Radiology. 2017 Dec;285(3):719-720. doi: 10.1148/radiol.2017171734. Kahn CE Jr1.
The field of artificial intelligence (AI) offers opportunities to improve the speed, accuracy, and quality of image interpretation and diagnosis in radiology. Advances in computing technology have enabled new and vastly more powerful tools to be brought to bear on medical images. The graphics processing unit (a specialized electronic circuit that arose to meet the needs of video games) uses highly parallel computing processes to speed up computations over large collections of data. It turns out that graphics processing units are surprisingly well-suited to the needs of so-called machine learning, a form of AI in that computers learn to classify images and other data. Graphics processing unit technology has enabled deep learning, an approach in which multiple layers of artificial neural networks are stacked one atop another to process imaging and clinical data (1). The architecture of such networks resembles that of the human visual cortex: The initial layers process simple features such as edges, and subsequent layers combine these features to recognize more complex features.