A Transfer Learning based GUI for Skin Cancer Diagnosis and Classification using Dermoscopic Images

Skin Cancer is considered the most common cancer worldwide. It accounts for more than one million new cases diagnosed each year. It is a form of cancer that affects cells of the skin, where they grow uncontrollably and show abnormal changes, thus resulting in a tumor. Although being a severe issue, in many areas with a lack of proper medical and technological facilities, the detection of skin cancer becomes a challenge. Many are unaware of the symptoms and severity of the disease. The probability of recovery is greatly increased by early diagnosis. Through the recent developments of deep learning technologies, the healthcare cost of identifying and classifying skin cancer can be drastically reduced. With this paper, we provide a study of the performance of different deep learning techniques – InceptionV3 Net, Xception Net, and InceptionResNetV2 on the MNIST:HAM (Human Against Machine) 10000 Skin Cancer dataset, with the help of transfer learning. The dataset, originally is imbalanced and to provide the best results, the study has been performed on both imbalanced and balanced datasets. The models have been compared on evaluation metrics such as accuracy, precision, recall, and f1-score. The study shows better results when compared to some state-of-the-art methods. The best working model Xception Net is used to develop a user-friendly Graphical User-Interface (GUI) to help doctors to detect and classify skin cancer in the early stage.