Background In recent years machine learning has rapidly evolved and found applications in a wide variety of health care tasks. Non contrasted head CTs are an important screening tool for detecting intracranial hemorrhages when patients present to the emergency room. We report our experience with training a neural network to predict the presence of spontaneous intracranial hemorrhages, specifically intraparenchymal, subarachnoid, and intraventricular hemorrhages.
Methods Non contrasted head CTs of patients with suspicion for acute ischemic stroke were obtained. A 3 dimensional convolutional network was constructed with a wide residual network configuration. Accuracy, sensitivity, specificity, ROC and AUC values were calculated on validation data, which was separate from training data.
Results A total of 203 head CTs were reviewed. There were 75 (37%) studies on which there was no hemorrhage. The remaining studies contained intraparenchymal, subarachnoid, and intraventricular hemorrhages, in many cases with overlap between classes. Accuracy on validation data was 0.905 on validation data, with a sensitivity of 0.846 and specificity of 1. The AUC, calculated from the ROC curve, was 0.92.
Conclusion The field of machine learning, powered by convolutional neural networks for the task of image recognition and processing, has quickly developed in recent years. We have constructed an artificial neural network that can identify the presence of spontaneous intracranial hemorrhages non contrasted head CT imaging.
Disclosures J. Vargas: 1; C; NVIDIA. 4; C; NVIDIA. A. Spiotta: None. A. Chatterjee: 2; C; Viz.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.