Neural Network News: How Deep Learning Predicts Traumatic Brain Injury Outcomes

The algorithm, created by researchers at the University of Pittsburgh School of Medicine, was trained to predict outcomes for patients with traumatic brain injury (TBI). The algorithm was trained on a wide range of data including computed tomography (CT) scans, vital signs, blood tests, heart function, and coma severity estimates for the patient. The large and varied dataset lends itself well to a deep learning algorithm. The model was a fusion model that combined clinical input with a dataset of head CT scans predicting mortality and unfavorable outcomes. The researchers used transfer learning and curriculum learning applied to a convolutional neural network (CNN) in order to specialize the network to the CT scans.