Synopsis

Keywords: Image acquisition: Machine learning

Deep learning (DL) methods can generally learn to spot anything that a radiologist can see. We will demonstrate impressive examples of segmentation of organs, automated radiologist scorings, and adaptive MR protocols that optimize the workflow and patient outcome across musculoskeletal and neurological disorders. We will add some intuition on how DL works but also highlight caveats since deep learning is not intelligent and only knows what it has been presented during training. This challenges generalization to other scanner models, protocols, pathological variations, and to other patient populations. Deployment of DL solutions therefore also relies on alert radiology experts.