Automated triaging of head MRI examinations using convolutional neural\n networks

Abstract

The growing demand for head magnetic resonance imaging (MRI) examinations,\nalong with a global shortage of radiologists, has led to an increase in the\ntime taken to report head MRI scans around the world. For many neurological\nconditions, this delay can result in increased morbidity and mortality. An\nautomated triaging tool could reduce reporting times for abnormal examinations\nby identifying abnormalities at the time of imaging and prioritizing the\nreporting of these scans. In this work, we present a convolutional neural\nnetwork for detecting clinically-relevant abnormalities in\n$\\text{T}_2$-weighted head MRI scans. Using a validated neuroradiology report\nclassifier, we generated a labelled dataset of 43,754 scans from two large UK\nhospitals for model training, and demonstrate accurate classification (area\nunder the receiver operating curve (AUC) = 0.943) on a test set of 800 scans\nlabelled by a team of neuroradiologists. Importantly, when trained on scans\nfrom only a single hospital the model generalized to scans from the other\nhospital ($\\Delta$AUC $\\leq$ 0.02). A simulation study demonstrated that our\nmodel would reduce the mean reporting time for abnormal examinations from 28\ndays to 14 days and from 9 days to 5 days at the two hospitals, demonstrating\nfeasibility for use in a clinical triage environment.\n

Ähnliche Arbeiten