schemes, ruler marker, low-contrast, variation in lesion sizes, and gel bubble. Different methodologies were proposed by the researchers in the Dermatology Pigmented lesion classification. Researchers work on the binary class problem for the detection of Melanocytic lesions from the normal one. This study makes use of the MNIST HAM10000 dataset published by International Skin Image Collaboration. The dataset consists of seven classes of skin cancer diseases. Furthermore in this research, our stacked CNN model proves its superiority by achieving 95.2% accuracy along with data augmentation and image preprocessing techniques.