Medical Images Classification with Deep Learning Techniques

The rate of skin tumors has risen due to increased ultraviolet radiation. Although most is harmless and does not affect survival, some of the more harmful skin tumors present a deadly threat if a delay in examination leaves them to become in advance level. Ideally, an analysis by an expert dermatologist would correctly detect malignant skin tumors in the early stage. The detection of Melanoma cancer in an early stage can be helpful to cure it. Computer vision can play an essential role in Medical Image Diagnosis, and many existing systems have proved it. This master thesis focuses on Melanoma detection from ISIC dataset of medical images using deep learning methods in the computer vision field. This paper contains a research how fine tuning of pretrained transfer learning models affects the quality of models and how to use transfer learning in this domain.