The 24 full papers presented were carefully reviewed and selected from 38 submissions. The conference focused on challenges and limitations of current deep learning methods applied to limited and noisy medical data and present new methods for training models using such imperfect data.