Dual attention multiple instance learning with unsupervised complementary loss for COVID-19 screening

•We propose a novel end-to-end model for weakly supervised classification of COVID-19 from bacterial pneumonia in chest computed tomography (CT).•We show the benefit of joint unsupervised contrastive and supervised learning of patient labels under the multiple instance learning (MIL) framework.•Experimental results show competitive performance of the proposed framework over existing state-of-the-art MIL methods including supervised methods.