Medical visual question answering using joint self-supervised learning