Unsupervised anomaly localization in high-resolution breast scans using deep pluralistic image completion

Nicholas Konz, Haoyu Dong, Maciej A. Mazurowski

Abstract: Highlights•A fast unsupervised tumor localization method for the challenging DBT modality.•Uses a novel anomaly metric based on pluralistic image completion.•Includes a formal theoretical analysis of the metric, supported by experiments.•Comprehensive empirical studies of both accuracy and speed.•Improves on performance of existing machine learning methods by a wide margin.