Taylor AG, et al. - In this retrospective study, researchers created a large human-annotated dataset of pneumothorax chest X-rays and developed deep convolutional networks to screen for potentially emergent moderate or large pneumothorax at the time of image acquisition. In all, the radiologists visually annotated 13,292 frontal chest X-rays (3,107 with pneumothorax). Images showing pneumothorax of large or moderate size were considered positive and those with a trace or no pneumothorax were considered negative. Small pneumothorax images were excluded from training. Automated image classifiers were developed that detect clinically significant pneumothorax (moderate and large in size as determined by radiologist consensus read) at high levels of performance within a single site while maintaining a reasonable false-positive rate. Such algorithms could be used by radiologists as a tool to increase the speed at which a serious pneumothorax is detected, even at times of lower staffing, when turnaround times are typically longer.