A major limitation in applying deep learning to artificial intelligence (AI)\nsystems is the scarcity of high-quality curated datasets. We investigate strong\naugmentation based self-supervised learning (SSL) techniques to address this\nproblem. Using breast cancer detection as an example, we first identify a\nmammogram-specific transformation paradigm and then systematically compare four\nrecent SSL methods representing a diversity of approaches. We develop a method\nto convert a pretrained model from making predictions on uniformly tiled\npatches to whole images, and an attention-based pooling method that improves\nthe classification performance. We found that the best SSL model substantially\noutperformed the baseline supervised model. The best SSL model also improved\nthe data efficiency of sample labeling by nearly 4-fold and was highly\ntransferrable from one dataset to another. SSL represents a major breakthrough\nin computer vision and may help the AI for medical imaging field to shift away\nfrom supervised learning and dependency on scarce labels.\n