Cardiac magnetic resonance (CMR) imaging is a popular technique for the diagnosis of myocarditis. However, there are some challenges for specialist physicians to identify myocarditis with CMR due to factors such as low contrast, various noises, and invisible mass or lesion in the images. On one hand, deep learning models requires a large amount of data to achieve good results, but expert manually-annotated medical images are expensive to acquire. On the other hand, deep learning pre-trained models suffer from domain discrepancy and overfitting when trained on a small-sized dataset.