The value in artificial intelligence

Abstract

Past 5 years have seen burgeoning applications of machine learning (ML) in diverse radiological domains including thoracic radiology, neuroimaging, abdominal imaging, musculoskeletal imaging, and breast imaging. Deep learning technologies have been applied to improve image resolution at ultralow radiation dose. Publications abound on ML in chest CT have focused on detection and characterization of pulmonary nodules, as well as for rib and spine straightening and labeling, vessel segmentation, and estimation of CT fractional flow reserve. ML has also been applied for detecting lines, tubes, pneumothorax, pleural effusions, cardiomegaly, and pneumonia, on chest radiographs. Applications of ML in cerebral hemorrhage detection and prediction of stroke outcomes, appendicitis and renal colic prediction, hand bone age calculation or rib unfolding for fracture detection, and characterization of breast macro-calcifications and masses are also shown. We review fundamentals, applications, and limitations of machine learning in thoracic radiology, neuroimaging, abdominal imaging, musculoskeletal imaging, and breast imaging.